ЛР6: OpenTelemetry, Collector, Prometheus/Loki/Tempo, Grafana, алерты, исследование цены наблюдаемости
@@ -18,3 +18,9 @@ __pycache__
|
||||
node_modules
|
||||
evidence
|
||||
.DS_Store
|
||||
research
|
||||
student-tests
|
||||
scripts
|
||||
docs
|
||||
requirements-research.txt
|
||||
*.md
|
||||
|
||||
@@ -1,30 +1,41 @@
|
||||
# Copy to .env for your LOCAL synthetic test environment; never commit .env.
|
||||
# These are logical interface suggestions; map actual framework names in REPORT.md.
|
||||
# Секреты (пароль БД, DATABASE_URL) создаёт scripts/init-secrets.sh в каталоге secrets/ — не в .env.
|
||||
APP_HOST=0.0.0.0
|
||||
APP_PORT=8080
|
||||
APP_ENV=lab
|
||||
# Срок жизни сессии; для проверки expiry запускайте с SESSION_TTL_SECONDS=2 (см. student-tests/test_expiry.sh)
|
||||
SESSION_TTL_SECONDS=3600
|
||||
# Choose credentials locally. File paths contain no secret values.
|
||||
DATABASE_URL_FILE=/run/secrets/database_url
|
||||
# Optional on 3, required use on 4 in lab2; Redis OR Valkey:
|
||||
# TTL записи в кэше Valkey (не больше остатка жизни сессии)
|
||||
SESSION_CACHE_TTL_SECONDS=60
|
||||
# Источник сессий: PostgreSQL (истина) + Valkey (кэш). Адрес кэша задан в compose.yaml (VALKEY_URL).
|
||||
SESSION_STORE=postgres
|
||||
CACHE_URL_FILE=/run/secrets/cache_url
|
||||
DB_POOL_MAX=10
|
||||
# Файлы секретов внутри контейнера (создаются scripts/init-secrets.sh)
|
||||
DATABASE_URL_FILE=/run/secrets/database_url
|
||||
# --- lab3 ---
|
||||
# Число API-реплик (compose deploy.replicas); также можно: docker compose up -d --scale api=4
|
||||
API_REPLICAS=2
|
||||
IDEMPOTENCY_TTL_SECONDS=86400
|
||||
S3_ENDPOINT=http://s3:9000
|
||||
S3_REGION=us-east-1
|
||||
API_CPUS=1.0
|
||||
API_MEM=256m
|
||||
POSTGRES_CPUS=1.0
|
||||
LB_CPUS=1.0
|
||||
# Пауза между SIGTERM и закрытием listener: LB успевает исключить реплику (rolling update без ошибок)
|
||||
DRAIN_SECONDS=4
|
||||
# --- lab4 ---
|
||||
S3_BUCKET=messenger-attachments
|
||||
S3_ACCESS_KEY_ID_FILE=/run/secrets/s3_access_key_id
|
||||
S3_SECRET_ACCESS_KEY_FILE=/run/secrets/s3_secret_access_key
|
||||
MAX_UPLOAD_BYTES=10485760
|
||||
MAX_IMAGE_PIXELS=40000000
|
||||
BROKER_URL_FILE=/run/secrets/broker_url
|
||||
S3_CPUS=1.0
|
||||
UPLOAD_CONCURRENCY=4
|
||||
# --- lab5 ---
|
||||
RABBITMQ_CPUS=1.0
|
||||
WORKER_CPUS=1.0
|
||||
WORKER_MEM=768m
|
||||
WORKER_REPLICAS=1
|
||||
WORKER_CONCURRENCY=2
|
||||
JOB_MAX_ATTEMPTS=5
|
||||
JOB_TIMEOUT_SECONDS=60
|
||||
STAGING_RETENTION_SECONDS=86400
|
||||
OTEL_SERVICE_NAME=messenger-api
|
||||
# Worker must set a distinct service.name.
|
||||
OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4318
|
||||
OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
|
||||
OTEL_RESOURCE_ATTRIBUTES=deployment.environment.name=lab
|
||||
MAX_ATTEMPTS=5
|
||||
LEASE_SECONDS=60
|
||||
MAX_QUEUED_JOBS=200
|
||||
# --- lab6: наблюдаемость ---
|
||||
OTEL_ENABLED=1
|
||||
# доля трасс при head-sampling в SDK (1.0 = все); профиль tail-sampling: COLLECTOR_CONFIG=/etc/otelcol/collector-tail.yaml
|
||||
OTEL_TRACES_SAMPLER_ARG=1.0
|
||||
COLLECTOR_CONFIG=/etc/otelcol/collector.yaml
|
||||
|
||||
@@ -0,0 +1,32 @@
|
||||
# syntax=docker/dockerfile:1
|
||||
# Базовый образ закреплён тегом и digest (python 3.12.12, Debian bookworm).
|
||||
ARG PYTHON_IMAGE=python:3.12.12-slim-bookworm@sha256:593bd06efe90efa80dc4eee3948be7c0fde4134606dd40d8dd8dbcade98e669c
|
||||
|
||||
# --- сборка зависимостей в отдельном слое (в runtime не попадают pip и кэш) ---
|
||||
FROM ${PYTHON_IMAGE} AS builder
|
||||
ENV PIP_DISABLE_PIP_VERSION_CHECK=1 PIP_NO_CACHE_DIR=1
|
||||
WORKDIR /build
|
||||
COPY requirements.txt .
|
||||
RUN python -m venv /opt/venv && /opt/venv/bin/pip install -r requirements.txt
|
||||
|
||||
# --- runtime ---
|
||||
FROM ${PYTHON_IMAGE}
|
||||
ENV PATH=/opt/venv/bin:$PATH \
|
||||
PYTHONUNBUFFERED=1 PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=/srv/src \
|
||||
APP_HOST=0.0.0.0 APP_PORT=8080 APP_ENV=lab LIMIT_CONCURRENCY=1000 MIGRATIONS_DIR=/srv/migrations
|
||||
RUN groupadd --system --gid 10001 app && useradd --system --uid 10001 --gid app --no-create-home app
|
||||
COPY --from=builder /opt/venv /opt/venv
|
||||
WORKDIR /srv
|
||||
COPY src/ /srv/src/
|
||||
COPY migrations/ /srv/migrations/
|
||||
# ревизия образа (меняется при rolling update; видна в журнале запуска)
|
||||
ARG REV=dev
|
||||
LABEL org.opencontainers.image.revision=$REV
|
||||
ENV APP_REVISION=$REV
|
||||
USER 10001:10001
|
||||
EXPOSE 8080
|
||||
HEALTHCHECK --interval=10s --timeout=3s --start-period=5s --retries=3 \
|
||||
CMD ["python", "-c", "import urllib.request,os,sys; sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:%s/health/ready' % os.environ.get('APP_PORT','8080'), timeout=2).status == 200 else 1)"]
|
||||
# exec-форма: сервер = PID 1. Состояние хранится в PostgreSQL/Valkey, процесс stateless;
|
||||
# масштабирование — репликами контейнера. SIGTERM: drain (readiness 503) -> graceful shutdown (см. app/serve.py).
|
||||
CMD ["python", "-m", "app.serve"]
|
||||
@@ -28,3 +28,20 @@ test-persistence:
|
||||
test-failover:
|
||||
@test -n "$(ACTION_SCRIPT)" || (echo "Set ACTION_SCRIPT to your implemented executable stop-one hook"; exit 2)
|
||||
$(PYTHON) tests/resilience.py --mode failover --action-script "$(ACTION_SCRIPT)" --base-url "$(BASE_URL)" $(TLS_ARGS)
|
||||
|
||||
# --- собственные цели студента ---
|
||||
.PHONY: secrets up down student-test
|
||||
secrets:
|
||||
sh scripts/init-secrets.sh
|
||||
|
||||
up: secrets
|
||||
docker compose --env-file .env up -d --build
|
||||
|
||||
down:
|
||||
docker compose --env-file .env down
|
||||
|
||||
student-test:
|
||||
$(PYTHON) -m unittest discover -s student-tests -p "test_api.py" -v
|
||||
$(PYTHON) -m unittest discover -s student-tests -p "test_attachments.py" -v
|
||||
$(PYTHON) -m unittest discover -s student-tests -p "test_pipeline.py" -v
|
||||
$(PYTHON) -m unittest discover -s student-tests -p "test_observability.py" -v
|
||||
|
||||
@@ -1,37 +1,154 @@
|
||||
# Отчёт — лабораторная 6
|
||||
|
||||
- Автор, группа: TODO
|
||||
- Целевая оценка; выбранный трек 5, если нужен: TODO
|
||||
- Commit/PR: TODO после публикации
|
||||
- Использованные источники, библиотеки и помощь ИИ: TODO
|
||||
- Автор, группа: [ФИО], [группа]
|
||||
- Целевая оценка; выбранный трек 5, если нужен: **5**, трек 1 «Цена наблюдаемости»
|
||||
- Commit/PR: локальный коммит ветки `solution/lab6`; PR — после публикации
|
||||
- Использованные источники, библиотеки и помощь ИИ: Python 3.12, OpenTelemetry Python SDK 1.45.0 (traces/metrics/logs, OTLP/HTTP), uvicorn, asyncpg, aiobotocore, aio-pika, Pillow; OpenTelemetry Collector contrib 0.161.0, Prometheus 3.15.0, Loki 3.7.8, Tempo 2.9.0, Grafana 13.2.3; RabbitMQ 4.1.8, SeaweedFS 4.48, PostgreSQL 17, Valkey 8, HAProxy 3.0. Код lab5 перенесён, бизнес-API не менялся. Использование ИИ: [заполняется автором]
|
||||
|
||||
## Запуск из чистого clone
|
||||
|
||||
TODO: версии runtime/образов, prerequisites, команды создания локальных secrets/.env, build, миграций, запуска, ожидания готовности, тестов и остановки. Указать необходимые действия оператора. Не вставлять секреты. Отдельно указать безопасную остановку и команду намеренного удаления учебных данных.
|
||||
Prerequisites: Docker Compose v2 (≈ 4 ГБ свободной RAM для всего стенда), bash (Git Bash на Windows), Python ≥ 3.9; на Windows `PYTHONUTF8=1`.
|
||||
Версии всех образов закреплены тегом и digest в `compose.yaml`; зависимости — `requirements.txt` (включая `opentelemetry-*==1.45.0`).
|
||||
|
||||
```sh
|
||||
sh scripts/init-secrets.sh # secrets/*: БД, S3, RabbitMQ, пароль Grafana (secrets/grafana_admin_password)
|
||||
cp .env.example .env
|
||||
docker compose --env-file .env config --quiet
|
||||
docker compose --env-file .env up -d --build
|
||||
curl http://127.0.0.1:8080/health/ready # API
|
||||
# Grafana: http://127.0.0.1:3000 пользователь admin, пароль — secrets/grafana_admin_password
|
||||
python3 tests/smoke.py --base-url http://127.0.0.1:8080 # make test (телеметрия включена)
|
||||
python3 -m unittest discover -s student-tests -p test_observability.py -v # метрики, логи, трассы, связь API -> worker, отсутствие секретов
|
||||
python3 student-tests/test_observability_retry.py # все попытки задания в одной трассе
|
||||
python3 research/incidents.py --out evidence # 4 инцидента, алерты firing -> resolved (~10 мин)
|
||||
python3 research/obs_cost.py --out evidence # трек 1 (~15 мин)
|
||||
```
|
||||
|
||||
Windows: `scripts/restart.cmd`, `scripts/stop-one.cmd`. Остановка с сохранением данных — `docker compose --env-file .env stop`; **намеренное удаление учебных данных** — `docker compose --env-file .env down -v` (volumes `pgdata`, `valkeydata`, `seaweeddata`, `rabbitmqdata`, `promdata`, `lokidata`, `tempodata`, `grafanadata`).
|
||||
Профиль без телеметрии: `OTEL_ENABLED=0`; head-sampling: `OTEL_TRACES_SAMPLER_ARG=0.1`; tail-sampling в Collector: `COLLECTOR_CONFIG=/etc/otelcol/collector-tail.yaml`.
|
||||
|
||||
## Архитектура и решения
|
||||
|
||||
TODO: схема процессов/хранилищ, источник истины, транзакционные границы, ограничения. Что перенесено из lab5, что изменено в этой работе. Соответствие переменных из .env.example реальной конфигурации.
|
||||
```
|
||||
API ×2 ─┐ ┌─▶ Prometheus (метрики, retention 7 д, volume promdata)
|
||||
├─OTLP/HTTP─▶ OTel Collector ─┼─▶ Tempo (трассы, retention 7 д, volume tempodata)
|
||||
worker ─┘ (memory_limiter, batch, └─▶ Loki (логи, retention 7 д, volume lokidata)
|
||||
scrub, tail_sampling*) ▲
|
||||
traceparent: API ─▶ outbox ─▶ RabbitMQ ─▶ worker Grafana (127.0.0.1:3000) читает все три источника
|
||||
```
|
||||
|
||||
| Роль | Сервис | Версия | Порты | Volume | Зачем |
|
||||
| --- | --- | --- | --- | --- | --- |
|
||||
| Collector | `otel-collector` | contrib 0.161.0 | OTLP 4317/4318, метрики 8888/8889 — **только сеть backend** | — | единая точка приёма, ограничение RAM, удаление чувствительных атрибутов, маршрутизация сигналов; это не хранилище |
|
||||
| Метрики | `prometheus` | 3.15.0 | не публикуется | `promdata` | временные ряды, SLI/recording rules, alert rules, firing/resolved |
|
||||
| Логи | `loki` | 3.7.8 | не публикуется | `lokidata` | структурные логи API и worker (OTLP), поиск по `trace_id` |
|
||||
| Трассы | `tempo` | 2.9.0 | не публикуется | `tempodata` | распределённые трассы, TraceQL |
|
||||
| UI | `grafana` | 13.2.3 | `127.0.0.1:3000` | `grafanadata` | datasources/dashboard из Git (provisioning), анонимный доступ выключен, пароль — secret |
|
||||
|
||||
**Что хранит что.** Grafana ничего не хранит из сигналов: это интерфейс запросов (PromQL, LogQL, TraceQL) и provisioning. Collector — маршрутизатор/обработчик (не БД). Метрики — Prometheus, логи — Loki, трассы — Tempo. Конфигурации (`infra/otel`, `infra/prometheus`, `infra/loki`, `infra/tempo`, `infra/grafana`) и dashboard JSON лежат в Git; провижининг datasources и dashboard автоматический.
|
||||
|
||||
### Сигналы и соответствие `contracts/observability.md`
|
||||
|
||||
| Сигнал контракта | Реализация (SDK) → имя в Prometheus / место |
|
||||
| --- | --- |
|
||||
| HTTP requests; HTTP duration (s) | `http.server.request.duration` (histogram, секунды, buckets 5 мс…10 с) → `http_server_request_duration_seconds_{bucket,sum,count}{http_request_method, http_route, http_response_status_code}`; число запросов = `_count` (отдельный счётчик не дублируется) |
|
||||
| Jobs processed/failed; duration | `messenger.jobs.processed{job_type,outcome=ready|failed|retried|noop}` → `messenger_jobs_processed_total`; `messenger.job.duration` → `messenger_job_duration_seconds_*` |
|
||||
| End-to-end time to ready | `messenger.job.time_to_ready` (от приёма до ready) → `messenger_job_time_to_ready_seconds_*` |
|
||||
| Queue depth / oldest job age | `messenger.queue.depth`, `messenger.queue.oldest_job_age` → `messenger_queue_depth`, `messenger_queue_oldest_job_age_seconds` (считает API, т. е. метрика живёт и при остановленном worker) |
|
||||
| DB pool used/waiting | `db.client.connections.usage{state}`, `db.client.connections.waiting` → `db_client_connections_usage`, `db_client_connections_waiting` |
|
||||
| Worker busy/capacity | `messenger_worker_busy`, `messenger_worker_capacity` |
|
||||
| CPU/RAM процессов | `process_memory_rss_bytes`, `process_cpu_time_seconds_total` (счётчик CPU-времени процесса) |
|
||||
| Логи API и worker | структурный JSON в stdout **и** OTLP: `ts, level, service.name, msg(event), trace_id, span_id` (при активном спане), `request_id`; resource: `service.name` (`messenger-api`/`messenger-worker`), `service.version`, `service.instance.id`, `deployment.environment.name=lab` |
|
||||
| Трассы API и worker | server-спан запроса (`HTTP POST /api/v1/…`), `db.connection`, `s3.put_object/get_object`, producer-спан `attachment.process publish`, consumer-спан `attachment.process process` — реальные, не статичные |
|
||||
| Связь upload → publish → consume → S3/DB | W3C `traceparent/tracestate`: контекст запроса сохраняется в `outbox` (миграция 0006), relay кладёт его в поля сообщения (`jobs.schema.json`) и заголовки AMQP; worker продолжает ту же трассу (consumer — потомок producer; producer — потомок server-спана) |
|
||||
|
||||
**Низкая кардинальность.** `user_id/chat_id/message_id/attachment_id/request_id/filename` не используются как labels метрик (проверяет `test_no_high_cardinality_or_secret_labels`); маршрут — шаблон `/api/v1/chats/{chat_id}/messages`. В логах/спанах допустимы технические ID синтетических данных. **Секреты не экспортируются**: токены, пароли, `Authorization`, тела сообщений, изображения, SQL-текст, query-строки — не попадают ни в атрибуты, ни в логи; дополнительно Collector удаляет `http.request.header.authorization`, `db.statement`, `url.query` и т. п. (`attributes/scrub`). Проверка `test_secrets_and_message_bodies_are_not_exported`: уникальный текст сообщения, токен и пароль не находятся ни в Loki, ни в трассе, ни в метриках.
|
||||
|
||||
**Health-проверки** LB (раз в секунду) не порождают спанов и access-логов (метрики остаются, исключены из SLI фильтром `http_route!~"/health.*"`); фоновые опросы БД (relay, sweeper) — без корневых спанов (`child_span`).
|
||||
|
||||
### Dashboard и SLI/SLO
|
||||
|
||||
Dashboard `Messenger: API, очередь, worker` (`infra/grafana/dashboards/messenger.json`, генератор `build_dashboard.py`): RPS по маршрутам и кодам, **доля 5xx**, **p50/p95/p99** (`histogram_quantile` из `sum by (le)` гистограмм всех реплик — не среднее p95), задержки по маршрутам, DB pool (занято/простаивает/ожидают), задания по исходам, p50/p95 длительности обработки и **времени до ready**, глубина очереди, **возраст старейшей задачи**, saturation worker (busy/capacity), RSS/CPU, очередь и отказы экспорта Collector, логи ошибок (Loki), список алертов, последние трассы. Единицы и запросы описаны в подсказках панелей.
|
||||
|
||||
- **SLI доступности**: доля бизнес-запросов (все маршруты кроме `/health`) без ответа 5xx. Числитель — 5xx, знаменатель — все запросы без `/health`; 4xx (неверный пароль, чужой чат) в отказ сервиса не входят.
|
||||
- **SLO**: 99.5 % за окно 30 суток (бюджет ошибок 0.5 %); для демонстрации на локальном стенде окна алертов короткие (rate по 1 мин, `for` 15–20 с).
|
||||
- Дополнительные SLI: p95 задержки (порог 0.5 с), доля 5xx **на маршруте загрузки** (малообъёмный, но критичный — общий SLI его деградации не заметит), p95 времени до ready.
|
||||
|
||||
### Alert rules (`infra/prometheus/rules/messenger.yml`, runbook — `docs/runbooks.md`)
|
||||
|
||||
| Алерт | Условие / окно | Для чего |
|
||||
| --- | --- | --- |
|
||||
| `HighHttpErrorRate` | 5xx > 5 % при ≥ 0.5 RPS, `for 20s`, rate[1m] | общий отказ сервиса |
|
||||
| `HighUploadErrorRate` | 5xx на маршруте загрузки > 20 %, `for 15s` | отказ S3/очереди |
|
||||
| `HighLatencyP95` | p95 > 0.5 с, `for 20s` | медленная БД/обработчик |
|
||||
| `QueueLagGrowing` | старейшая задача > 30 с, `for 20s` | остановка/перегрузка worker |
|
||||
| `WorkerSilent` | `absent(messenger_worker_capacity)`, `for 30s` | worker не шлёт метрики |
|
||||
| `TelemetryExportFailing` | отказы экспорта Collector > 0, `for 15s` | сбой хранилищ телеметрии |
|
||||
|
||||
Alerts вычисляет Prometheus; состояние (pending/firing/resolved) видно в Prometheus и на панели «Алерты» dashboard. Внешняя доставка не требуется и не настроена.
|
||||
|
||||
### Collector: ограничение ресурсов и поведение при backpressure
|
||||
`memory_limiter` (limit 280 MiB при лимите контейнера 384 MiB, spike 60 MiB; при приближении данные отклоняются, SDK отбрасывает пакет, приложение продолжает работать), `batch` (1024/2 с), экспортёры с **ограниченной очередью** (`sending_queue` 500) и повторами до 30 с, затем отбрасывание; SDK: очередь спанов/логов 2048, таймаут экспорта 3 с, **ошибки экспорта не влияют на ответы API** (проверено). `metric_expiration: 20s` — остановившийся источник метрик быстро исчезает (алерт `WorkerSilent`).
|
||||
|
||||
## Доказательства критериев
|
||||
|
||||
| Критерий README | Команда/сценарий | Наблюдаемый результат | Файл доказательства |
|
||||
| --- | --- | --- | --- |
|
||||
| Публичный smoke | make test | TODO | TODO |
|
||||
| Остальные пункты уровня 3 | TODO | TODO | TODO |
|
||||
| Все пункты уровня 4, если заявлен | TODO | TODO | TODO |
|
||||
| Выбранный трек 5, если заявлен | TODO | TODO | TODO |
|
||||
| Весь минимум lab5 работает с включённой телеметрией | `tests/smoke.py`; `student-tests/test_pipeline.py` | PASS; 9 тестов OK | `evidence/checks.txt` |
|
||||
| Collector, 3 хранилища и Grafana в стенде, конфиги в Git, volumes | `docker compose up -d`; `infra/**` | стенд поднимается одной командой; datasources и dashboard импортируются provisioning-ом | `evidence/grafana_provisioning.txt` |
|
||||
| Метрики RPS/errors/duration API; jobs ok/failed и duration worker | `test_observability.py::test_metrics_for_http_jobs_queue_pool_worker` | все 12 серий имеют данные | `evidence/checks.txt` |
|
||||
| Структурные логи и реальные трассы API и worker, связь запрос ↔ лог, job ↔ лог | `test_one_trace_spans_…`, `test_logs_are_structured_and_linked_by_trace_id` | одна трасса содержит спаны `messenger-api` и `messenger-worker`; в Loki по `trace_id` находятся логи обоих сервисов, `span_id` у каждой записи | `evidence/checks.txt` |
|
||||
| Низкая кардинальность, нет секретов | `test_no_high_cardinality…`, `test_secrets_and_message_bodies_are_not_exported` | OK | `evidence/checks.txt` |
|
||||
| W3C trace context API → broker → worker (parent-child), цепочка до S3/DB, включая retry | `test_one_trace…`; `test_observability_retry.py` | server → producer → consumer; `s3.put/get_object`, `db.connection` в той же трассе; 3 попытки задания — 3 consumer-спана одной трассы, 2 лога `job_rescheduled` | `evidence/checks.txt` |
|
||||
| Dashboard: p50/p95/p99, error ratio, queue depth/age, DB pool, saturation worker | `infra/grafana/dashboards/messenger.json` | импортируется автоматически (`/api/search` возвращает dashboard) | `evidence/grafana_provisioning.txt` |
|
||||
| SLI/SLO, alert rules, firing → resolved на искусственных отказах | `research/incidents.py` | см. «Инциденты» | `evidence/incidents.json`, `incidents.png` |
|
||||
| Collector с memory limiter/batch; остановка хранилища телеметрии не ломает API | инцидент 4 | API без ошибок (ratio 0.0, RPS ≈ 67), очередь Collector ограничена (≈23–28), RSS Collector ≈ 247 MiB < лимита 384 MiB | `evidence/incidents.json` |
|
||||
| Два+ инцидента с диагностикой | см. ниже | 4 инцидента | отчёт и `evidence/incidents.json` |
|
||||
| **Трек 1** | `research/obs_cost.py` | см. ниже | `evidence/obs_cost.json` |
|
||||
|
||||
TODO: добавить отдельную строку для каждого заявленного критерия. PASS без команды и наблюдения не является доказательством. Проверка инфраструктурного hook требует подтверждения, какие реальные процессы/хранилища перезапущены.
|
||||
## Инциденты (диагностика только по сохранённым наблюдениям)
|
||||
|
||||
## Эксперименты и отказы
|
||||
Фоновая нагрузка: ≈32 запроса/с сообщений и ≈25 загрузок/с (мелкие PNG). `research/incidents.py` — оркестрация и запись состояний алертов.
|
||||
|
||||
TODO: гипотеза, hardware/лимиты, версии, объём данных, профиль нагрузки, concurrency/arrival rate, длительность/прогрев, повторы, успешные RPS, ошибки, p50/p95/p99, CPU/RAM. Сохранить raw summary без tokens и объяснить ограничения измерения.
|
||||
| № | Воздействие | Алерт (время обнаружения от воздействия) | Разрешён после восстановления | Как найдена причина |
|
||||
| --- | --- | --- | --- | --- |
|
||||
| 1 | остановка worker | `QueueLagGrowing` через 64 с (порог 30 с + `for 20 s` + шаг rate); также `WorkerSilent`; каскад: переполнение очереди → backpressure 503 на загрузке → `HighHttpErrorRate`/`HighUploadErrorRate` | через 24 с (очередь рассосалась) | возраст старейшей задачи растёт (56.7 с) при **нулевой** скорости `messenger_jobs_processed_total`, серия `messenger_worker_capacity` исчезла → остановлен worker, а не медленная БД; логи API показывают 503 «Очередь обработки заполнена» |
|
||||
| 2 | медленная БД (триггер `pg_sleep(0.8)` на вставку сообщений) | `HighLatencyP95` через 72 с (p95 = 0.80 с) | через 46 с | p95 растёт, а в TraceQL `{ name = "db.connection" && duration > 300ms }` и `{ name =~ "HTTP POST.*messages" && duration > 300ms }` найдено по 20 трасс: почти всё время запроса в спанах БД, CPU процессов не растёт → медленная БД, а не обработчик |
|
||||
| 3 | остановка S3 | `HighUploadErrorRate` через 88 с (доля 5xx на загрузке 32 %; **общая** доля 0.8 % — общий SLI не заметил бы) | через 78 с | `{ status = error }` — спаны `s3.put_object` с ошибкой, 139 логов `s3_put_failed` с `trace_id`; остальные маршруты без ошибок → недоступен S3 |
|
||||
| 4 | остановка Tempo и Loki | `TelemetryExportFailing` через 49 с | через 85 с | бизнес-метрики API без ошибок (ratio 0.0, RPS ≈ 67), растёт `otelcol_exporter_send_failed_spans` (≈66/с), очередь экспорта ограничена (23), RSS Collector 247 MiB → сбой хранилищ телеметрии, не приложения; метрики Prometheus не потеряны (путь Collector → Prometheus независим), трассы/логи за простой потеряны |
|
||||
|
||||
| Воздействие | Что ожидаем | Что наблюдали | Время восстановления / потеря данных |
|
||||
| --- | --- | --- | --- |
|
||||
| TODO | TODO | TODO | TODO |
|
||||
График: `evidence/incidents.png` (возраст очереди, p95, доля 5xx, состояние алертов; серые линии — воздействие, зелёные — восстановление). Время обнаружения включает окно `rate[1m]` — для более быстрой реакции нужны более короткие окна/burn-rate (трек 2, не выбран).
|
||||
|
||||
## Эксперименты (трек 1): цена наблюдаемости
|
||||
|
||||
**Гипотеза.** Телеметрия имеет измеримую цену в CPU API (SDK: спаны, метрики, логи), которая слабо зависит от head-sampling (логи и метрики остаются), но head-sampling теряет редкие ошибки и медленные запросы пропорционально доле выборки; tail-sampling в Collector сохраняет их почти полностью, перенося цену из хранилища в Collector.
|
||||
|
||||
**Методика** (`research/obs_cost.py`, график `evidence/obs_cost.png`). Одинаковая нагрузка во всех профилях: 8 процессов-клиентов, 64 одновременных, 90 % `POST` сообщений + 10 % `GET`, прогрев 5 с, замер 30 с, **2 повтора**; перед профилем пересоздаются Collector, API и worker. В API внедрены отказы (одинаково для всех профилей): 2 % запросов → 500 и 2 % → задержка 0.6 с (`API_FAULT_INJECTION=http500:0.02,slow:0.02:0.6`) — это «редкие ошибки и медленные запросы»; после каждого прогона 25 с на дозапись, затем TraceQL в Tempo считает сохранённые ошибочные (`status = error`) и медленные (`duration > 500ms`) трассы. Число внедрённых отказов считается на стороне клиента. Объём телеметрии — счётчики Collector (`otelcol_receiver_accepted_*`, `otelcol_exporter_sent_spans`). Docker Desktop (WSL2), 2 реплики API по 1 CPU, Collector 384 MiB.
|
||||
|
||||
| Профиль | RPS | p95, мс | CPU API (на реплику) | CPU % / пик RAM MiB Collector | спаны: SDK → Tempo | логи (записей) | сохранено ошибочных трасс | сохранено медленных трасс |
|
||||
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
||||
| off (`OTEL_ENABLED=0`) | 1340 | 74 | 53 % | 0.1 / 39 | — | — | — | — |
|
||||
| on: head 100 % | 1324 | 72 | 87 % | 3.1 / 70 | 96 тыс. → 96 тыс. | 47 тыс. | ≈100 % | ≈100 % |
|
||||
| on: 100 %, очередь SDK 65 536 / пакет 2048 | 1384 | 75 | 84 % | 2.9 / 72 | 96 тыс. → 96 тыс. | 47 тыс. | ≈100 % | ≈100 % |
|
||||
| on: head 10 % | 1400 | 67 | 77 % | 1.7 / 71 | 9 тыс. → 10 тыс. | 49 тыс. | 13 % | 9 % |
|
||||
| on: SDK 100 % + tail-sampling в Collector | 1399 | 67 | 85 % | 4.0 / 110 | 101 тыс. → 14 тыс. | 49 тыс. | ≈100 % | ≈100 % |
|
||||
|
||||
(RPS и p95 — среднее двух прогонов; различия RPS/p95 между профилями ≤ ±5 % — в пределах разброса между повторами; p99 ≈ 0.62 с во всех профилях определяется внедрённой задержкой 0.6 с. Объём — за окно 30 с нагрузки + дозапись. «Сохранено» считается как число трасс с ошибкой/медленных в Tempo к числу внедрённых: значение ≈100 % означает, что потерь не обнаружено; точное отношение немного >100 % из-за 5 с прогрева, попавших в окно поиска.)
|
||||
|
||||
**Результаты и выводы.**
|
||||
1. **Цена в CPU.** Включение телеметрии поднимает CPU API с **53 % до 87 %** одного ядра при той же пропускной способности (+34 п. п., ≈ +60 % CPU на запрос); пропускная способность и p95 на этом профиле не изменились в пределах погрешности, потому что запас CPU у реплик (лимит 100 %) не исчерпан — на насыщенной системе цена перешла бы в потерю RPS. RSS API +≈12 MiB; Collector — 3–4 % CPU и 71 MiB RAM (110 MiB при tail-sampling: буфер трасс).
|
||||
2. **Объём.** 100 %: ≈96 тыс. спанов и ≈48 тыс. записей логов за окно. **Head 10 %** сокращает спаны на 90 % (9.9 тыс.), но **логи (≈49 тыс.) и метрики не меняются**, поэтому CPU API снижается лишь до 77 % (−10 п. п.): большая часть цены — логи и метрики, а не трассы. **Tail-sampling**: SDK по-прежнему шлёт 100 % спанов (CPU API 85 %), но в Tempo уходит ≈15 тыс. из 101 тыс. (−85 % хранения).
|
||||
3. **Потери редких событий.** Head 10 % сохраняет лишь **≈13 % ошибочных и ≈9 % медленных** трасс (ровно доля выборки): 87–91 % редких ошибок и медленных запросов невидимы в Tempo (метрики и логи при этом их фиксируют). 100 % и tail-sampling сохраняют ≈100 % ошибочных и медленных трасс; tail выигрывает по хранилищу при почти той же цене в API.
|
||||
4. **Увеличение очереди SDK** (65 536 / пакет 2048) не изменило ни объём, ни долю сохранённых трасс: потерь спанов на стороне SDK в этой нагрузке не было.
|
||||
5. **Находка по дороге.** В первом прогоне «100 %» сохранялось только ≈57 % ошибочных трасс: Tempo с настройками по умолчанию **отбрасывал ≈29 % спанов** (`tempo_discarded_spans_total{reason="live_traces_exceeded"}`: 224 тыс. из 767 тыс.) из-за лимита «живых» трасс (10 000 на тенанта). Лимиты подняты в `infra/tempo/tempo.yaml` (`overrides.defaults.ingestion`), повторные прогоны — без потерь; цена — память Tempo: при лимите контейнера 512 MiB он был убит OOM (код 137) на тестах после нагрузки, поэтому `mem_limit` Tempo увеличен до 1 GiB. Вывод: потери надо искать метриками самих хранилищ, а не только Collector.
|
||||
|
||||
**Сбалансированная конфигурация (проверена):** SDK 100 % + **tail-sampling** в Collector (все ошибки, все запросы >500 мс, 10 % остальных): редкие проблемные трассы сохраняются полностью, хранилище трасс экономится на 85 %, цена — CPU SDK как при 100 % и ≈ +40 MiB RAM Collector. Если CPU API критичен — снижать объём логов (access-log без health уже отключён) и использовать head-sampling только для трасс успешных запросов.
|
||||
|
||||
**Ограничения методики.** Единичный ноутбук; 2 прогона на профиль; запас CPU API не исчерпан (RPS не падает) — цена показана в CPU, а не в RPS; доля сохранённых трасс оценивается TraceQL-поиском с лимитом 5 000 результатов (для «всех HTTP-трасс» лимит достигнут, для ошибочных/медленных — нет); размер дисковых томов Tempo/Loki не используется как метрика: он меняется компакцией и flush-ом и недостоверен на таких интервалах.
|
||||
|
||||
## Ограничения и следующие шаги
|
||||
|
||||
TODO: какие отказы выдерживаются и какие нет; что не реализовано; где учебное упрощение; какое улучшение подтверждено, а какое пока гипотеза.
|
||||
- Выдерживается: недоступность Tempo/Loki/Collector без влияния на API (данные трасс/логов за простой теряются, метрики — нет). Не выдерживается: потеря volume хранилищ телеметрии; Prometheus/Loki/Tempo — по одному экземпляру.
|
||||
- Алерты локальные (без доставки наружу); окна коротки для демонстрации — для production нужны burn-rate по нескольким окнам.
|
||||
- Логи в stdout остаются в docker json-file (ротация 10 МБ × 3) и дублируются в Loki; `trace_id` присутствует в обоих.
|
||||
- Подтверждено экспериментами: связность трасс через очередь, алерты firing/resolved на 4 отказах, цена телеметрии. Гипотеза без доказательства: поведение при >2 репликах API и при нагрузке порядка тысяч RPS на одном Collector.
|
||||
|
||||
@@ -0,0 +1,337 @@
|
||||
# Стенд lab6: HAProxy -> API x N -> PostgreSQL (метаданные) + SeaweedFS (приватный S3, байты) + Valkey (кэш сессий).
|
||||
# Секреты: scripts/init-secrets.sh создаёт secrets/* (в Git не попадают). Запуск и проверки — см. REPORT.md.
|
||||
name: mtusi-lab6
|
||||
|
||||
x-common: &common
|
||||
restart: unless-stopped
|
||||
logging:
|
||||
driver: json-file
|
||||
options: {max-size: "10m", max-file: "3"}
|
||||
|
||||
services:
|
||||
postgres:
|
||||
<<: *common
|
||||
image: postgres:17.11-alpine@sha256:b0f9560a2de083e2cc7382e75f808c7381a32852a7ec49117deedb300e552b24
|
||||
environment:
|
||||
POSTGRES_USER: messenger
|
||||
POSTGRES_DB: messenger
|
||||
POSTGRES_PASSWORD_FILE: /run/secrets/postgres_password
|
||||
secrets: [postgres_password]
|
||||
volumes:
|
||||
- pgdata:/var/lib/postgresql/data
|
||||
command: ["postgres", "-c", "max_connections=200", "-c", "shared_buffers=128MB", "-c", "log_min_duration_statement=500"]
|
||||
healthcheck:
|
||||
test: ["CMD-SHELL", "pg_isready -U messenger -d messenger"]
|
||||
interval: 5s
|
||||
timeout: 3s
|
||||
retries: 12
|
||||
mem_limit: 512m
|
||||
cpus: ${POSTGRES_CPUS:-1.0}
|
||||
networks: [backend]
|
||||
|
||||
valkey:
|
||||
<<: *common
|
||||
image: valkey/valkey:8.1.10-alpine@sha256:081c2f5cb575efc901aa80ff9cdbd1ec6a301682fd35e1ebb4b0990a4a4a8507
|
||||
command: ["valkey-server", "--appendonly", "yes", "--appendfsync", "everysec",
|
||||
"--maxmemory", "64mb", "--maxmemory-policy", "volatile-ttl", "--save", ""]
|
||||
volumes:
|
||||
- valkeydata:/data
|
||||
healthcheck:
|
||||
test: ["CMD", "valkey-cli", "ping"]
|
||||
interval: 5s
|
||||
timeout: 3s
|
||||
retries: 12
|
||||
mem_limit: 128m
|
||||
cpus: 0.5
|
||||
networks: [backend]
|
||||
|
||||
# Приватное S3-совместимое хранилище SeaweedFS (all-in-one: master + volume + filer + S3-шлюз).
|
||||
# Порты наружу не публикуются: доступ только из сети backend и только по ключам из s3.json (identities).
|
||||
# Анонимный доступ запрещён (при заданных identities запрос без подписи получает 403).
|
||||
seaweedfs:
|
||||
<<: *common
|
||||
image: chrislusf/seaweedfs:4.48@sha256:4e61d15fd35994cb1e43e1e553dff106794841fd9a99ade2fc8c8bfce4d7872d
|
||||
entrypoint: ["/bin/sh", "-c"]
|
||||
command: ["exec weed server -dir=/data -ip=seaweedfs -master.volumeSizeLimitMB=512 -volume.max=30 -filer -s3 -s3.port=8333 -s3.config=/run/secrets/seaweed_s3_config"]
|
||||
secrets: [seaweed_s3_config]
|
||||
volumes:
|
||||
- seaweeddata:/data
|
||||
healthcheck:
|
||||
test: ["CMD-SHELL", "wget -q -O /dev/null http://127.0.0.1:9333/cluster/status && nc -z 127.0.0.1 8333"]
|
||||
interval: 5s
|
||||
timeout: 3s
|
||||
retries: 24
|
||||
start_period: 10s
|
||||
mem_limit: 768m
|
||||
cpus: ${S3_CPUS:-1.0}
|
||||
networks: [backend]
|
||||
|
||||
# Идемпотентный bootstrap: bucket создаётся, если его нет; проверяется, что анонимный доступ запрещён и что ключ API работает.
|
||||
s3-init:
|
||||
build: .
|
||||
image: mtusi-messenger:lab6
|
||||
command: ["python", "-m", "app.s3_bootstrap"]
|
||||
environment:
|
||||
S3_ENDPOINT: http://seaweedfs:8333
|
||||
S3_BUCKET: ${S3_BUCKET:-messenger-attachments}
|
||||
S3_ADMIN_ACCESS_KEY_FILE: /run/secrets/s3_admin_access_key
|
||||
S3_ADMIN_SECRET_KEY_FILE: /run/secrets/s3_admin_secret_key
|
||||
S3_ACCESS_KEY_FILE: /run/secrets/s3_access_key
|
||||
S3_SECRET_KEY_FILE: /run/secrets/s3_secret_key
|
||||
secrets: [s3_admin_access_key, s3_admin_secret_key, s3_access_key, s3_secret_key]
|
||||
depends_on:
|
||||
seaweedfs: {condition: service_healthy}
|
||||
restart: "no"
|
||||
networks: [backend]
|
||||
|
||||
# Брокер: durable quorum-очередь, persistent-сообщения, publisher confirms; данные на volume rabbitmqdata.
|
||||
# hostname фиксирован: имя узла RabbitMQ определяет каталог данных, иначе очередь не переживёт пересоздание контейнера.
|
||||
rabbitmq:
|
||||
<<: *common
|
||||
image: rabbitmq:4.1-alpine@sha256:6a0f7984185cd2cdbf78a184b4c17abe88ecbbda6006a7acf17dbc7e4307f4e2
|
||||
hostname: rabbitmq
|
||||
# RabbitMQ 4.x: RABBITMQ_DEFAULT_USER/PASS объявлены устаревшими -> учётные данные в конфигурационном файле (secret)
|
||||
secrets:
|
||||
- source: rabbitmq_conf
|
||||
target: /etc/rabbitmq/conf.d/20-credentials.conf
|
||||
mode: 0444
|
||||
volumes:
|
||||
- rabbitmqdata:/var/lib/rabbitmq
|
||||
healthcheck:
|
||||
test: ["CMD", "rabbitmq-diagnostics", "-q", "ping"]
|
||||
interval: 10s
|
||||
timeout: 8s
|
||||
retries: 12
|
||||
mem_limit: 512m
|
||||
cpus: ${RABBITMQ_CPUS:-1.0}
|
||||
networks: [backend]
|
||||
|
||||
migrate:
|
||||
build: .
|
||||
image: mtusi-messenger:lab6
|
||||
command: ["python", "-m", "app.migrate"]
|
||||
environment:
|
||||
DATABASE_URL_FILE: /run/secrets/database_url
|
||||
secrets: [database_url]
|
||||
depends_on:
|
||||
postgres: {condition: service_healthy}
|
||||
restart: "no"
|
||||
networks: [backend]
|
||||
|
||||
api:
|
||||
<<: *common
|
||||
build: .
|
||||
image: mtusi-messenger:lab6
|
||||
environment:
|
||||
APP_ENV: ${APP_ENV:-lab}
|
||||
DATABASE_URL_FILE: /run/secrets/database_url
|
||||
VALKEY_URL: redis://valkey:6379/0
|
||||
S3_ENDPOINT: http://seaweedfs:8333
|
||||
S3_BUCKET: ${S3_BUCKET:-messenger-attachments}
|
||||
S3_ACCESS_KEY_FILE: /run/secrets/s3_access_key
|
||||
S3_SECRET_KEY_FILE: /run/secrets/s3_secret_key
|
||||
SESSION_TTL_SECONDS: ${SESSION_TTL_SECONDS:-3600}
|
||||
SESSION_CACHE_TTL_SECONDS: ${SESSION_CACHE_TTL_SECONDS:-60}
|
||||
DB_POOL_MAX: ${DB_POOL_MAX:-10}
|
||||
DRAIN_SECONDS: ${DRAIN_SECONDS:-4}
|
||||
MAX_INFLIGHT: ${MAX_INFLIGHT:-64}
|
||||
ACCESS_LOG: ${ACCESS_LOG:-1}
|
||||
UPLOAD_CONCURRENCY: ${UPLOAD_CONCURRENCY:-4}
|
||||
FAULT_INJECTION: ${API_FAULT_INJECTION:-}
|
||||
BROKER_URL_FILE: /run/secrets/broker_url
|
||||
MAX_QUEUED_JOBS: ${MAX_QUEUED_JOBS:-200}
|
||||
SERVICE_NAME: messenger-api
|
||||
OTEL_ENABLED: ${OTEL_ENABLED:-1}
|
||||
OTEL_EXPORTER_OTLP_ENDPOINT: http://otel-collector:4318
|
||||
OTEL_TRACES_SAMPLER_ARG: ${OTEL_TRACES_SAMPLER_ARG:-1.0}
|
||||
OTEL_QUEUE_SIZE: ${OTEL_QUEUE_SIZE:-2048}
|
||||
OTEL_BATCH_SIZE: ${OTEL_BATCH_SIZE:-512}
|
||||
secrets: [database_url, s3_access_key, s3_secret_key, broker_url]
|
||||
depends_on:
|
||||
migrate: {condition: service_completed_successfully}
|
||||
s3-init: {condition: service_completed_successfully}
|
||||
valkey: {condition: service_healthy}
|
||||
stop_grace_period: 15s
|
||||
mem_limit: ${API_MEM:-256m}
|
||||
cpus: ${API_CPUS:-1.0}
|
||||
deploy:
|
||||
replicas: ${API_REPLICAS:-2}
|
||||
networks: [backend]
|
||||
|
||||
# Worker: отдельный процесс (тот же образ). Берёт задания из RabbitMQ, обрабатывает изображения, пишет результат в S3 и БД.
|
||||
worker:
|
||||
<<: *common
|
||||
build: .
|
||||
image: mtusi-messenger:lab6
|
||||
command: ["python", "-m", "app.worker"]
|
||||
environment:
|
||||
SERVICE_NAME: messenger-worker
|
||||
DATABASE_URL_FILE: /run/secrets/database_url
|
||||
S3_ENDPOINT: http://seaweedfs:8333
|
||||
S3_BUCKET: ${S3_BUCKET:-messenger-attachments}
|
||||
S3_ACCESS_KEY_FILE: /run/secrets/s3_access_key
|
||||
S3_SECRET_KEY_FILE: /run/secrets/s3_secret_key
|
||||
BROKER_URL_FILE: /run/secrets/broker_url
|
||||
WORKER_CONCURRENCY: ${WORKER_CONCURRENCY:-2}
|
||||
MAX_ATTEMPTS: ${MAX_ATTEMPTS:-5}
|
||||
LEASE_SECONDS: ${LEASE_SECONDS:-60}
|
||||
RETRY_BASE_DELAY: ${RETRY_BASE_DELAY:-2}
|
||||
FAULT_INJECTION: ${WORKER_FAULT_INJECTION:-}
|
||||
OTEL_ENABLED: ${OTEL_ENABLED:-1}
|
||||
OTEL_EXPORTER_OTLP_ENDPOINT: http://otel-collector:4318
|
||||
OTEL_TRACES_SAMPLER_ARG: ${OTEL_TRACES_SAMPLER_ARG:-1.0}
|
||||
OTEL_QUEUE_SIZE: ${OTEL_QUEUE_SIZE:-2048}
|
||||
OTEL_BATCH_SIZE: ${OTEL_BATCH_SIZE:-512}
|
||||
ACCESS_LOG: "0"
|
||||
secrets: [database_url, s3_access_key, s3_secret_key, broker_url]
|
||||
depends_on:
|
||||
migrate: {condition: service_completed_successfully}
|
||||
s3-init: {condition: service_completed_successfully}
|
||||
rabbitmq: {condition: service_healthy}
|
||||
healthcheck:
|
||||
test: ["CMD", "python", "-c", "import os,time,sys; sys.exit(0 if time.time()-os.path.getmtime('/tmp/worker.alive') < 30 else 1)"]
|
||||
interval: 10s
|
||||
timeout: 5s
|
||||
retries: 3
|
||||
start_period: 20s
|
||||
stop_grace_period: 30s
|
||||
mem_limit: ${WORKER_MEM:-768m}
|
||||
cpus: ${WORKER_CPUS:-1.0}
|
||||
deploy:
|
||||
replicas: ${WORKER_REPLICAS:-1}
|
||||
networks: [backend]
|
||||
|
||||
# ------------------------------ наблюдаемость (lab6) ------------------------------
|
||||
# OTLP-порты 4317/4318 и метрики Collector 8888/8889 доступны только внутри сети backend (наружу не публикуются).
|
||||
otel-collector:
|
||||
<<: *common
|
||||
image: otel/opentelemetry-collector-contrib:0.161.0@sha256:fd328de2552466ad78385e1b1289c3f2402b1c45f265b252aab1955b42845ac1
|
||||
command: ["--config=${COLLECTOR_CONFIG:-/etc/otelcol/collector.yaml}"]
|
||||
volumes:
|
||||
- ./infra/otel:/etc/otelcol:ro
|
||||
mem_limit: 384m
|
||||
cpus: ${COLLECTOR_CPUS:-1.0}
|
||||
networks: [backend]
|
||||
|
||||
prometheus:
|
||||
<<: *common
|
||||
image: prom/prometheus:v3.15.0@sha256:efd719c99d83b060d9daefdcf00360461adf279f45ef5391f8d111892118753e
|
||||
command: ["--config.file=/etc/prometheus/prometheus.yml", "--storage.tsdb.path=/prometheus", "--storage.tsdb.retention.time=7d",
|
||||
"--web.enable-lifecycle"]
|
||||
volumes:
|
||||
- ./infra/prometheus/prometheus.yml:/etc/prometheus/prometheus.yml:ro
|
||||
- ./infra/prometheus/rules:/etc/prometheus/rules:ro
|
||||
- promdata:/prometheus
|
||||
healthcheck:
|
||||
test: ["CMD", "wget", "-q", "-O", "/dev/null", "http://127.0.0.1:9090/-/ready"]
|
||||
interval: 10s
|
||||
timeout: 3s
|
||||
retries: 6
|
||||
mem_limit: 512m
|
||||
cpus: 1.0
|
||||
networks: [backend]
|
||||
|
||||
loki:
|
||||
<<: *common
|
||||
image: grafana/loki:3.7.8@sha256:1107dd5274e0ada47e42472b7a7e71f3b2a2fe878878108f3e2f9e51528f0193
|
||||
command: ["-config.file=/etc/loki/loki.yaml"]
|
||||
volumes:
|
||||
- ./infra/loki/loki.yaml:/etc/loki/loki.yaml:ro
|
||||
- lokidata:/loki
|
||||
mem_limit: 512m
|
||||
cpus: 1.0
|
||||
networks: [backend]
|
||||
|
||||
tempo:
|
||||
<<: *common
|
||||
image: grafana/tempo:2.9.0@sha256:65a5789759435f1ef696f1953258b9bbdb18eb571d5ce711ff812d2e128288a4
|
||||
command: ["-config.file=/etc/tempo/tempo.yaml"]
|
||||
volumes:
|
||||
- ./infra/tempo/tempo.yaml:/etc/tempo/tempo.yaml:ro
|
||||
- tempodata:/var/tempo
|
||||
mem_limit: 1g
|
||||
cpus: 1.0
|
||||
networks: [backend]
|
||||
|
||||
# Grafana: единственный опубликованный сервис наблюдаемости, только 127.0.0.1:3000. Datasources, dashboard и alert-вью
|
||||
# подключаются provisioning-ом из Git (infra/grafana). Анонимный доступ выключен; пароль admin — secret.
|
||||
grafana:
|
||||
<<: *common
|
||||
image: grafana/grafana:13.2.3@sha256:b28bae15e219c998fb0e0424ed724930cc61b1f61fb404d47c862f9a23f9e572
|
||||
environment:
|
||||
GF_SECURITY_ADMIN_USER: admin
|
||||
GF_SECURITY_ADMIN_PASSWORD__FILE: /run/secrets/grafana_admin_password
|
||||
GF_AUTH_ANONYMOUS_ENABLED: "false"
|
||||
GF_USERS_ALLOW_SIGN_UP: "false"
|
||||
GF_ANALYTICS_REPORTING_ENABLED: "false"
|
||||
GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH: /var/lib/grafana/dashboards/messenger.json
|
||||
secrets: [grafana_admin_password]
|
||||
volumes:
|
||||
- ./infra/grafana/provisioning:/etc/grafana/provisioning:ro
|
||||
- ./infra/grafana/dashboards:/var/lib/grafana/dashboards:ro
|
||||
- grafanadata:/var/lib/grafana
|
||||
depends_on: [prometheus, loki, tempo]
|
||||
ports:
|
||||
- "127.0.0.1:3000:3000"
|
||||
healthcheck:
|
||||
test: ["CMD-SHELL", "wget -q -O /dev/null http://127.0.0.1:3000/api/health"]
|
||||
interval: 10s
|
||||
timeout: 3s
|
||||
retries: 12
|
||||
mem_limit: 512m
|
||||
cpus: 1.0
|
||||
networks: [backend]
|
||||
|
||||
lb:
|
||||
<<: *common
|
||||
image: haproxy:3.0-alpine@sha256:56b887da77428b7a6621e59e480cdbd330cc805c22d3cedb66ceea76ffdea2c6
|
||||
volumes:
|
||||
- ./infra/haproxy/haproxy.cfg:/usr/local/etc/haproxy/haproxy.cfg:ro
|
||||
depends_on:
|
||||
- api
|
||||
ports:
|
||||
- "127.0.0.1:8080:8080"
|
||||
healthcheck:
|
||||
test: ["CMD-SHELL", "wget -q -O /dev/null http://127.0.0.1:8081/healthz"]
|
||||
interval: 5s
|
||||
timeout: 3s
|
||||
retries: 6
|
||||
mem_limit: 128m
|
||||
cpus: ${LB_CPUS:-1.0}
|
||||
stop_grace_period: 10s
|
||||
networks: [backend]
|
||||
|
||||
secrets:
|
||||
postgres_password:
|
||||
file: ./secrets/postgres_password
|
||||
database_url:
|
||||
file: ./secrets/database_url
|
||||
seaweed_s3_config:
|
||||
file: ./secrets/seaweed_s3.json
|
||||
s3_admin_access_key:
|
||||
file: ./secrets/s3_admin_access_key
|
||||
s3_admin_secret_key:
|
||||
file: ./secrets/s3_admin_secret_key
|
||||
s3_access_key:
|
||||
file: ./secrets/s3_access_key
|
||||
s3_secret_key:
|
||||
file: ./secrets/s3_secret_key
|
||||
rabbitmq_conf:
|
||||
file: ./secrets/rabbitmq.conf
|
||||
grafana_admin_password:
|
||||
file: ./secrets/grafana_admin_password
|
||||
broker_url:
|
||||
file: ./secrets/broker_url
|
||||
|
||||
volumes:
|
||||
pgdata:
|
||||
valkeydata:
|
||||
seaweeddata:
|
||||
rabbitmqdata:
|
||||
promdata:
|
||||
lokidata:
|
||||
tempodata:
|
||||
grafanadata:
|
||||
|
||||
networks:
|
||||
backend:
|
||||
@@ -0,0 +1,34 @@
|
||||
# Runbooks (локальный стенд lab6)
|
||||
|
||||
Grafana — `http://127.0.0.1:3000` (пользователь `admin`, пароль — `secrets/grafana_admin_password`), dashboard «Messenger: API, очередь, worker».
|
||||
Prometheus/Loki/Tempo доступны только внутри сети стенда: запросы — через Grafana (Explore) либо `docker compose exec prometheus wget -qO- http://loki:3100/...`.
|
||||
|
||||
Общий порядок диагностики: **метрика → трасса → лог**. Метрика показывает *что* и *когда* (алерт, dashboard), трасса — *где* потрачено время/где ошибка
|
||||
(по `trace_id` из лога или из Tempo), лог — *почему* (событие и безопасные поля без тел сообщений и токенов).
|
||||
Связка лог → трасса: в Loki у каждой записи есть `trace_id`/`span_id` (в Grafana — ссылка «Tempo»); трасса → логи: кнопка «Logs for this span».
|
||||
|
||||
## HighErrorRate
|
||||
Алерт `HighHttpErrorRate`: доля ответов 5xx среди бизнес-запросов (без `/health`) > 5 % при ≥ 0.5 запроса/с, 20 с. 4xx (неверный пароль, 404) в SLI не входят.
|
||||
1. Dashboard → «RPS по статус-кодам» и «Доля ошибок»: какой маршрут даёт 5xx (`http_route` в `sum by (http_route)`).
|
||||
2. Tempo: TraceQL `{ status = error && resource.service.name = "messenger-api" }` — в трассе виден спан с ошибкой (`s3.put_object`, `db.connection`, …) и `http.response.status_code`.
|
||||
3. Loki: `{service_name="messenger-api"} | status >= 500` или поиск событий `s3_put_failed`, `db_unavailable`, `unhandled_exception` (поле `error_type`).
|
||||
4. Действия: S3 недоступен → `docker compose ps seaweedfs`, `docker compose start seaweedfs`; БД → `docker compose ps postgres`, readiness реплик; затем алерт возвращается в resolved через ~1 мин (окно `rate[1m]`).
|
||||
|
||||
## HighLatency
|
||||
Алерт `HighLatencyP95`: p95 всех бизнес-запросов (из суммарной гистограммы всех реплик) > 0.5 с.
|
||||
1. Dashboard «Latency p50/p95/p99» и «Latency p95 по маршрутам» → какой маршрут.
|
||||
2. Отличить **медленную БД** от **медленного обработчика**: «DB pool: занято/ожидают» (рост `waiting`) и длительность спанов `db.connection` в медленной трассе (`{ name = "db.connection" && duration > 300ms }`). Если почти всё время запроса внутри `db.connection` — БД; если время в самом серверном спане вне дочерних — код приложения/CPU (`process_cpu_time`).
|
||||
3. БД: `log_min_duration_statement=500` пишет медленные запросы в лог `postgres`; проверить блокировки/триггеры (`pg_stat_activity`).
|
||||
|
||||
## QueueLag
|
||||
Алерт `QueueLagGrowing`: возраст старейшего задания (`queued`/`processing`) > 30 с, 20 с.
|
||||
1. Dashboard «Возраст старейшей задачи», «Глубина очереди», «Worker: занято/ёмкость», «Задания по исходам».
|
||||
2. Нет метрик worker / `busy=0` и растёт возраст → worker остановлен или не подключён к брокеру: `docker compose ps worker rabbitmq`, `docker compose logs worker`. `busy == capacity` при росте очереди → не хватает мощности (увеличить `WORKER_CONCURRENCY`/реплики, см. отчёт lab5).
|
||||
3. Много `retried`/`failed` → причина в логах worker (`job_rescheduled`, `reason`) и в трассе consumer-спана (S3/БД). Реестр: `docker compose run --rm -T worker python -m app.replay list`.
|
||||
4. После устранения причины очередь рассасывается, алерт переходит в resolved.
|
||||
|
||||
## Telemetry
|
||||
Алерт `TelemetryExportFailing`: Collector не может отправить данные в Tempo/Loki.
|
||||
1. Это сбой наблюдаемости, а не приложения: бизнес-метрики API (ошибки, RPS) не меняются, приложение работает (SDK и Collector ограничивают очередь/RAM и отбрасывают данные).
|
||||
2. Проверить: `docker compose ps tempo loki`; метрики Collector `otelcol_exporter_queue_size`, `otelcol_exporter_send_failed_*`, `otelcol_process_memory_rss`.
|
||||
3. Запустить хранилища: `docker compose start tempo loki`. Данные за время простоя трасс/логов потеряны (счётчики `send_failed`), метрики не затронуты (путь Collector → Prometheus независим).
|
||||
@@ -0,0 +1,66 @@
|
||||
$ python3 scripts/check_contract.py
|
||||
PASS: offline skeleton structure, local refs and Python syntax; application NOT tested.
|
||||
$ python3 tests/smoke.py (телеметрия включена)
|
||||
PASS: lab6 public API scenario; infrastructure and grading evidence remain separate.
|
||||
# контейнеры до перезапуска:
|
||||
/mtusi-lab6-api-2 StartedAt=2026-10-03T02:04:24.831646035Z
|
||||
/mtusi-lab6-api-1 StartedAt=2026-10-03T02:04:24.660308554Z
|
||||
/mtusi-lab6-worker-1 StartedAt=2026-10-03T02:04:23.350829139Z
|
||||
/mtusi-lab6-rabbitmq-1 StartedAt=2026-10-03T02:04:19.141968204Z
|
||||
/mtusi-lab6-seaweedfs-1 StartedAt=2026-10-03T02:04:19.613928237Z
|
||||
/mtusi-lab6-postgres-1 StartedAt=2026-10-03T02:04:18.665413808Z
|
||||
/mtusi-lab6-valkey-1 StartedAt=2026-10-03T02:04:18.435301589Z
|
||||
/mtusi-lab6-otel-collector-1 StartedAt=2026-10-03T02:04:19.399662403Z
|
||||
/mtusi-lab6-prometheus-1 StartedAt=2026-10-03T02:04:16.212964886Z
|
||||
/mtusi-lab6-loki-1 StartedAt=2026-10-03T02:04:19.815321273Z
|
||||
/mtusi-lab6-tempo-1 StartedAt=2026-10-03T02:04:18.87934529Z
|
||||
/mtusi-lab6-grafana-1 StartedAt=2026-10-03T02:04:18.083530623Z
|
||||
$ python3 tests/resilience.py --mode persistence --action-script scripts/restart.sh (Windows: restart.cmd; перезапускаются API, worker, брокер, S3, БД, кэш и стек наблюдаемости)
|
||||
Executing explicit persistence hook; hook output suppressed to avoid leaking secrets.
|
||||
PASS: persistence preserved confirmed data and existing session. Hook scope needs operator evidence.
|
||||
# контейнеры после перезапуска:
|
||||
/mtusi-lab6-api-2 StartedAt=2026-10-03T02:05:18.916843054Z
|
||||
/mtusi-lab6-api-1 StartedAt=2026-10-03T02:05:20.476426854Z
|
||||
/mtusi-lab6-worker-1 StartedAt=2026-10-03T02:05:19.075989142Z
|
||||
/mtusi-lab6-rabbitmq-1 StartedAt=2026-10-03T02:05:15.668495976Z
|
||||
/mtusi-lab6-seaweedfs-1 StartedAt=2026-10-03T02:05:15.807584366Z
|
||||
/mtusi-lab6-postgres-1 StartedAt=2026-10-03T02:05:14.948208653Z
|
||||
/mtusi-lab6-valkey-1 StartedAt=2026-10-03T02:05:14.108577454Z
|
||||
/mtusi-lab6-otel-collector-1 StartedAt=2026-10-03T02:05:13.357193004Z
|
||||
/mtusi-lab6-prometheus-1 StartedAt=2026-10-03T02:05:14.678083736Z
|
||||
/mtusi-lab6-loki-1 StartedAt=2026-10-03T02:05:15.497739002Z
|
||||
/mtusi-lab6-tempo-1 StartedAt=2026-10-03T02:05:15.099095895Z
|
||||
/mtusi-lab6-grafana-1 StartedAt=2026-10-03T02:05:15.320432594Z
|
||||
$ failover (1-я реплика)
|
||||
Executing explicit failover hook; hook output suppressed to avoid leaking secrets.
|
||||
{"observed_instances_before": 2, "successful_writes": 112, "failed_attempts": 0, "first_recovery_seconds": 2.36}
|
||||
PASS: failover preserved confirmed data and existing session. Hook scope needs operator evidence.
|
||||
$ failover (2-я реплика)
|
||||
Executing explicit failover hook; hook output suppressed to avoid leaking secrets.
|
||||
{"observed_instances_before": 2, "successful_writes": 109, "failed_attempts": 0, "first_recovery_seconds": 2.375}
|
||||
PASS: failover preserved confirmed data and existing session. Hook scope needs operator evidence.
|
||||
$ python -m unittest test_api.py
|
||||
Ran 12 tests in 2.950s
|
||||
OK
|
||||
$ python -m unittest test_idempotency.py
|
||||
Ran 9 tests in 2.070s
|
||||
OK
|
||||
$ python -m unittest test_attachments.py
|
||||
Ran 11 tests in 6.104s
|
||||
OK
|
||||
$ python student-tests/test_pipeline.py (Pillow)
|
||||
Ran 9 tests in 11.272s
|
||||
OK
|
||||
$ python student-tests/test_topology.py
|
||||
Ran 6 tests in 2.857s
|
||||
OK
|
||||
$ python student-tests/test_observability.py (метрики, логи, трассы, связь API->worker, нет секретов)
|
||||
Ran 5 tests in 7.658s
|
||||
OK
|
||||
$ python student-tests/test_observability_retry.py (все попытки задания — одна трасса)
|
||||
Ran 1 test in 18.312s
|
||||
OK
|
||||
$ python student-tests/test_resilience5.py (окна отказа конвейера, телеметрия включена)
|
||||
Ran 7 tests in 139.292s
|
||||
OK
|
||||
CHECKS-DONE
|
||||
@@ -0,0 +1,35 @@
|
||||
$ curl -u admin:*** http://127.0.0.1:3000/api/search (dashboard импортирован provisioning-ом, а не вручную)
|
||||
dash-folder Messenger eg028qble26m8e
|
||||
dash-db Messenger: API, очередь, worker messenger-overview
|
||||
$ curl -u admin:*** .../api/datasources (datasources из infra/grafana/provisioning)
|
||||
Loki loki loki provisioned= True
|
||||
Prometheus prometheus prometheus provisioned= True
|
||||
Tempo tempo tempo provisioned= True
|
||||
health prometheus: OK
|
||||
health loki: OK
|
||||
health tempo: OK
|
||||
$ curl -u admin:*** .../api/dashboards/uid/messenger-overview (панели)
|
||||
provisioned: True | �������: 22
|
||||
- RPS по маршрутам
|
||||
- Доля ошибок 5xx (SLI)
|
||||
- Latency p50 / p95 / p99
|
||||
- RPS по статус-кодам
|
||||
- Latency p95 по маршрутам
|
||||
- DB pool: занято / простаивает / ожидают
|
||||
- Задания по исходам
|
||||
- Длительность обработки p50 / p95 (worker)
|
||||
- Время до ready p50 / p95
|
||||
- Глубина очереди (queued + processing)
|
||||
- Возраст старейшей задачи
|
||||
- Worker: занято / ёмкость
|
||||
- RSS по сервисам
|
||||
- CPU по сервисам (ядра)
|
||||
- Collector: очередь экспорта и отказы
|
||||
- Структурные логи (API и worker), ошибки и предупреждения
|
||||
- Алерты (firing / pending)
|
||||
- Последние трассы API (Tempo)
|
||||
$ анонимный доступ выключен:
|
||||
GET /api/search ��� ������ -> HTTP 401
|
||||
$ порты наружу (docker compose ps):
|
||||
grafana 127.0.0.1:3000->3000/tcp
|
||||
lb 127.0.0.1:8080->8080/tcp
|
||||
|
After Width: | Height: | Size: 101 KiB |
|
After Width: | Height: | Size: 67 KiB |
@@ -0,0 +1,115 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Генератор dashboard JSON (infra/grafana/dashboards/messenger.json). Запуск: python infra/grafana/build_dashboard.py
|
||||
Dashboard хранится в Git как JSON и импортируется provisioning-ом автоматически; генератор — только удобство правки."""
|
||||
import json
|
||||
import os
|
||||
|
||||
PROM = {"type": "prometheus", "uid": "prometheus"}
|
||||
LOKI = {"type": "loki", "uid": "loki"}
|
||||
TEMPO = {"type": "tempo", "uid": "tempo"}
|
||||
panel_id = [0]
|
||||
|
||||
|
||||
def _id():
|
||||
panel_id[0] += 1
|
||||
return panel_id[0]
|
||||
|
||||
|
||||
def stat_or_ts(title, exprs, x, y, w=8, h=7, unit="short", kind="timeseries", desc="", thresholds=None, ds=PROM, legends=None):
|
||||
p = {"id": _id(), "type": kind, "title": title, "description": desc, "datasource": ds,
|
||||
"gridPos": {"x": x, "y": y, "w": w, "h": h},
|
||||
"targets": [{"refId": chr(65 + i), "expr": e, "datasource": ds, "legendFormat": (legends or [""] * len(exprs))[i]}
|
||||
for i, e in enumerate(exprs)],
|
||||
"fieldConfig": {"defaults": {"unit": unit, "custom": {"lineWidth": 1, "fillOpacity": 10}}, "overrides": []},
|
||||
"options": {"legend": {"displayMode": "list", "placement": "bottom"}}}
|
||||
if thresholds:
|
||||
p["fieldConfig"]["defaults"]["thresholds"] = {"mode": "absolute", "steps": thresholds}
|
||||
return p
|
||||
|
||||
|
||||
def row(title, y):
|
||||
return {"id": _id(), "type": "row", "title": title, "gridPos": {"x": 0, "y": y, "w": 24, "h": 1}, "collapsed": False, "panels": []}
|
||||
|
||||
|
||||
NB = '{http_route!~"/health.*"}'
|
||||
panels = []
|
||||
y = 0
|
||||
panels.append(row("API (бизнес-запросы без /health) — запросы, ошибки, задержки", y)); y += 1
|
||||
panels += [
|
||||
stat_or_ts("RPS по маршрутам", ['sum by (http_route) (rate(http_server_request_duration_seconds_count%s[1m]))' % NB], 0, y,
|
||||
unit="reqps", legends=["{{http_route}}"], desc="Единицы: запросов в секунду. Источник: http_server_request_duration_seconds_count."),
|
||||
stat_or_ts("Доля ошибок 5xx (SLI)", ['sli:http_error_ratio:1m'], 8, y, unit="percentunit",
|
||||
desc="Числитель: 5xx; знаменатель: все запросы кроме /health. 4xx (в т.ч. неверный пароль) не считаются отказом сервиса.",
|
||||
thresholds=[{"color": "green", "value": None}, {"color": "red", "value": 0.05}]),
|
||||
stat_or_ts("Latency p50 / p95 / p99", [
|
||||
'histogram_quantile(0.50, sum by (le) (rate(http_server_request_duration_seconds_bucket%s[1m])))' % NB,
|
||||
'histogram_quantile(0.95, sum by (le) (rate(http_server_request_duration_seconds_bucket%s[1m])))' % NB,
|
||||
'histogram_quantile(0.99, sum by (le) (rate(http_server_request_duration_seconds_bucket%s[1m])))' % NB],
|
||||
16, y, unit="s", legends=["p50", "p95", "p99"],
|
||||
desc="Процентили считаются из суммарной гистограммы всех реплик (sum by le), а не усреднением p95 реплик."),
|
||||
]
|
||||
y += 7
|
||||
panels += [
|
||||
stat_or_ts("RPS по статус-кодам", ['sum by (http_response_status_code) (rate(http_server_request_duration_seconds_count%s[1m]))' % NB],
|
||||
0, y, unit="reqps", legends=["{{http_response_status_code}}"]),
|
||||
stat_or_ts("Latency p95 по маршрутам", ['histogram_quantile(0.95, sum by (le, http_route) (rate(http_server_request_duration_seconds_bucket%s[1m])))' % NB],
|
||||
8, y, unit="s", legends=["{{http_route}}"]),
|
||||
stat_or_ts("DB pool: занято / простаивает / ожидают", ['sum(db_client_connections_usage{state="used"})', 'sum(db_client_connections_usage{state="idle"})',
|
||||
'sum(db_client_connections_waiting)'], 16, y, unit="short",
|
||||
legends=["used", "idle", "waiting"], desc="Пул соединений PostgreSQL по всем репликам API. Рост waiting — насыщение пула."),
|
||||
]
|
||||
y += 7
|
||||
panels.append(row("Задания (worker) и очередь", y)); y += 1
|
||||
panels += [
|
||||
stat_or_ts("Задания по исходам", ['sum by (outcome) (rate(messenger_jobs_processed_total[1m]))'], 0, y, unit="ops", legends=["{{outcome}}"],
|
||||
desc="ready — успешно; failed — финальная ошибка; retried — временная ошибка, запланирован повтор; noop — повторная доставка."),
|
||||
stat_or_ts("Длительность обработки p50 / p95 (worker)", [
|
||||
'histogram_quantile(0.50, sum by (le) (rate(messenger_job_duration_seconds_bucket[1m])))',
|
||||
'histogram_quantile(0.95, sum by (le) (rate(messenger_job_duration_seconds_bucket[1m])))'], 8, y, unit="s", legends=["p50", "p95"]),
|
||||
stat_or_ts("Время до ready p50 / p95", [
|
||||
'histogram_quantile(0.50, sum by (le) (rate(messenger_job_time_to_ready_seconds_bucket[1m])))',
|
||||
'histogram_quantile(0.95, sum by (le) (rate(messenger_job_time_to_ready_seconds_bucket[1m])))'], 16, y, unit="s", legends=["p50", "p95"],
|
||||
desc="От устойчивого приёма загрузки (202) до ready. Включает ожидание в очереди."),
|
||||
]
|
||||
y += 7
|
||||
panels += [
|
||||
stat_or_ts("Глубина очереди (queued + processing)", ['max(messenger_queue_depth)'], 0, y, unit="short", legends=["jobs"]),
|
||||
stat_or_ts("Возраст старейшей задачи", ['max(messenger_queue_oldest_job_age_seconds)'], 8, y, unit="s", legends=["oldest"],
|
||||
thresholds=[{"color": "green", "value": None}, {"color": "red", "value": 30}],
|
||||
desc="Backlog — не задержка обработки: показывает, как долго ждёт самое старое задание."),
|
||||
stat_or_ts("Worker: занято / ёмкость", ['sum(messenger_worker_busy)', 'sum(messenger_worker_capacity)'], 16, y, legends=["busy", "capacity"],
|
||||
desc="Насыщение worker: busy == capacity при растущей очереди означает нехватку мощности."),
|
||||
]
|
||||
y += 7
|
||||
panels.append(row("Ресурсы процессов", y)); y += 1
|
||||
panels += [
|
||||
stat_or_ts("RSS по сервисам", ['sum by (service_name) (process_memory_rss_bytes)'], 0, y, unit="bytes", legends=["{{service_name}}"]),
|
||||
stat_or_ts("CPU по сервисам (ядра)", ['sum by (service_name) (rate(process_cpu_time_seconds_total[1m]))'], 8, y, unit="short", legends=["{{service_name}}"],
|
||||
desc="rate накопленного CPU-времени процесса: 1.0 = одно ядро."),
|
||||
stat_or_ts("Collector: очередь экспорта и отказы", ['sum by (exporter) (otelcol_exporter_queue_size)',
|
||||
'sum by (exporter) (rate(otelcol_exporter_send_failed_spans[1m]))',
|
||||
'sum(rate(otelcol_processor_memory_limiter_refused_spans[1m]))'], 16, y, legends=["queue {{exporter}}", "failed spans {{exporter}}", "refused (memory_limiter)"],
|
||||
desc="Поведение Collector при недоступном backend: ограниченная очередь, отказы, memory_limiter."),
|
||||
]
|
||||
y += 7
|
||||
panels.append(row("Логи и алерты", y)); y += 1
|
||||
panels.append({"id": _id(), "type": "logs", "title": "Структурные логи (API и worker), ошибки и предупреждения", "datasource": LOKI,
|
||||
"gridPos": {"x": 0, "y": y, "w": 16, "h": 10},
|
||||
"targets": [{"refId": "A", "datasource": LOKI, "expr": '{service_name=~"messenger-.*"} | detected_level=~"warn|warning|error|ERROR|WARN"'}],
|
||||
"options": {"showTime": True, "wrapLogMessage": True, "enableLogDetails": True, "sortOrder": "Descending"}})
|
||||
panels.append({"id": _id(), "type": "alertlist", "title": "Алерты (firing / pending)", "datasource": PROM,
|
||||
"gridPos": {"x": 16, "y": y, "w": 8, "h": 10},
|
||||
"options": {"alertInstanceLabelFilter": "", "alertName": "", "dashboardAlerts": False, "groupBy": [], "groupMode": "default",
|
||||
"maxItems": 20, "sortOrder": 1, "stateFilter": {"firing": True, "pending": True, "noData": True, "normal": False, "error": True}}})
|
||||
y += 10
|
||||
panels.append({"id": _id(), "type": "traces", "title": "Последние трассы API (Tempo)", "datasource": TEMPO,
|
||||
"gridPos": {"x": 0, "y": y, "w": 24, "h": 8},
|
||||
"targets": [{"refId": "A", "datasource": TEMPO, "queryType": "traceql", "query": '{ resource.service.name = "messenger-api" }', "limit": 20}]})
|
||||
|
||||
dashboard = {"uid": "messenger-overview", "title": "Messenger: API, очередь, worker", "tags": ["messenger", "lab6"], "timezone": "browser",
|
||||
"schemaVersion": 39, "version": 1, "refresh": "5s", "time": {"from": "now-15m", "to": "now"}, "panels": panels,
|
||||
"annotations": {"list": []}, "templating": {"list": []}}
|
||||
out = os.path.join(os.path.dirname(os.path.abspath(__file__)), "dashboards", "messenger.json")
|
||||
with open(out, "w", encoding="utf-8") as fh:
|
||||
json.dump(dashboard, fh, ensure_ascii=False, indent=1)
|
||||
print("saved", out)
|
||||
@@ -0,0 +1,917 @@
|
||||
{
|
||||
"uid": "messenger-overview",
|
||||
"title": "Messenger: API, очередь, worker",
|
||||
"tags": [
|
||||
"messenger",
|
||||
"lab6"
|
||||
],
|
||||
"timezone": "browser",
|
||||
"schemaVersion": 39,
|
||||
"version": 1,
|
||||
"refresh": "5s",
|
||||
"time": {
|
||||
"from": "now-15m",
|
||||
"to": "now"
|
||||
},
|
||||
"panels": [
|
||||
{
|
||||
"id": 1,
|
||||
"type": "row",
|
||||
"title": "API (бизнес-запросы без /health) — запросы, ошибки, задержки",
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 0,
|
||||
"w": 24,
|
||||
"h": 1
|
||||
},
|
||||
"collapsed": false,
|
||||
"panels": []
|
||||
},
|
||||
{
|
||||
"id": 2,
|
||||
"type": "timeseries",
|
||||
"title": "RPS по маршрутам",
|
||||
"description": "Единицы: запросов в секунду. Источник: http_server_request_duration_seconds_count.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 1,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (http_route) (rate(http_server_request_duration_seconds_count{http_route!~\"/health.*\"}[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{http_route}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "reqps",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 3,
|
||||
"type": "timeseries",
|
||||
"title": "Доля ошибок 5xx (SLI)",
|
||||
"description": "Числитель: 5xx; знаменатель: все запросы кроме /health. 4xx (в т.ч. неверный пароль) не считаются отказом сервиса.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 8,
|
||||
"y": 1,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sli:http_error_ratio:1m",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": ""
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "percentunit",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
},
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
},
|
||||
{
|
||||
"color": "red",
|
||||
"value": 0.05
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 4,
|
||||
"type": "timeseries",
|
||||
"title": "Latency p50 / p95 / p99",
|
||||
"description": "Процентили считаются из суммарной гистограммы всех реплик (sum by le), а не усреднением p95 реплик.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 1,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "histogram_quantile(0.50, sum by (le) (rate(http_server_request_duration_seconds_bucket{http_route!~\"/health.*\"}[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p50"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "histogram_quantile(0.95, sum by (le) (rate(http_server_request_duration_seconds_bucket{http_route!~\"/health.*\"}[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p95"
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"expr": "histogram_quantile(0.99, sum by (le) (rate(http_server_request_duration_seconds_bucket{http_route!~\"/health.*\"}[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p99"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "s",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 5,
|
||||
"type": "timeseries",
|
||||
"title": "RPS по статус-кодам",
|
||||
"description": "",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 8,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (http_response_status_code) (rate(http_server_request_duration_seconds_count{http_route!~\"/health.*\"}[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{http_response_status_code}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "reqps",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 6,
|
||||
"type": "timeseries",
|
||||
"title": "Latency p95 по маршрутам",
|
||||
"description": "",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 8,
|
||||
"y": 8,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "histogram_quantile(0.95, sum by (le, http_route) (rate(http_server_request_duration_seconds_bucket{http_route!~\"/health.*\"}[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{http_route}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "s",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 7,
|
||||
"type": "timeseries",
|
||||
"title": "DB pool: занято / простаивает / ожидают",
|
||||
"description": "Пул соединений PostgreSQL по всем репликам API. Рост waiting — насыщение пула.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 8,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(db_client_connections_usage{state=\"used\"})",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "used"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "sum(db_client_connections_usage{state=\"idle\"})",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "idle"
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"expr": "sum(db_client_connections_waiting)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "waiting"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 8,
|
||||
"type": "row",
|
||||
"title": "Задания (worker) и очередь",
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 15,
|
||||
"w": 24,
|
||||
"h": 1
|
||||
},
|
||||
"collapsed": false,
|
||||
"panels": []
|
||||
},
|
||||
{
|
||||
"id": 9,
|
||||
"type": "timeseries",
|
||||
"title": "Задания по исходам",
|
||||
"description": "ready — успешно; failed — финальная ошибка; retried — временная ошибка, запланирован повтор; noop — повторная доставка.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 16,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (outcome) (rate(messenger_jobs_processed_total[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{outcome}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "ops",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 10,
|
||||
"type": "timeseries",
|
||||
"title": "Длительность обработки p50 / p95 (worker)",
|
||||
"description": "",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 8,
|
||||
"y": 16,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "histogram_quantile(0.50, sum by (le) (rate(messenger_job_duration_seconds_bucket[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p50"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "histogram_quantile(0.95, sum by (le) (rate(messenger_job_duration_seconds_bucket[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p95"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "s",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 11,
|
||||
"type": "timeseries",
|
||||
"title": "Время до ready p50 / p95",
|
||||
"description": "От устойчивого приёма загрузки (202) до ready. Включает ожидание в очереди.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 16,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "histogram_quantile(0.50, sum by (le) (rate(messenger_job_time_to_ready_seconds_bucket[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p50"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "histogram_quantile(0.95, sum by (le) (rate(messenger_job_time_to_ready_seconds_bucket[1m])))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "p95"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "s",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 12,
|
||||
"type": "timeseries",
|
||||
"title": "Глубина очереди (queued + processing)",
|
||||
"description": "",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 23,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "max(messenger_queue_depth)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "jobs"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 13,
|
||||
"type": "timeseries",
|
||||
"title": "Возраст старейшей задачи",
|
||||
"description": "Backlog — не задержка обработки: показывает, как долго ждёт самое старое задание.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 8,
|
||||
"y": 23,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "max(messenger_queue_oldest_job_age_seconds)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "oldest"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "s",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
},
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{
|
||||
"color": "green",
|
||||
"value": null
|
||||
},
|
||||
{
|
||||
"color": "red",
|
||||
"value": 30
|
||||
}
|
||||
]
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 14,
|
||||
"type": "timeseries",
|
||||
"title": "Worker: занято / ёмкость",
|
||||
"description": "Насыщение worker: busy == capacity при растущей очереди означает нехватку мощности.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 23,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(messenger_worker_busy)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "busy"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "sum(messenger_worker_capacity)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "capacity"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 15,
|
||||
"type": "row",
|
||||
"title": "Ресурсы процессов",
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 30,
|
||||
"w": 24,
|
||||
"h": 1
|
||||
},
|
||||
"collapsed": false,
|
||||
"panels": []
|
||||
},
|
||||
{
|
||||
"id": 16,
|
||||
"type": "timeseries",
|
||||
"title": "RSS по сервисам",
|
||||
"description": "",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 31,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (service_name) (process_memory_rss_bytes)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{service_name}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "bytes",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 17,
|
||||
"type": "timeseries",
|
||||
"title": "CPU по сервисам (ядра)",
|
||||
"description": "rate накопленного CPU-времени процесса: 1.0 = одно ядро.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 8,
|
||||
"y": 31,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (service_name) (rate(process_cpu_time_seconds_total[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "{{service_name}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 18,
|
||||
"type": "timeseries",
|
||||
"title": "Collector: очередь экспорта и отказы",
|
||||
"description": "Поведение Collector при недоступном backend: ограниченная очередь, отказы, memory_limiter.",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 31,
|
||||
"w": 8,
|
||||
"h": 7
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (exporter) (otelcol_exporter_queue_size)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "queue {{exporter}}"
|
||||
},
|
||||
{
|
||||
"refId": "B",
|
||||
"expr": "sum by (exporter) (rate(otelcol_exporter_send_failed_spans[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "failed spans {{exporter}}"
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"expr": "sum(rate(otelcol_processor_memory_limiter_refused_spans[1m]))",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"legendFormat": "refused (memory_limiter)"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"custom": {
|
||||
"lineWidth": 1,
|
||||
"fillOpacity": 10
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": {
|
||||
"displayMode": "list",
|
||||
"placement": "bottom"
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 19,
|
||||
"type": "row",
|
||||
"title": "Логи и алерты",
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 38,
|
||||
"w": 24,
|
||||
"h": 1
|
||||
},
|
||||
"collapsed": false,
|
||||
"panels": []
|
||||
},
|
||||
{
|
||||
"id": 20,
|
||||
"type": "logs",
|
||||
"title": "Структурные логи (API и worker), ошибки и предупреждения",
|
||||
"datasource": {
|
||||
"type": "loki",
|
||||
"uid": "loki"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 39,
|
||||
"w": 16,
|
||||
"h": 10
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"datasource": {
|
||||
"type": "loki",
|
||||
"uid": "loki"
|
||||
},
|
||||
"expr": "{service_name=~\"messenger-.*\"} | detected_level=~\"warn|warning|error|ERROR|WARN\""
|
||||
}
|
||||
],
|
||||
"options": {
|
||||
"showTime": true,
|
||||
"wrapLogMessage": true,
|
||||
"enableLogDetails": true,
|
||||
"sortOrder": "Descending"
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 21,
|
||||
"type": "alertlist",
|
||||
"title": "Алерты (firing / pending)",
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "prometheus"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 16,
|
||||
"y": 39,
|
||||
"w": 8,
|
||||
"h": 10
|
||||
},
|
||||
"options": {
|
||||
"alertInstanceLabelFilter": "",
|
||||
"alertName": "",
|
||||
"dashboardAlerts": false,
|
||||
"groupBy": [],
|
||||
"groupMode": "default",
|
||||
"maxItems": 20,
|
||||
"sortOrder": 1,
|
||||
"stateFilter": {
|
||||
"firing": true,
|
||||
"pending": true,
|
||||
"noData": true,
|
||||
"normal": false,
|
||||
"error": true
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": 22,
|
||||
"type": "traces",
|
||||
"title": "Последние трассы API (Tempo)",
|
||||
"datasource": {
|
||||
"type": "tempo",
|
||||
"uid": "tempo"
|
||||
},
|
||||
"gridPos": {
|
||||
"x": 0,
|
||||
"y": 49,
|
||||
"w": 24,
|
||||
"h": 8
|
||||
},
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"datasource": {
|
||||
"type": "tempo",
|
||||
"uid": "tempo"
|
||||
},
|
||||
"queryType": "traceql",
|
||||
"query": "{ resource.service.name = \"messenger-api\" }",
|
||||
"limit": 20
|
||||
}
|
||||
]
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"list": []
|
||||
},
|
||||
"templating": {
|
||||
"list": []
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,11 @@
|
||||
apiVersion: 1
|
||||
providers:
|
||||
- name: messenger
|
||||
orgId: 1
|
||||
folder: Messenger
|
||||
type: file
|
||||
disableDeletion: false
|
||||
allowUiUpdates: false
|
||||
updateIntervalSeconds: 10
|
||||
options:
|
||||
path: /var/lib/grafana/dashboards
|
||||
@@ -0,0 +1,41 @@
|
||||
apiVersion: 1
|
||||
# Источники данных Grafana: Prometheus (метрики), Loki (логи), Tempo (трассы). Связи: лог -> трасса (derivedFields),
|
||||
# трасса -> логи (tracesToLogsV2), трасса -> метрики.
|
||||
datasources:
|
||||
- name: Prometheus
|
||||
uid: prometheus
|
||||
type: prometheus
|
||||
access: proxy
|
||||
url: http://prometheus:9090
|
||||
isDefault: true
|
||||
jsonData:
|
||||
timeInterval: 5s
|
||||
- name: Loki
|
||||
uid: loki
|
||||
type: loki
|
||||
access: proxy
|
||||
url: http://loki:3100
|
||||
jsonData:
|
||||
derivedFields:
|
||||
- name: trace_id
|
||||
matcherType: label
|
||||
matcherRegex: trace_id
|
||||
url: "$${__value.raw}"
|
||||
datasourceUid: tempo
|
||||
- name: Tempo
|
||||
uid: tempo
|
||||
type: tempo
|
||||
access: proxy
|
||||
url: http://tempo:3200
|
||||
jsonData:
|
||||
tracesToLogsV2:
|
||||
datasourceUid: loki
|
||||
spanStartTimeShift: "-1m"
|
||||
spanEndTimeShift: "1m"
|
||||
filterByTraceID: true
|
||||
customQuery: true
|
||||
query: '{service_name="$${__span.tags["service.name"]}"} | trace_id="$${__span.traceId}"'
|
||||
tracesToMetrics:
|
||||
datasourceUid: prometheus
|
||||
serviceMap:
|
||||
datasourceUid: prometheus
|
||||
@@ -0,0 +1,50 @@
|
||||
# Балансировщик lab3: HAProxy 3.0. Единственная публичная точка входа; API-реплики наружу не опубликованы.
|
||||
global
|
||||
log stdout format raw local0 notice
|
||||
maxconn 4096
|
||||
|
||||
defaults
|
||||
mode http
|
||||
log global
|
||||
option dontlognull
|
||||
option forwardfor
|
||||
# connect короткий: недоступная реплика быстро заменяется другой (redispatch)
|
||||
timeout connect 1s
|
||||
timeout client 30s
|
||||
timeout server 30s
|
||||
timeout http-request 10s
|
||||
timeout http-keep-alive 5s
|
||||
timeout queue 5s
|
||||
retries 3
|
||||
option redispatch 1
|
||||
retry-on conn-failure empty-response response-timeout 0rtt-rejected
|
||||
|
||||
# Docker DNS (127.0.0.11): адреса реплик API обновляются без перезапуска LB (scale up/down, пересоздание контейнеров).
|
||||
resolvers docker
|
||||
nameserver dns 127.0.0.11:53
|
||||
accepted_payload_size 8192
|
||||
resolve_retries 3
|
||||
timeout resolve 1s
|
||||
timeout retry 1s
|
||||
hold valid 2s
|
||||
hold nx 2s
|
||||
hold other 2s
|
||||
hold obsolete 2s
|
||||
|
||||
frontend public
|
||||
bind :8080
|
||||
default_backend api
|
||||
|
||||
# Внутренний healthcheck самого LB (контейнерный HEALTHCHECK); наружу не публикуется.
|
||||
frontend lb_health
|
||||
bind :8081
|
||||
monitor-uri /healthz
|
||||
|
||||
backend api
|
||||
balance roundrobin
|
||||
# Проверка готовности каждые 1 с; 2 неудачи подряд -> сервер исключён (обнаружение отказа <= ~3 с), 2 успеха -> возвращён.
|
||||
option httpchk
|
||||
http-check send meth GET uri /health/ready
|
||||
http-check expect status 200
|
||||
default-server check inter 1s fall 2 rise 2 resolvers docker init-addr none
|
||||
server-template api 1-8 api:8080
|
||||
@@ -0,0 +1,37 @@
|
||||
# Loki: хранилище логов (filesystem, volume lokidata). Принимает OTLP/HTTP от Collector (/otlp/v1/logs).
|
||||
auth_enabled: false
|
||||
server:
|
||||
http_listen_port: 3100
|
||||
grpc_listen_port: 9096
|
||||
log_level: warn
|
||||
common:
|
||||
instance_addr: 127.0.0.1
|
||||
path_prefix: /loki
|
||||
storage:
|
||||
filesystem:
|
||||
chunks_directory: /loki/chunks
|
||||
rules_directory: /loki/rules
|
||||
replication_factor: 1
|
||||
ring:
|
||||
kvstore:
|
||||
store: inmemory
|
||||
schema_config:
|
||||
configs:
|
||||
- from: 2024-04-01
|
||||
store: tsdb
|
||||
object_store: filesystem
|
||||
schema: v13
|
||||
index:
|
||||
prefix: index_
|
||||
period: 24h
|
||||
limits_config:
|
||||
allow_structured_metadata: true
|
||||
retention_period: 168h
|
||||
reject_old_samples: false
|
||||
volume_enabled: true
|
||||
compactor:
|
||||
working_directory: /loki/compactor
|
||||
retention_enabled: true
|
||||
delete_request_store: filesystem
|
||||
analytics:
|
||||
reporting_enabled: false
|
||||
@@ -0,0 +1,64 @@
|
||||
# Профиль «tail sampling» (трек 1, lab6): SDK отправляет 100 % спанов, Collector оставляет
|
||||
# - все трассы с ошибками, - все медленные (> 500 мс), - 10 % остальных.
|
||||
# Метрики и логи без изменений. Подключается через COLLECTOR_CONFIG=/etc/otelcol/collector-tail.yaml.
|
||||
receivers:
|
||||
otlp:
|
||||
protocols:
|
||||
grpc: {endpoint: 0.0.0.0:4317}
|
||||
http: {endpoint: 0.0.0.0:4318}
|
||||
|
||||
processors:
|
||||
memory_limiter: {check_interval: 1s, limit_mib: 280, spike_limit_mib: 60}
|
||||
batch: {send_batch_size: 1024, send_batch_max_size: 2048, timeout: 2s}
|
||||
attributes/scrub:
|
||||
actions:
|
||||
- {key: http.request.header.authorization, action: delete}
|
||||
- {key: http.request.header.cookie, action: delete}
|
||||
- {key: db.statement, action: delete}
|
||||
- {key: url.query, action: delete}
|
||||
tail_sampling:
|
||||
decision_wait: 5s
|
||||
num_traces: 20000
|
||||
expected_new_traces_per_sec: 500
|
||||
policies:
|
||||
- {name: errors, type: status_code, status_code: {status_codes: [ERROR]}}
|
||||
- {name: slow, type: latency, latency: {threshold_ms: 500}}
|
||||
- {name: baseline, type: probabilistic, probabilistic: {sampling_percentage: 10}}
|
||||
|
||||
exporters:
|
||||
otlp_grpc/tempo:
|
||||
endpoint: tempo:4317
|
||||
tls: {insecure: true}
|
||||
sending_queue: {enabled: true, num_consumers: 2, queue_size: 500}
|
||||
retry_on_failure: {enabled: true, initial_interval: 1s, max_interval: 10s, max_elapsed_time: 30s}
|
||||
otlp_http/loki:
|
||||
endpoint: http://loki:3100/otlp
|
||||
sending_queue: {enabled: true, num_consumers: 2, queue_size: 500}
|
||||
retry_on_failure: {enabled: true, initial_interval: 1s, max_interval: 10s, max_elapsed_time: 30s}
|
||||
prometheus:
|
||||
endpoint: 0.0.0.0:8889
|
||||
resource_to_telemetry_conversion: {enabled: true}
|
||||
metric_expiration: 20s
|
||||
|
||||
service:
|
||||
telemetry:
|
||||
metrics:
|
||||
level: normal
|
||||
readers:
|
||||
- pull:
|
||||
exporter:
|
||||
prometheus: {host: 0.0.0.0, port: 8888}
|
||||
logs: {level: warn}
|
||||
pipelines:
|
||||
traces:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, attributes/scrub, tail_sampling, batch]
|
||||
exporters: [otlp_grpc/tempo]
|
||||
metrics:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, batch]
|
||||
exporters: [prometheus]
|
||||
logs:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, attributes/scrub, batch]
|
||||
exporters: [otlp_http/loki]
|
||||
@@ -0,0 +1,75 @@
|
||||
# OpenTelemetry Collector (contrib). Один пайплайн на сигнал, у каждого — реальный backend:
|
||||
# traces -> Tempo (OTLP/gRPC)
|
||||
# metrics -> Prometheus (Prometheus-экспортёр: Prometheus сам забирает /metrics у Collector)
|
||||
# logs -> Loki (OTLP/HTTP, нативный приём Loki)
|
||||
# Collector — маршрутизатор/обработчик, не долговременное хранилище. Порты 4317/4318 доступны ТОЛЬКО во внутренней сети backend.
|
||||
receivers:
|
||||
otlp:
|
||||
protocols:
|
||||
grpc:
|
||||
endpoint: 0.0.0.0:4317
|
||||
http:
|
||||
endpoint: 0.0.0.0:4318
|
||||
|
||||
processors:
|
||||
# Защита RAM Collector: при приближении к лимиту новые данные отклоняются (клиент SDK получает ошибку и отбрасывает
|
||||
# пакет, приложение продолжает работать). limit_mib согласован с mem_limit контейнера.
|
||||
memory_limiter:
|
||||
check_interval: 1s
|
||||
limit_mib: 280
|
||||
spike_limit_mib: 60
|
||||
batch:
|
||||
send_batch_size: 1024
|
||||
send_batch_max_size: 2048
|
||||
timeout: 2s
|
||||
# Гарантированное удаление чувствительных атрибутов, даже если приложение их случайно передаст.
|
||||
attributes/scrub:
|
||||
actions:
|
||||
- {key: http.request.header.authorization, action: delete}
|
||||
- {key: http.request.header.cookie, action: delete}
|
||||
- {key: http.request.body, action: delete}
|
||||
- {key: db.statement, action: delete}
|
||||
- {key: url.query, action: delete}
|
||||
- {key: http.target, action: delete}
|
||||
# Tail-sampling (профиль «tail», см. collector-tail.yaml): здесь не используется.
|
||||
|
||||
exporters:
|
||||
otlp_grpc/tempo:
|
||||
endpoint: tempo:4317
|
||||
tls: {insecure: true}
|
||||
# Ограниченная буферизация при недоступном backend: очередь фиксированного размера, повторы до 30 с, затем отбрасывание
|
||||
# (потери считаются метрикой otelcol_exporter_send_failed_spans / otelcol_exporter_queue_size).
|
||||
sending_queue: {enabled: true, num_consumers: 2, queue_size: 500}
|
||||
retry_on_failure: {enabled: true, initial_interval: 1s, max_interval: 10s, max_elapsed_time: 30s}
|
||||
otlp_http/loki:
|
||||
endpoint: http://loki:3100/otlp
|
||||
sending_queue: {enabled: true, num_consumers: 2, queue_size: 500}
|
||||
retry_on_failure: {enabled: true, initial_interval: 1s, max_interval: 10s, max_elapsed_time: 30s}
|
||||
prometheus:
|
||||
endpoint: 0.0.0.0:8889
|
||||
resource_to_telemetry_conversion: {enabled: true}
|
||||
metric_expiration: 20s
|
||||
|
||||
service:
|
||||
telemetry:
|
||||
metrics:
|
||||
level: normal
|
||||
readers:
|
||||
- pull:
|
||||
exporter:
|
||||
prometheus: {host: 0.0.0.0, port: 8888}
|
||||
logs:
|
||||
level: warn
|
||||
pipelines:
|
||||
traces:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, attributes/scrub, batch]
|
||||
exporters: [otlp_grpc/tempo]
|
||||
metrics:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, batch]
|
||||
exporters: [prometheus]
|
||||
logs:
|
||||
receivers: [otlp]
|
||||
processors: [memory_limiter, attributes/scrub, batch]
|
||||
exporters: [otlp_http/loki]
|
||||
@@ -0,0 +1,20 @@
|
||||
# Prometheus: хранилище метрик (retention 7 дней, volume promdata). Забирает метрики приложений через Collector
|
||||
# (Prometheus-экспортёр :8889) и собственные метрики Collector (:8888): очереди экспортёров, отказы, drop из memory_limiter.
|
||||
global:
|
||||
scrape_interval: 5s
|
||||
evaluation_interval: 5s
|
||||
|
||||
rule_files:
|
||||
- /etc/prometheus/rules/*.yml
|
||||
|
||||
scrape_configs:
|
||||
- job_name: otel-collector-app-metrics
|
||||
honor_labels: true
|
||||
static_configs:
|
||||
- targets: ["otel-collector:8889"]
|
||||
- job_name: otel-collector-internal
|
||||
static_configs:
|
||||
- targets: ["otel-collector:8888"]
|
||||
- job_name: prometheus
|
||||
static_configs:
|
||||
- targets: ["localhost:9090"]
|
||||
@@ -0,0 +1,77 @@
|
||||
# Правила записи (SLI) и алерты. Runbook для каждого алерта — docs/runbooks.md (ссылка в аннотации).
|
||||
# SLI доступности API: доля бизнес-запросов (без /health) без ответа 5xx. 4xx — ошибка клиента, в SLI не входит.
|
||||
# SLO: 99.5 % за скользящее окно 30 дней => бюджет ошибок 0.5 %. Для демонстрации окна короткие (локальный стенд).
|
||||
groups:
|
||||
- name: messenger-sli
|
||||
interval: 5s
|
||||
rules:
|
||||
- record: sli:http_requests:rate1m
|
||||
expr: sum(rate(http_server_request_duration_seconds_count{http_route!~"/health.*"}[1m]))
|
||||
- record: sli:http_errors:rate1m
|
||||
expr: sum(rate(http_server_request_duration_seconds_count{http_route!~"/health.*",http_response_status_code=~"5.."}[1m]))
|
||||
- record: sli:http_error_ratio:1m
|
||||
expr: (sli:http_errors:rate1m / sli:http_requests:rate1m) or vector(0)
|
||||
- record: sli:http_error_ratio:5m
|
||||
expr: |
|
||||
(sum(rate(http_server_request_duration_seconds_count{http_route!~"/health.*",http_response_status_code=~"5.."}[5m]))
|
||||
/ sum(rate(http_server_request_duration_seconds_count{http_route!~"/health.*"}[5m]))) or vector(0)
|
||||
- record: sli:http_latency_p95:5m
|
||||
expr: histogram_quantile(0.95, sum by (le) (rate(http_server_request_duration_seconds_bucket{http_route!~"/health.*"}[5m])))
|
||||
- record: sli:http_latency_p95:1m
|
||||
expr: histogram_quantile(0.95, sum by (le) (rate(http_server_request_duration_seconds_bucket{http_route!~"/health.*"}[1m])))
|
||||
- record: sli:upload_error_ratio:1m
|
||||
expr: |
|
||||
(sum(rate(http_server_request_duration_seconds_count{http_route="/api/v1/chats/{chat_id}/attachments",http_response_status_code=~"5.."}[1m]))
|
||||
/ sum(rate(http_server_request_duration_seconds_count{http_route="/api/v1/chats/{chat_id}/attachments"}[1m]))) or vector(0)
|
||||
- record: sli:upload_requests:rate1m
|
||||
expr: sum(rate(http_server_request_duration_seconds_count{http_route="/api/v1/chats/{chat_id}/attachments"}[1m]))
|
||||
- record: sli:time_to_ready_p95:5m
|
||||
expr: histogram_quantile(0.95, sum by (le) (rate(messenger_job_time_to_ready_seconds_bucket[5m])))
|
||||
- name: messenger-alerts
|
||||
rules:
|
||||
- alert: HighHttpErrorRate
|
||||
expr: sli:http_error_ratio:1m > 0.05 and sli:http_requests:rate1m > 0.5
|
||||
for: 20s
|
||||
labels: {severity: page}
|
||||
annotations:
|
||||
summary: "Доля 5xx бизнес-запросов API выше 5 %"
|
||||
description: "Числитель — ответы 5xx, знаменатель — все запросы кроме /health. Значение: {{ $value | humanizePercentage }}."
|
||||
runbook: "docs/runbooks.md#higherrorrate"
|
||||
- alert: HighUploadErrorRate
|
||||
expr: sli:upload_error_ratio:1m > 0.2 and sli:upload_requests:rate1m > 0.2
|
||||
for: 15s
|
||||
labels: {severity: page}
|
||||
annotations:
|
||||
summary: "Более 20 % загрузок вложений завершаются 5xx"
|
||||
description: "Маршрут загрузки критичен, но малообъёмен: общий SLI 5xx его деградацию не заметит, поэтому есть отдельный алерт. Значение: {{ $value | humanizePercentage }}."
|
||||
runbook: "docs/runbooks.md#higherrorrate"
|
||||
- alert: HighLatencyP95
|
||||
expr: sli:http_latency_p95:1m > 0.5 and sli:http_requests:rate1m > 0.5
|
||||
for: 20s
|
||||
labels: {severity: ticket}
|
||||
annotations:
|
||||
summary: "p95 задержки бизнес-запросов выше 500 мс"
|
||||
description: "p95 из суммарной гистограммы всех реплик: {{ $value | humanizeDuration }}. Различайте медленную БД (spans db.connection, db_client_connections_waiting) и медленный обработчик."
|
||||
runbook: "docs/runbooks.md#highlatency"
|
||||
- alert: QueueLagGrowing
|
||||
expr: messenger_queue_oldest_job_age_seconds > 30
|
||||
for: 20s
|
||||
labels: {severity: ticket}
|
||||
annotations:
|
||||
summary: "Самая старая задача в очереди ждёт дольше 30 с"
|
||||
description: "Возраст старейшего задания: {{ $value }} с. Проверьте worker, брокер, S3."
|
||||
runbook: "docs/runbooks.md#queuelag"
|
||||
- alert: WorkerSilent
|
||||
expr: absent(messenger_worker_capacity)
|
||||
for: 30s
|
||||
labels: {severity: ticket}
|
||||
annotations:
|
||||
summary: "Worker не присылает метрики"
|
||||
runbook: "docs/runbooks.md#queuelag"
|
||||
- alert: TelemetryExportFailing
|
||||
expr: sum(rate(otelcol_exporter_send_failed_spans[1m])) + sum(rate(otelcol_exporter_send_failed_log_records[1m])) > 0
|
||||
for: 15s
|
||||
labels: {severity: ticket}
|
||||
annotations:
|
||||
summary: "Collector не может отправить телеметрию в хранилище"
|
||||
runbook: "docs/runbooks.md#telemetry"
|
||||
@@ -0,0 +1,33 @@
|
||||
# Tempo: хранилище трасс (локальный диск, volume tempodata, retention 7 суток). Принимает OTLP/gRPC от Collector.
|
||||
stream_over_http_enabled: true
|
||||
server:
|
||||
http_listen_port: 3200
|
||||
log_level: warn
|
||||
distributor:
|
||||
receivers:
|
||||
otlp:
|
||||
protocols:
|
||||
grpc:
|
||||
endpoint: 0.0.0.0:4317
|
||||
ingester:
|
||||
max_block_duration: 1m
|
||||
compactor:
|
||||
compaction:
|
||||
block_retention: 168h
|
||||
storage:
|
||||
trace:
|
||||
backend: local
|
||||
wal:
|
||||
path: /var/tempo/wal
|
||||
local:
|
||||
path: /var/tempo/blocks
|
||||
# Лимиты приёма: значения по умолчанию (10 000 «живых» трасс на тенанта) отбрасывали до 30 % спанов под нагрузкой
|
||||
# (метрика tempo_discarded_spans_total{reason="live_traces_exceeded"}) — найдено экспериментом «цена наблюдаемости».
|
||||
overrides:
|
||||
defaults:
|
||||
ingestion:
|
||||
max_traces_per_user: 150000
|
||||
rate_limit_bytes: 50000000
|
||||
burst_size_bytes: 100000000
|
||||
usage_report:
|
||||
reporting_enabled: false
|
||||
@@ -0,0 +1,52 @@
|
||||
-- 0001: пользователи, чаты, участники, сообщения, сессии.
|
||||
-- Источник истины для всех данных — PostgreSQL. Каждая миграция выполняется в транзакции.
|
||||
CREATE TABLE users (
|
||||
id uuid PRIMARY KEY,
|
||||
username text NOT NULL,
|
||||
display_name text NOT NULL,
|
||||
password_hash text NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
CONSTRAINT users_username_format CHECK (username ~ '^[a-z0-9_]{3,32}$'),
|
||||
CONSTRAINT users_display_name_len CHECK (char_length(display_name) BETWEEN 1 AND 100)
|
||||
);
|
||||
-- Уникальность username: гонка двух регистраций решается ограничением БД, а не проверкой в приложении.
|
||||
CREATE UNIQUE INDEX users_username_key ON users (username);
|
||||
|
||||
CREATE TABLE chats (
|
||||
id uuid PRIMARY KEY,
|
||||
title text NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
CONSTRAINT chats_title_len CHECK (char_length(title) BETWEEN 1 AND 100)
|
||||
);
|
||||
|
||||
-- chat_created_at денормализован ради keyset-пагинации списка чатов пользователя без join+sort.
|
||||
CREATE TABLE chat_members (
|
||||
chat_id uuid NOT NULL REFERENCES chats (id) ON DELETE CASCADE,
|
||||
user_id uuid NOT NULL REFERENCES users (id),
|
||||
chat_created_at timestamptz NOT NULL,
|
||||
ord smallint NOT NULL,
|
||||
PRIMARY KEY (chat_id, user_id)
|
||||
);
|
||||
CREATE INDEX chat_members_user_chats ON chat_members (user_id, chat_created_at, chat_id);
|
||||
|
||||
CREATE TABLE messages (
|
||||
id uuid PRIMARY KEY,
|
||||
chat_id uuid NOT NULL REFERENCES chats (id) ON DELETE CASCADE,
|
||||
sender_id uuid NOT NULL REFERENCES users (id),
|
||||
text text NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
CONSTRAINT messages_text_len CHECK (char_length(text) <= 4000)
|
||||
);
|
||||
-- История чата: WHERE chat_id = ? AND (created_at, id) > (?, ?) ORDER BY created_at, id LIMIT n
|
||||
-- обслуживается этим индексом без сортировки и без сканирования всей истории.
|
||||
CREATE INDEX messages_chat_history ON messages (chat_id, created_at, id);
|
||||
|
||||
-- Серверные сессии. В БД хранится только SHA-256 токена: утечка БД не раскрывает рабочие токены.
|
||||
CREATE TABLE sessions (
|
||||
token_hash bytea PRIMARY KEY,
|
||||
user_id uuid NOT NULL REFERENCES users (id),
|
||||
created_at timestamptz NOT NULL DEFAULT now(),
|
||||
expires_at timestamptz NOT NULL,
|
||||
revoked_at timestamptz
|
||||
);
|
||||
CREATE INDEX sessions_expires_at ON sessions (expires_at);
|
||||
@@ -0,0 +1,17 @@
|
||||
-- 0002: общая идемпотентность отправки сообщений (lab3).
|
||||
-- Область уникальности: пользователь + чат + ключ (метод и путь фиксированы: POST /chats/{id}/messages).
|
||||
-- Запись ключа и сообщения — одна транзакция: побочный эффект и результат фиксируются атомарно.
|
||||
CREATE TABLE idempotency_keys (
|
||||
user_id uuid NOT NULL REFERENCES users (id),
|
||||
chat_id uuid NOT NULL REFERENCES chats (id) ON DELETE CASCADE,
|
||||
idem_key text NOT NULL,
|
||||
request_hash bytea NOT NULL,
|
||||
message_id uuid NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT now(),
|
||||
PRIMARY KEY (user_id, chat_id, idem_key),
|
||||
CONSTRAINT idem_key_format CHECK (idem_key ~ '^[A-Za-z0-9._:-]{1,128}$'),
|
||||
-- сообщение вставляется после ключа в той же транзакции -> проверка внешнего ключа откладывается до COMMIT
|
||||
CONSTRAINT idem_message_fk FOREIGN KEY (message_id) REFERENCES messages (id) ON DELETE CASCADE
|
||||
DEFERRABLE INITIALLY DEFERRED
|
||||
);
|
||||
CREATE INDEX idempotency_keys_created_at ON idempotency_keys (created_at);
|
||||
@@ -0,0 +1,35 @@
|
||||
-- 0003: отправка сообщения одной серверной функцией (оптимизация по итогам трека 1: меньше обращений к БД на запрос).
|
||||
-- Семантика идентична версии «несколько запросов в транзакции» (migrations 0002 + код приложения):
|
||||
-- проверка членства -> просроченный ключ удаляется -> INSERT ключа (ON CONFLICT DO NOTHING) -> INSERT сообщения.
|
||||
-- Функция выполняется атомарно (один вызов = одна транзакция); гонку одинаковых ключей по-прежнему решает уникальный индекс.
|
||||
CREATE FUNCTION send_message(p_chat uuid, p_user uuid, p_key text, p_hash bytea, p_text text, p_ttl double precision)
|
||||
RETURNS TABLE (outcome text, msg_id uuid, msg_chat uuid, msg_sender uuid, msg_text text, msg_created timestamptz,
|
||||
request_hash bytea)
|
||||
LANGUAGE plpgsql AS $$
|
||||
DECLARE
|
||||
v_new uuid := gen_random_uuid();
|
||||
v_ins uuid;
|
||||
v_created timestamptz;
|
||||
BEGIN
|
||||
IF NOT EXISTS (SELECT 1 FROM chat_members m WHERE m.chat_id = p_chat AND m.user_id = p_user) THEN
|
||||
RETURN QUERY SELECT 'not_member'::text, NULL::uuid, NULL::uuid, NULL::uuid, NULL::text, NULL::timestamptz, NULL::bytea;
|
||||
RETURN;
|
||||
END IF;
|
||||
DELETE FROM idempotency_keys k
|
||||
WHERE k.user_id = p_user AND k.chat_id = p_chat AND k.idem_key = p_key
|
||||
AND k.created_at < now() - make_interval(secs => p_ttl);
|
||||
INSERT INTO idempotency_keys (user_id, chat_id, idem_key, request_hash, message_id)
|
||||
VALUES (p_user, p_chat, p_key, p_hash, v_new)
|
||||
ON CONFLICT DO NOTHING
|
||||
RETURNING message_id INTO v_ins;
|
||||
IF v_ins IS NOT NULL THEN
|
||||
INSERT INTO messages (id, chat_id, sender_id, text) VALUES (v_new, p_chat, p_user, p_text)
|
||||
RETURNING created_at INTO v_created;
|
||||
RETURN QUERY SELECT 'created'::text, v_new, p_chat, p_user, p_text, v_created, p_hash;
|
||||
ELSE
|
||||
RETURN QUERY SELECT 'replay'::text, m.id, m.chat_id, m.sender_id, m.text, m.created_at, k.request_hash
|
||||
FROM idempotency_keys k JOIN messages m ON m.id = k.message_id
|
||||
WHERE k.user_id = p_user AND k.chat_id = p_chat AND k.idem_key = p_key;
|
||||
END IF;
|
||||
END;
|
||||
$$;
|
||||
@@ -0,0 +1,90 @@
|
||||
-- 0004 (lab4): метаданные вложений в PostgreSQL, байты — в приватном S3-бакете.
|
||||
-- Ключ объекта назначает сервер (att/<id>/original); filename — только отображаемое имя.
|
||||
CREATE TABLE attachments (
|
||||
id uuid PRIMARY KEY,
|
||||
chat_id uuid NOT NULL REFERENCES chats (id) ON DELETE CASCADE,
|
||||
owner_id uuid NOT NULL REFERENCES users (id),
|
||||
filename text NOT NULL,
|
||||
content_type text NOT NULL,
|
||||
size_bytes bigint NOT NULL,
|
||||
sha256 text NOT NULL,
|
||||
width integer NOT NULL,
|
||||
height integer NOT NULL,
|
||||
status text NOT NULL DEFAULT 'ready',
|
||||
error_code text,
|
||||
s3_key text NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
updated_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
CONSTRAINT attachments_content_type CHECK (content_type IN ('image/png', 'image/jpeg')),
|
||||
CONSTRAINT attachments_size CHECK (size_bytes BETWEEN 1 AND 10485760),
|
||||
CONSTRAINT attachments_sha256 CHECK (sha256 ~ '^[a-f0-9]{64}$'),
|
||||
CONSTRAINT attachments_status CHECK (status IN ('queued', 'processing', 'ready', 'failed')),
|
||||
CONSTRAINT attachments_filename_len CHECK (char_length(filename) BETWEEN 1 AND 255)
|
||||
);
|
||||
CREATE UNIQUE INDEX attachments_s3_key_key ON attachments (s3_key);
|
||||
CREATE INDEX attachments_chat ON attachments (chat_id);
|
||||
|
||||
-- Связь сообщение <-> вложения (порядок сохраняется). Вложение может быть привязано к нескольким сообщениям чата.
|
||||
CREATE TABLE message_attachments (
|
||||
message_id uuid NOT NULL REFERENCES messages (id) ON DELETE CASCADE,
|
||||
attachment_id uuid NOT NULL REFERENCES attachments (id),
|
||||
position smallint NOT NULL,
|
||||
PRIMARY KEY (message_id, attachment_id)
|
||||
);
|
||||
CREATE INDEX message_attachments_attachment ON message_attachments (attachment_id);
|
||||
|
||||
-- Отправка сообщения с вложениями: заменяет функцию из 0003 (добавлен параметр p_atts и результат msg_atts).
|
||||
DROP FUNCTION send_message(uuid, uuid, text, bytea, text, double precision);
|
||||
CREATE FUNCTION send_message(p_chat uuid, p_user uuid, p_key text, p_hash bytea, p_text text, p_ttl double precision,
|
||||
p_atts uuid[])
|
||||
RETURNS TABLE (outcome text, msg_id uuid, msg_chat uuid, msg_sender uuid, msg_text text, msg_created timestamptz,
|
||||
request_hash bytea, msg_atts uuid[])
|
||||
LANGUAGE plpgsql AS $$
|
||||
DECLARE
|
||||
v_new uuid := gen_random_uuid();
|
||||
v_ins uuid;
|
||||
v_created timestamptz;
|
||||
v_found int;
|
||||
v_failed int;
|
||||
BEGIN
|
||||
IF NOT EXISTS (SELECT 1 FROM chat_members m WHERE m.chat_id = p_chat AND m.user_id = p_user) THEN
|
||||
RETURN QUERY SELECT 'not_member'::text, NULL::uuid, NULL::uuid, NULL::uuid, NULL::text, NULL::timestamptz,
|
||||
NULL::bytea, NULL::uuid[];
|
||||
RETURN;
|
||||
END IF;
|
||||
IF cardinality(p_atts) > 0 THEN
|
||||
SELECT count(*), count(*) FILTER (WHERE a.status = 'failed') INTO v_found, v_failed
|
||||
FROM attachments a WHERE a.id = ANY (p_atts) AND a.chat_id = p_chat;
|
||||
IF v_found <> cardinality(p_atts) THEN
|
||||
RETURN QUERY SELECT 'attachment_not_found'::text, NULL::uuid, NULL::uuid, NULL::uuid, NULL::text,
|
||||
NULL::timestamptz, NULL::bytea, NULL::uuid[];
|
||||
RETURN;
|
||||
END IF;
|
||||
IF v_failed > 0 THEN
|
||||
RETURN QUERY SELECT 'attachment_failed'::text, NULL::uuid, NULL::uuid, NULL::uuid, NULL::text,
|
||||
NULL::timestamptz, NULL::bytea, NULL::uuid[];
|
||||
RETURN;
|
||||
END IF;
|
||||
END IF;
|
||||
DELETE FROM idempotency_keys k
|
||||
WHERE k.user_id = p_user AND k.chat_id = p_chat AND k.idem_key = p_key
|
||||
AND k.created_at < now() - make_interval(secs => p_ttl);
|
||||
INSERT INTO idempotency_keys (user_id, chat_id, idem_key, request_hash, message_id)
|
||||
VALUES (p_user, p_chat, p_key, p_hash, v_new)
|
||||
ON CONFLICT DO NOTHING
|
||||
RETURNING message_id INTO v_ins;
|
||||
IF v_ins IS NOT NULL THEN
|
||||
INSERT INTO messages (id, chat_id, sender_id, text) VALUES (v_new, p_chat, p_user, p_text)
|
||||
RETURNING created_at INTO v_created;
|
||||
INSERT INTO message_attachments (message_id, attachment_id, position)
|
||||
SELECT v_new, a, (o - 1)::smallint FROM unnest(p_atts) WITH ORDINALITY AS t(a, o);
|
||||
RETURN QUERY SELECT 'created'::text, v_new, p_chat, p_user, p_text, v_created, p_hash, p_atts;
|
||||
ELSE
|
||||
RETURN QUERY SELECT 'replay'::text, m.id, m.chat_id, m.sender_id, m.text, m.created_at, k.request_hash,
|
||||
COALESCE((SELECT array_agg(ma.attachment_id ORDER BY ma.position)
|
||||
FROM message_attachments ma WHERE ma.message_id = m.id), '{}'::uuid[])
|
||||
FROM idempotency_keys k JOIN messages m ON m.id = k.message_id
|
||||
WHERE k.user_id = p_user AND k.chat_id = p_chat AND k.idem_key = p_key;
|
||||
END IF;
|
||||
END;
|
||||
$$;
|
||||
@@ -0,0 +1,60 @@
|
||||
-- 0005 (lab5): асинхронная обработка вложений — состояния, варианты, outbox, журнал событий.
|
||||
-- Расширение обратно совместимо с lab4: старые строки (ready) остаются валидными.
|
||||
ALTER TABLE attachments
|
||||
ADD COLUMN staging_key text,
|
||||
ADD COLUMN pipeline_version integer NOT NULL DEFAULT 1,
|
||||
ADD COLUMN attempt integer NOT NULL DEFAULT 0,
|
||||
ADD COLUMN job_id uuid,
|
||||
ADD COLUMN lease_owner text,
|
||||
ADD COLUMN lease_expires_at timestamptz,
|
||||
ADD COLUMN processing_started_at timestamptz,
|
||||
ADD COLUMN accepted_at timestamptz NOT NULL DEFAULT clock_timestamp(),
|
||||
ADD COLUMN ready_at timestamptz;
|
||||
|
||||
-- очередь/backlog и поиск зависших processing без полного скана
|
||||
CREATE INDEX attachments_active ON attachments (status, accepted_at) WHERE status IN ('queued', 'processing');
|
||||
CREATE INDEX attachments_lease ON attachments (lease_expires_at) WHERE status = 'processing';
|
||||
|
||||
-- Результаты обработки. Ключ эффекта (attachment_id, pipeline_version, name): повторная запись не создаёт второй набор вариантов.
|
||||
CREATE TABLE attachment_variants (
|
||||
attachment_id uuid NOT NULL REFERENCES attachments (id) ON DELETE CASCADE,
|
||||
pipeline_version integer NOT NULL,
|
||||
name text NOT NULL,
|
||||
content_type text NOT NULL,
|
||||
size_bytes bigint NOT NULL,
|
||||
width integer NOT NULL,
|
||||
height integer NOT NULL,
|
||||
sha256 text NOT NULL,
|
||||
s3_key text NOT NULL,
|
||||
created_at timestamptz NOT NULL DEFAULT now(),
|
||||
PRIMARY KEY (attachment_id, pipeline_version, name),
|
||||
CONSTRAINT variants_name CHECK (name IN ('thumbnail', 'crop', 'optimized')),
|
||||
CONSTRAINT variants_content_type CHECK (content_type IN ('image/jpeg', 'image/png', 'image/webp'))
|
||||
);
|
||||
|
||||
-- Transactional outbox: намерение «обработать файл» фиксируется В ТОЙ ЖЕ транзакции, что и строка вложения.
|
||||
-- Relay публикует строки в брокер (publisher confirm) и только затем ставит published_at. available_at — отложенная
|
||||
-- публикация (backoff между попытками). Доставка at-least-once; обработка идемпотентна.
|
||||
CREATE TABLE outbox (
|
||||
id bigserial PRIMARY KEY,
|
||||
job_id uuid NOT NULL,
|
||||
attachment_id uuid NOT NULL REFERENCES attachments (id) ON DELETE CASCADE,
|
||||
pipeline_version integer NOT NULL DEFAULT 1,
|
||||
attempt integer NOT NULL,
|
||||
available_at timestamptz NOT NULL DEFAULT now(),
|
||||
published_at timestamptz,
|
||||
created_at timestamptz NOT NULL DEFAULT now()
|
||||
);
|
||||
CREATE INDEX outbox_pending ON outbox (available_at) WHERE published_at IS NULL;
|
||||
|
||||
-- Журнал событий пайплайна (аудит операторского replay, причины отказов; без секретов).
|
||||
CREATE TABLE job_events (
|
||||
id bigserial PRIMARY KEY,
|
||||
attachment_id uuid NOT NULL,
|
||||
job_id uuid,
|
||||
event text NOT NULL,
|
||||
attempt integer,
|
||||
detail text,
|
||||
created_at timestamptz NOT NULL DEFAULT now()
|
||||
);
|
||||
CREATE INDEX job_events_attachment ON job_events (attachment_id, id);
|
||||
@@ -0,0 +1,3 @@
|
||||
-- 0006 (lab6): контекст трассировки в outbox. traceparent/tracestate запроса upload сохраняются вместе с заданием,
|
||||
-- relay передаёт их в брокер (поля jobs.schema.json), worker продолжает ту же трассу: API -> outbox -> broker -> worker.
|
||||
ALTER TABLE outbox ADD COLUMN traceparent text, ADD COLUMN tracestate text;
|
||||
@@ -0,0 +1,43 @@
|
||||
aio-pika==10.1.0
|
||||
aiobotocore==3.9.2
|
||||
aiohappyeyeballs==2.7.1
|
||||
aiohttp==3.14.3
|
||||
aioitertools==0.13.0
|
||||
aiormq==7.2.1
|
||||
aiosignal==1.4.0
|
||||
argon2-cffi-bindings==26.1.0
|
||||
argon2-cffi==23.1.0
|
||||
asyncpg==0.30.0
|
||||
attrs==26.1.0
|
||||
botocore==1.43.106
|
||||
cffi==2.1.1
|
||||
click==8.5.0
|
||||
frozenlist==1.8.0
|
||||
googleapis-common-protos==1.75.5
|
||||
h11==0.16.0
|
||||
httptools==0.8.0
|
||||
idna==3.20
|
||||
jmespath==1.1.0
|
||||
multidict==6.9.1
|
||||
opentelemetry-api==1.45.0
|
||||
opentelemetry-exporter-http-transport==0.66b0
|
||||
opentelemetry-exporter-otlp-common==0.66b0
|
||||
opentelemetry-exporter-otlp-proto-common==1.45.0
|
||||
opentelemetry-exporter-otlp-proto-http==1.45.0
|
||||
opentelemetry-proto==1.45.0
|
||||
opentelemetry-sdk==1.45.0
|
||||
opentelemetry-semantic-conventions==0.66b0
|
||||
pamqp==4.0.1
|
||||
pillow==12.3.0
|
||||
propcache==0.5.4
|
||||
protobuf==7.36.2
|
||||
pycparser==3.0
|
||||
python-dateutil==2.9.0.post0
|
||||
redis==5.2.1
|
||||
six==1.17.0
|
||||
typing_extensions==4.16.0
|
||||
urllib3==2.8.0
|
||||
uvicorn==0.34.0
|
||||
uvloop==0.21.0; sys_platform != "win32"
|
||||
wrapt==2.5.0
|
||||
yarl==1.25.1
|
||||
@@ -0,0 +1,113 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Уровень 4: последовательная остановка каждой API-реплики под нагрузкой (с восстановлением предыдущей).
|
||||
|
||||
Сценарий (стенд `docker compose up -d` с 2 репликами; нагрузка идёт через LB непрерывно ~70 с):
|
||||
t=10 остановить реплику A (docker stop --time 0 = мгновенный отказ процесса)
|
||||
t=25 восстановить (scripts/restore-replicas.sh)
|
||||
t=40 остановить реплику B
|
||||
t=55 восстановить
|
||||
Для каждого отказа считаем: RPS до/во время/после, число ошибочных попыток, время до прекращения ошибок
|
||||
(= обнаружение отказа балансировщиком), сохранность подтверждённых сообщений и отсутствие дублей.
|
||||
|
||||
python research/failover_sequence.py --out evidence
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
import loadgen # noqa: E402
|
||||
|
||||
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
|
||||
BASE = "http://127.0.0.1:8080"
|
||||
|
||||
|
||||
def sh(*args):
|
||||
return subprocess.run(args, cwd=ROOT, capture_output=True, text=True).stdout.strip()
|
||||
|
||||
|
||||
def api_ids():
|
||||
return sh("docker", "ps", "-q", "--filter", "label=com.docker.compose.project=mtusi-lab6",
|
||||
"--filter", "label=com.docker.compose.service=api").split()
|
||||
|
||||
|
||||
def scenario(rec, plan):
|
||||
"""Выполняется в отдельном потоке: действия по расписанию относительно rec.t0."""
|
||||
for at, action in plan:
|
||||
time.sleep(max(0, rec.t0 + at - time.time()))
|
||||
t = time.time() - rec.t0
|
||||
if action == "stop":
|
||||
target = api_ids()[0]
|
||||
subprocess.run(["docker", "stop", "--time", "0", target], capture_output=True)
|
||||
rec.marks.append({"t": round(t, 2), "action": "stop", "container": target})
|
||||
else:
|
||||
subprocess.run(["sh", "scripts/restore-replicas.sh"], cwd=ROOT, capture_output=True)
|
||||
rec.marks.append({"t": round(t, 2), "action": "restore"})
|
||||
|
||||
|
||||
async def main(args):
|
||||
stand = loadgen.Stand(BASE)
|
||||
await stand.setup()
|
||||
rec = loadgen.Recorder()
|
||||
rec.marks = []
|
||||
plan = [(10, "stop"), (25, "restore"), (40, "stop"), (55, "restore")]
|
||||
threading.Thread(target=scenario, args=(rec, plan), daemon=True).start()
|
||||
await loadgen.run_load(stand, rec, args.seconds, args.concurrency)
|
||||
await asyncio.sleep(2)
|
||||
check = await loadgen.verify(stand, rec)
|
||||
stops = [m for m in rec.marks if m["action"] == "stop"]
|
||||
restores = [m for m in rec.marks if m["action"] == "restore"]
|
||||
windows = []
|
||||
for i, stop in enumerate(stops):
|
||||
t0 = stop["t"]
|
||||
errs = [e for e in rec.events if e[1] == "POST" and e[2] != 201 and t0 - 0.2 <= e[0] <= t0 + 10]
|
||||
last_err = max((e[0] for e in errs), default=t0)
|
||||
restore_t = restores[i]["t"] if i < len(restores) else t0 + 15
|
||||
windows.append({"stopped_container": stop["container"], "stop_at_s": t0,
|
||||
"rps_before": loadgen.summarize(rec, t0 - 8, t0 - 0.5).get("ok_rps"),
|
||||
"rps_one_replica": loadgen.summarize(rec, t0 + 5, restore_t - 1).get("ok_rps"),
|
||||
"rps_after_restore": loadgen.summarize(rec, restore_t + 8, restore_t + 14).get("ok_rps"),
|
||||
"failed_attempts_within_10s": len(errs),
|
||||
"error_statuses": sorted({str(e[2]) for e in errs}),
|
||||
"seconds_until_errors_stop": round(last_err - t0, 2) if errs else 0.0,
|
||||
"p95_ms_one_replica": loadgen.summarize(rec, t0 + 5, restore_t - 1).get("p95_ms")})
|
||||
series = loadgen.per_second(rec)
|
||||
report = {"date": time.strftime("%Y-%m-%d %H:%M:%S"), "args": vars(args), "marks": rec.marks, "windows": windows,
|
||||
"total": loadgen.summarize(rec), "verification": check,
|
||||
"invariants_ok": check["confirmed_missing_in_history"] == 0 and check["duplicate_texts_in_history"] == 0,
|
||||
"timeline": [{"t": t, "ok": b["ok"], "err": b["err"], "p95_ms": loadgen.pct(b["lat"], .95), "instances": len(b["instances"])}
|
||||
for t, b in sorted(series.items())]}
|
||||
with open(os.path.join(args.out, "failover_sequence.json"), "w", encoding="utf-8") as fh:
|
||||
json.dump(report, fh, ensure_ascii=False, indent=2)
|
||||
plot(report, os.path.join(args.out, "failover_sequence.png"))
|
||||
print(json.dumps({k: v for k, v in report.items() if k != "timeline"}, ensure_ascii=False, indent=2))
|
||||
|
||||
|
||||
def plot(report, path):
|
||||
import matplotlib
|
||||
matplotlib.use("Agg")
|
||||
import matplotlib.pyplot as plt
|
||||
tl = report["timeline"]
|
||||
t = [x["t"] for x in tl]
|
||||
fig, ax = plt.subplots(3, 1, figsize=(11, 8.5), sharex=True)
|
||||
ax[0].plot(t, [x["ok"] for x in tl], color="#2e7d32"); ax[0].set_ylabel("успешных POST/с"); ax[0].set_title("Пропускная способность")
|
||||
ax[1].bar(t, [x["err"] for x in tl], color="#c62828"); ax[1].set_ylabel("ошибочных попыток/с"); ax[1].set_title("Ошибки (до повтора с тем же ключом)")
|
||||
ax[2].plot(t, [x["p95_ms"] or 0 for x in tl], color="#1565c0"); ax[2].set_ylabel("p95, мс"); ax[2].set_xlabel("время, с"); ax[2].set_title("Латентность успешных POST")
|
||||
for a in ax:
|
||||
for m in report["marks"]:
|
||||
a.axvline(m["t"], color="red" if m["action"] == "stop" else "green", ls="--", lw=1)
|
||||
ax[0].text(0.01, 0.05, "красная линия — остановка реплики, зелёная — восстановление", transform=ax[0].transAxes, fontsize=9)
|
||||
fig.tight_layout(); fig.savefig(path, dpi=110)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
ap.add_argument("--seconds", type=int, default=70)
|
||||
ap.add_argument("--concurrency", type=int, default=16)
|
||||
asyncio.run(main(ap.parse_args()))
|
||||
@@ -0,0 +1,258 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Уровень 4: контролируемые инциденты и проверка алертов firing -> resolved.
|
||||
|
||||
Фоном идёт нагрузка (сообщения + загрузки). Инциденты и ожидаемые алерты:
|
||||
1. worker_down — остановка worker: растёт возраст старейшей задачи -> QueueLagGrowing; после запуска — resolved.
|
||||
2. slow_db — триггер в БД добавляет pg_sleep(0.8) на вставку сообщения: растёт p95 -> HighLatencyP95;
|
||||
причина отличается от медленного обработчика по spans db.connection и метрике ожидания пула.
|
||||
3. s3_down — остановка S3: загрузки дают 503 -> HighUploadErrorRate (маршрут критичен, но малообъёмен).
|
||||
4. telemetry_backend — остановка Tempo и Loki: API продолжает отвечать, Collector ограничивает очередь/RAM -> TelemetryExportFailing.
|
||||
Для каждого фиксируем времена перехода алертов (по Prometheus API), ключевые метрики и «улики» из Tempo/Loki.
|
||||
python research/incidents.py --out evidence
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import base64
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
import urllib.parse
|
||||
import uuid
|
||||
|
||||
import aiohttp
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
|
||||
BASE = "http://127.0.0.1:8080"
|
||||
|
||||
|
||||
def sh(*args, stdin=None):
|
||||
return subprocess.run(args, cwd=ROOT, capture_output=True, text=True, encoding="utf-8", input=stdin).stdout.strip()
|
||||
|
||||
|
||||
def dc(*args):
|
||||
return sh("docker", "compose", "--env-file", ".env", *args)
|
||||
|
||||
|
||||
def prom_get(path):
|
||||
return json.loads(dc("exec", "-T", "prometheus", "wget", "-q", "-O", "-", "http://127.0.0.1:9090" + path) or "{}")
|
||||
|
||||
|
||||
def prom_value(query):
|
||||
r = prom_get("/api/v1/query?query=" + urllib.parse.quote(query))
|
||||
res = r.get("data", {}).get("result", [])
|
||||
return float(res[0]["value"][1]) if res else None
|
||||
|
||||
|
||||
def alerts():
|
||||
r = prom_get("/api/v1/alerts")
|
||||
return {a["labels"]["alertname"]: a["state"] for a in r.get("data", {}).get("alerts", [])}
|
||||
|
||||
|
||||
def net_get(url, accept=None):
|
||||
cmd = ["exec", "-T", "prometheus", "wget", "-q", "-O", "-"] + (["--header", "Accept: " + accept] if accept else []) + [url]
|
||||
return dc(*cmd)
|
||||
|
||||
|
||||
class Recorder(threading.Thread):
|
||||
def __init__(self):
|
||||
super().__init__(daemon=True)
|
||||
self.t0, self.rows, self.stop = time.time(), [], threading.Event()
|
||||
|
||||
def run(self):
|
||||
while not self.stop.is_set():
|
||||
t = time.time() - self.t0
|
||||
row = {"t": round(t, 1), "alerts": alerts(),
|
||||
"queue_oldest_age_s": prom_value("max(messenger_queue_oldest_job_age_seconds)"),
|
||||
"error_ratio_1m": prom_value("sli:http_error_ratio:1m"),
|
||||
"p95_s": prom_value("sli:http_latency_p95:1m"),
|
||||
"db_waiting": prom_value("sum(db_client_connections_waiting)"),
|
||||
"db_used": prom_value('sum(db_client_connections_usage{state="used"})'),
|
||||
"rps": prom_value("sli:http_requests:rate1m")}
|
||||
self.rows.append(row)
|
||||
self.stop.wait(3)
|
||||
|
||||
|
||||
async def register(s):
|
||||
name = "inc_%s" % uuid.uuid4().hex[:12]
|
||||
pw = "incident-test-pw-" + uuid.uuid4().hex[:6]
|
||||
async with s.post(BASE + "/api/v1/users", json={"username": name, "display_name": "i", "password": pw}) as r:
|
||||
user = await r.json()
|
||||
cred = base64.b64encode(("%s:%s" % (name, pw)).encode()).decode()
|
||||
async with s.post(BASE + "/api/v1/auth/sessions", headers={"Authorization": "Basic " + cred}) as r:
|
||||
token = (await r.json())["token"]
|
||||
async with s.post(BASE + "/api/v1/chats", json={"title": "inc", "member_ids": [user["id"]]}, headers={"Authorization": "Bearer " + token}) as r:
|
||||
chat = await r.json()
|
||||
return token, chat["id"]
|
||||
|
||||
|
||||
async def traffic(stop, png, counters):
|
||||
async with aiohttp.ClientSession(timeout=aiohttp.ClientTimeout(total=15), connector=aiohttp.TCPConnector(limit=40)) as s:
|
||||
token, chat = await register(s)
|
||||
hdr = {"Authorization": "Bearer " + token}
|
||||
|
||||
async def msgs():
|
||||
while not stop.is_set():
|
||||
try:
|
||||
async with s.post("%s/api/v1/chats/%s/messages" % (BASE, chat), json={"text": "inc-" + uuid.uuid4().hex[:8]},
|
||||
headers=dict(hdr, **{"Idempotency-Key": uuid.uuid4().hex})) as r:
|
||||
await r.read()
|
||||
counters["msg_%d" % (r.status // 100)] = counters.get("msg_%d" % (r.status // 100), 0) + 1
|
||||
except Exception:
|
||||
counters["msg_err"] = counters.get("msg_err", 0) + 1
|
||||
await asyncio.sleep(0.1)
|
||||
|
||||
async def uploads():
|
||||
while not stop.is_set():
|
||||
try:
|
||||
async with s.post("%s/api/v1/chats/%s/attachments?filename=i.png" % (BASE, chat), data=png,
|
||||
headers=dict(hdr, **{"Content-Type": "image/png"})) as r:
|
||||
await r.read()
|
||||
counters["up_%d" % (r.status // 100)] = counters.get("up_%d" % (r.status // 100), 0) + 1
|
||||
except Exception:
|
||||
counters["up_err"] = counters.get("up_err", 0) + 1
|
||||
await asyncio.sleep(0.3)
|
||||
await asyncio.gather(*([msgs() for _ in range(4)] + [uploads() for _ in range(10)]))
|
||||
|
||||
|
||||
def tempo_search(traceql, limit=20):
|
||||
now = int(time.time())
|
||||
url = "http://tempo:3200/api/search?q=%s&limit=%d&start=%d&end=%d" % (urllib.parse.quote(traceql), limit, now - 900, now + 5)
|
||||
raw = net_get(url)
|
||||
try:
|
||||
return json.loads(raw).get("traces", [])
|
||||
except ValueError:
|
||||
return []
|
||||
|
||||
|
||||
def loki_count(query):
|
||||
now = int(time.time() * 1e9)
|
||||
url = "http://loki:3100/loki/api/v1/query_range?limit=500&start=%d&end=%d&query=%s" % (now - 900 * 10 ** 9, now, urllib.parse.quote(query))
|
||||
try:
|
||||
data = json.loads(net_get(url))
|
||||
return sum(len(s["values"]) for s in data["data"]["result"])
|
||||
except ValueError:
|
||||
return None
|
||||
|
||||
|
||||
def wait_until(pred, seconds):
|
||||
deadline = time.time() + seconds
|
||||
while time.time() < deadline:
|
||||
if pred():
|
||||
return True
|
||||
time.sleep(2)
|
||||
return False
|
||||
|
||||
|
||||
def run_incident(name, expected_alert, start, stop, hold, evidence, rec):
|
||||
t_start = time.time() - rec.t0
|
||||
start()
|
||||
fired = wait_until(lambda: alerts().get(expected_alert) == "firing", hold)
|
||||
t_fired = time.time() - rec.t0 if fired else None
|
||||
ev = evidence() if evidence else {}
|
||||
ev["alerts_in_state_firing"] = sorted(k for k, v in alerts().items() if v == "firing")
|
||||
stop()
|
||||
t_recovery = time.time() - rec.t0
|
||||
resolved = wait_until(lambda: expected_alert not in alerts(), 120)
|
||||
t_resolved = time.time() - rec.t0 if resolved else None
|
||||
result = {"incident": name, "alert": expected_alert, "injected_at_s": round(t_start, 1),
|
||||
"firing_at_s": round(t_fired, 1) if t_fired else None, "detection_seconds": round(t_fired - t_start, 1) if t_fired else None,
|
||||
"recovered_cmd_at_s": round(t_recovery, 1), "resolved_at_s": round(t_resolved, 1) if t_resolved else None,
|
||||
"fired": fired, "resolved": resolved, "evidence": ev}
|
||||
if t_fired and t_resolved:
|
||||
result["time_to_resolve_after_recovery_s"] = round(t_resolved - t_recovery, 1)
|
||||
time.sleep(6)
|
||||
return result
|
||||
|
||||
|
||||
def orchestrate(rec, results, args):
|
||||
time.sleep(25) # прогрев: метрики и алерты стабилизируются
|
||||
# 1. worker down
|
||||
results.append(run_incident(
|
||||
"worker_down", "QueueLagGrowing", lambda: dc("stop", "worker"), lambda: dc("start", "worker"), 150,
|
||||
lambda: {"queue_oldest_age_s": prom_value("max(messenger_queue_oldest_job_age_seconds)"), "worker_capacity_series_present": prom_value("sum(messenger_worker_capacity)") is not None,
|
||||
"jobs_ready_rate_1m": prom_value('sum(rate(messenger_jobs_processed_total{outcome="ready"}[1m]))'),
|
||||
"diagnosis": "возраст старейшей задачи растёт при нулевой скорости обработки jobs (worker не отдаёт метрики) -> остановлен worker, а не медленная БД"},
|
||||
rec))
|
||||
# 2. slow DB
|
||||
trigger_sql = ("CREATE OR REPLACE FUNCTION slow_insert() RETURNS trigger LANGUAGE plpgsql AS $$ BEGIN PERFORM pg_sleep(0.8); RETURN NEW; END $$;"
|
||||
"CREATE TRIGGER slow_messages BEFORE INSERT ON messages FOR EACH ROW EXECUTE FUNCTION slow_insert();")
|
||||
pg = ["exec", "-T", "postgres", "psql", "-U", "messenger", "-d", "messenger", "-v", "ON_ERROR_STOP=1", "-c"]
|
||||
results.append(run_incident(
|
||||
"slow_db", "HighLatencyP95", lambda: dc(*pg, trigger_sql), lambda: dc(*pg, "DROP TRIGGER IF EXISTS slow_messages ON messages; DROP FUNCTION IF EXISTS slow_insert();"), 120,
|
||||
lambda: {"p95_s": prom_value("sli:http_latency_p95:1m"), "db_pool_waiting": prom_value("sum(db_client_connections_waiting)"),
|
||||
"slow_traces_with_long_db_span": len(tempo_search('{ resource.service.name = "messenger-api" && name = "db.connection" && duration > 300ms }')),
|
||||
"slow_http_post_messages_traces": len(tempo_search('{ name =~ "HTTP POST.*messages" && duration > 300ms }')),
|
||||
"diagnosis": "p95 высокий, а время в спанах db.connection сопоставимо с длительностью запроса и растёт ожидание пула -> медленная БД, а не обработчик"},
|
||||
rec))
|
||||
# 3. S3 down
|
||||
results.append(run_incident(
|
||||
"s3_down", "HighUploadErrorRate", lambda: dc("stop", "seaweedfs"), lambda: dc("start", "seaweedfs"), 120,
|
||||
lambda: {"upload_error_ratio_1m": prom_value("sli:upload_error_ratio:1m"), "global_error_ratio_1m": prom_value("sli:http_error_ratio:1m"), "error_traces": len(tempo_search('{ status = error && resource.service.name = "messenger-api" }')),
|
||||
"s3_put_failed_log_lines": loki_count('{service_name="messenger-api"} |= "s3_put_failed"'),
|
||||
"diagnosis": "рост доли 5xx на маршруте загрузки при нормальных остальных; спаны s3.put_object с ошибкой и логи s3_put_failed -> недоступен S3"},
|
||||
rec))
|
||||
# 4. telemetry backend down
|
||||
results.append(run_incident(
|
||||
"telemetry_backend", "TelemetryExportFailing", lambda: dc("stop", "tempo", "loki"), lambda: dc("start", "tempo", "loki"), 120,
|
||||
lambda: {"api_error_ratio_1m": prom_value("sli:http_error_ratio:1m"), "api_rps": prom_value("sli:http_requests:rate1m"),
|
||||
"collector_queue_size": prom_value("max(otelcol_exporter_queue_size)"), "collector_send_failed_spans_rate": prom_value("sum(rate(otelcol_exporter_send_failed_spans[1m]))"),
|
||||
"collector_rss_mib": round((prom_value("max(otelcol_process_memory_rss)") or 0) / 1048576, 1),
|
||||
"diagnosis": "ошибки экспорта растут, очередь ограничена; бизнес-метрики API без роста ошибок -> сбой хранилища телеметрии, не приложения"},
|
||||
rec))
|
||||
|
||||
|
||||
async def main(args):
|
||||
from PIL import Image
|
||||
import io
|
||||
buf = io.BytesIO()
|
||||
Image.new("RGB", (320, 200), (30, 90, 160)).save(buf, "PNG")
|
||||
stop_flag, counters, results = asyncio.Event(), {}, []
|
||||
rec = Recorder()
|
||||
rec.start()
|
||||
t = threading.Thread(target=orchestrate, args=(rec, results, args), daemon=True)
|
||||
t.start()
|
||||
traffic_task = asyncio.create_task(traffic(stop_flag, buf.getvalue(), counters))
|
||||
while t.is_alive():
|
||||
await asyncio.sleep(1)
|
||||
stop_flag.set()
|
||||
await asyncio.sleep(1)
|
||||
traffic_task.cancel()
|
||||
rec.stop.set()
|
||||
collector_rss = None
|
||||
report = {"date": time.strftime("%Y-%m-%d %H:%M:%S"), "traffic_counters": counters, "incidents": results, "timeline": rec.rows}
|
||||
with open(os.path.join(args.out, "incidents.json"), "w", encoding="utf-8") as fh:
|
||||
json.dump(report, fh, ensure_ascii=False, indent=2)
|
||||
plot(report, os.path.join(args.out, "incidents.png"))
|
||||
print(json.dumps({"incidents": results, "traffic": counters}, ensure_ascii=False, indent=2))
|
||||
|
||||
|
||||
def plot(report, path):
|
||||
import matplotlib
|
||||
matplotlib.use("Agg")
|
||||
import matplotlib.pyplot as plt
|
||||
rows = report["timeline"]
|
||||
t = [r["t"] for r in rows]
|
||||
fig, ax = plt.subplots(4, 1, figsize=(12, 11), sharex=True)
|
||||
ax[0].plot(t, [r["queue_oldest_age_s"] or 0 for r in rows], color="#6a1b9a"); ax[0].axhline(30, color="red", ls="--"); ax[0].set_ylabel("с"); ax[0].set_title("Возраст старейшей задачи (порог алерта 30 с)")
|
||||
ax[1].plot(t, [r["p95_s"] or 0 for r in rows], color="#1565c0"); ax[1].axhline(0.5, color="red", ls="--"); ax[1].set_ylabel("с"); ax[1].set_title("p95 задержки API (порог 0.5 с)")
|
||||
ax[2].plot(t, [(r["error_ratio_1m"] or 0) * 100 for r in rows], color="#c62828"); ax[2].axhline(5, color="red", ls="--"); ax[2].set_ylabel("%"); ax[2].set_title("Доля 5xx (порог 5 %)")
|
||||
names = ["QueueLagGrowing", "HighLatencyP95", "HighHttpErrorRate", "TelemetryExportFailing"]
|
||||
for i, n in enumerate(names):
|
||||
ax[3].fill_between(t, i, [i + 0.8 if r["alerts"].get(n) == "firing" else i for r in rows], step="post", alpha=0.7, label=n)
|
||||
ax[3].set_yticks([i + 0.4 for i in range(4)]); ax[3].set_yticklabels(names); ax[3].set_title("Алерты в состоянии firing"); ax[3].set_xlabel("время, с")
|
||||
for inc in report["incidents"]:
|
||||
for a in ax:
|
||||
a.axvline(inc["injected_at_s"], color="gray", ls=":")
|
||||
if inc["resolved_at_s"]:
|
||||
a.axvline(inc["resolved_at_s"], color="green", ls=":")
|
||||
fig.tight_layout(); fig.savefig(path, dpi=105)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
asyncio.run(main(ap.parse_args()))
|
||||
@@ -0,0 +1,172 @@
|
||||
"""Генератор нагрузки для экспериментов lab6+ (asyncio + aiohttp, closed-loop).
|
||||
|
||||
Каждый воркер выполняет цикл: POST сообщения с новым Idempotency-Key (90 %) или GET первой страницы истории (10 %).
|
||||
При ошибке соединения/5xx POST повторяется с ТЕМ ЖЕ ключом (до RETRIES раз) — именно так клиент безопасно
|
||||
повторяет неподтверждённую запись. Результат: по секундам успехи/ошибки/латентности, список подтверждённых
|
||||
сообщений (для сверки с историей), число повторов.
|
||||
"""
|
||||
import asyncio
|
||||
import base64
|
||||
import random
|
||||
import time
|
||||
import uuid
|
||||
|
||||
import aiohttp
|
||||
|
||||
RETRIES = 3
|
||||
|
||||
|
||||
class Stand:
|
||||
"""Набор пользователей/чатов, созданных через API (через LB), для одного эксперимента."""
|
||||
|
||||
def __init__(self, base, users=20, chats=20):
|
||||
self.base, self.n_users, self.n_chats = base, users, chats
|
||||
self.users, self.chats = [], []
|
||||
|
||||
async def setup(self):
|
||||
timeout = aiohttp.ClientTimeout(total=30)
|
||||
async with aiohttp.ClientSession(timeout=timeout) as s:
|
||||
for i in range(self.n_users):
|
||||
name = "ld_%s" % uuid.uuid4().hex[:14]
|
||||
pw = "load-test-password-" + uuid.uuid4().hex[:8]
|
||||
async with s.post(self.base + "/api/v1/users", json={"username": name, "display_name": "load %d" % i, "password": pw}) as r:
|
||||
user = await r.json()
|
||||
assert r.status == 201, user
|
||||
cred = base64.b64encode(("%s:%s" % (name, pw)).encode()).decode()
|
||||
async with s.post(self.base + "/api/v1/auth/sessions", headers={"Authorization": "Basic " + cred}) as r:
|
||||
sess = await r.json()
|
||||
assert r.status == 201, sess
|
||||
self.users.append({"id": user["id"], "token": sess["token"]})
|
||||
for i in range(self.n_chats):
|
||||
members = random.sample(self.users, 3)
|
||||
owner = members[0]
|
||||
async with s.post(self.base + "/api/v1/chats", json={"title": "load %d" % i, "member_ids": [m["id"] for m in members]},
|
||||
headers={"Authorization": "Bearer " + owner["token"]}) as r:
|
||||
chat = await r.json()
|
||||
assert r.status == 201, chat
|
||||
self.chats.append({"id": chat["id"], "members": members})
|
||||
|
||||
|
||||
class Recorder:
|
||||
def __init__(self):
|
||||
self.t0 = time.time()
|
||||
self.events = [] # (t, kind, status, latency_ms, instance)
|
||||
self.confirmed = [] # (chat_id, message_id, text)
|
||||
self.retries = 0
|
||||
self.final_failures = 0
|
||||
self.unconfirmed = [] # (chat_id, key, text) — финально не подтверждены (могли записаться!)
|
||||
|
||||
|
||||
async def worker(stand, rec, deadline_fn, read_ratio=0.1, text_size=0, think=0.0):
|
||||
connector = aiohttp.TCPConnector(limit=2)
|
||||
timeout = aiohttp.ClientTimeout(total=5)
|
||||
async with aiohttp.ClientSession(connector=connector, timeout=timeout) as s:
|
||||
n = 0
|
||||
while not deadline_fn():
|
||||
chat = random.choice(stand.chats)
|
||||
user = random.choice(chat["members"])
|
||||
hdr = {"Authorization": "Bearer " + user["token"]}
|
||||
if random.random() < read_ratio:
|
||||
t = time.perf_counter()
|
||||
try:
|
||||
async with s.get("%s/api/v1/chats/%s/messages?limit=20" % (stand.base, chat["id"]), headers=hdr) as r:
|
||||
await r.read()
|
||||
rec.events.append((time.time() - rec.t0, "GET", r.status, (time.perf_counter() - t) * 1000, r.headers.get("X-Instance-Id")))
|
||||
except Exception as exc:
|
||||
rec.events.append((time.time() - rec.t0, "GET", type(exc).__name__, (time.perf_counter() - t) * 1000, None))
|
||||
continue
|
||||
key = uuid.uuid4().hex
|
||||
text = "m-%s-%d" % (key[:8], n) + "я" * text_size
|
||||
n += 1
|
||||
for attempt in range(RETRIES + 1):
|
||||
t = time.perf_counter()
|
||||
try:
|
||||
async with s.post("%s/api/v1/chats/%s/messages" % (stand.base, chat["id"]), json={"text": text},
|
||||
headers=dict(hdr, **{"Idempotency-Key": key})) as r:
|
||||
body = await r.json(content_type=None)
|
||||
status, inst = r.status, r.headers.get("X-Instance-Id")
|
||||
except Exception as exc:
|
||||
status, inst, body = type(exc).__name__, None, None
|
||||
rec.events.append((time.time() - rec.t0, "POST", status, (time.perf_counter() - t) * 1000, inst))
|
||||
if status == 201:
|
||||
rec.confirmed.append((chat["id"], body["id"], text))
|
||||
break
|
||||
if attempt < RETRIES and (not isinstance(status, int) or status >= 500 or status == 429):
|
||||
rec.retries += 1
|
||||
await asyncio.sleep(0.05 * (attempt + 1))
|
||||
continue
|
||||
rec.final_failures += 1
|
||||
rec.unconfirmed.append((chat["id"], key, text))
|
||||
break
|
||||
if think:
|
||||
await asyncio.sleep(think)
|
||||
|
||||
|
||||
async def run_load(stand, rec, seconds, concurrency, **kw):
|
||||
end = time.time() + seconds
|
||||
await asyncio.gather(*(worker(stand, rec, lambda: time.time() > end, **kw) for _ in range(concurrency)))
|
||||
|
||||
|
||||
async def verify(stand, rec):
|
||||
"""Сверка: каждое подтверждённое сообщение читается, дублей по тексту нет, неподтверждённые встречаются не более 1 раза."""
|
||||
timeout = aiohttp.ClientTimeout(total=30)
|
||||
stored = {}
|
||||
async with aiohttp.ClientSession(timeout=timeout) as s:
|
||||
for chat in stand.chats:
|
||||
hdr = {"Authorization": "Bearer " + chat["members"][0]["token"]}
|
||||
cursor, items = None, []
|
||||
while True:
|
||||
url = "%s/api/v1/chats/%s/messages?limit=100%s" % (stand.base, chat["id"], "&cursor=" + cursor if cursor else "")
|
||||
async with s.get(url, headers=hdr) as r:
|
||||
page = await r.json()
|
||||
assert r.status == 200, page
|
||||
items += page["items"]
|
||||
cursor = page["next_cursor"]
|
||||
if not cursor:
|
||||
break
|
||||
stored[chat["id"]] = items
|
||||
ids = {m["id"] for items in stored.values() for m in items}
|
||||
texts = [m["text"] for items in stored.values() for m in items]
|
||||
confirmed_missing = [c for c in rec.confirmed if c[1] not in ids]
|
||||
dup_texts = len(texts) - len(set(texts))
|
||||
unconfirmed_present = sum(1 for c in rec.unconfirmed if c[2] in set(texts))
|
||||
return {"stored_messages": len(ids), "confirmed_201": len(rec.confirmed), "confirmed_missing_in_history": len(confirmed_missing),
|
||||
"duplicate_texts_in_history": dup_texts, "unconfirmed_requests": len(rec.unconfirmed),
|
||||
"unconfirmed_that_were_written": unconfirmed_present, "retries": rec.retries, "final_failures": rec.final_failures}
|
||||
|
||||
|
||||
def per_second(rec, kind="POST"):
|
||||
buckets = {}
|
||||
for t, k, status, lat, inst in rec.events:
|
||||
if k != kind:
|
||||
continue
|
||||
b = buckets.setdefault(int(t), {"ok": 0, "err": 0, "lat": [], "instances": set()})
|
||||
if status == 201 or (kind == "GET" and status == 200):
|
||||
b["ok"] += 1
|
||||
b["lat"].append(lat)
|
||||
if inst:
|
||||
b["instances"].add(inst)
|
||||
else:
|
||||
b["err"] += 1
|
||||
return buckets
|
||||
|
||||
|
||||
def pct(values, p):
|
||||
if not values:
|
||||
return None
|
||||
values = sorted(values)
|
||||
return round(values[max(0, int(len(values) * p + 0.9999) - 1)], 2)
|
||||
|
||||
|
||||
def summarize(rec, t_from=None, t_to=None, kind="POST"):
|
||||
rows = [e for e in rec.events if e[1] == kind and (t_from is None or e[0] >= t_from) and (t_to is None or e[0] < t_to)]
|
||||
if not rows:
|
||||
return {}
|
||||
dur = (t_to if t_to is not None else max(e[0] for e in rows)) - (t_from if t_from is not None else min(e[0] for e in rows))
|
||||
ok = [e[3] for e in rows if e[2] == 201 or (kind == "GET" and e[2] == 200)]
|
||||
errs = {}
|
||||
for e in rows:
|
||||
if not (e[2] == 201 or (kind == "GET" and e[2] == 200)):
|
||||
errs[str(e[2])] = errs.get(str(e[2]), 0) + 1
|
||||
return {"requests": len(rows), "ok": len(ok), "errors": errs, "seconds": round(dur, 2), "ok_rps": round(len(ok) / dur, 1) if dur > 0 else None,
|
||||
"error_ratio": round((len(rows) - len(ok)) / len(rows), 4), "p50_ms": pct(ok, .5), "p95_ms": pct(ok, .95), "p99_ms": pct(ok, .99)}
|
||||
@@ -0,0 +1,217 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Трек 1 (цена наблюдаемости), lab6.
|
||||
|
||||
Профили при ОДИНАКОВОЙ нагрузке (8 процессов-клиентов, 64 одновременных, 90 % POST сообщений / 10 % GET, 30 с после прогрева):
|
||||
off — OTEL_ENABLED=0 (SDK не создаётся, экспорта нет)
|
||||
on100 — телеметрия включена, head-sampling 100 % трасс
|
||||
head10 — head-sampling 10 % (ParentBased + TraceIdRatio в SDK)
|
||||
tail — SDK отправляет 100 %, Collector оставляет errors + медленные (>500 мс) + 10 % остальных (tail_sampling)
|
||||
Во время замера внедряются отказы: 2 % запросов дают 500 и 2 % задерживаются на 0.6 с (одинаково для всех профилей) — это «редкие
|
||||
ошибки и медленные запросы», судьбу которых в Tempo мы сравниваем.
|
||||
Метрики: RPS, p50/p95/p99, CPU/RSS API и Collector (docker stats), объём телеметрии (принятые Collector спаны/логи/точки метрик),
|
||||
потери (отказы приёма Collector), размер хранилищ Tempo/Loki, доля сохранённых ошибочных и медленных трасс.
|
||||
python research/obs_cost.py --out evidence
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import multiprocessing as mp
|
||||
import os
|
||||
import statistics
|
||||
import subprocess
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
import urllib.parse
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
import loadgen # noqa: E402
|
||||
|
||||
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
|
||||
BASE = "http://127.0.0.1:8080"
|
||||
PROFILES = {
|
||||
"off": {"OTEL_ENABLED": "0", "OTEL_TRACES_SAMPLER_ARG": "1.0", "COLLECTOR_CONFIG": "/etc/otelcol/collector.yaml"},
|
||||
"on100": {"OTEL_ENABLED": "1", "OTEL_TRACES_SAMPLER_ARG": "1.0", "COLLECTOR_CONFIG": "/etc/otelcol/collector.yaml"},
|
||||
"head10": {"OTEL_ENABLED": "1", "OTEL_TRACES_SAMPLER_ARG": "0.1", "COLLECTOR_CONFIG": "/etc/otelcol/collector.yaml"},
|
||||
"tail": {"OTEL_ENABLED": "1", "OTEL_TRACES_SAMPLER_ARG": "1.0", "COLLECTOR_CONFIG": "/etc/otelcol/collector-tail.yaml"},
|
||||
# то же, что on100, но большая очередь/пакет экспорта SDK: проверка гипотезы «часть спанов теряется очередью SDK»
|
||||
"on100_bigq": {"OTEL_ENABLED": "1", "OTEL_TRACES_SAMPLER_ARG": "1.0", "COLLECTOR_CONFIG": "/etc/otelcol/collector.yaml",
|
||||
"OTEL_QUEUE_SIZE": "65536", "OTEL_BATCH_SIZE": "2048"},
|
||||
}
|
||||
|
||||
|
||||
def sh(*args, env=None):
|
||||
return subprocess.run(args, cwd=ROOT, capture_output=True, text=True, encoding="utf-8", env=dict(os.environ, **(env or {}))).stdout.strip()
|
||||
|
||||
|
||||
def dc(*args, env=None):
|
||||
return sh("docker", "compose", "--env-file", ".env", *args, env=env)
|
||||
|
||||
|
||||
def net_get(url):
|
||||
return dc("exec", "-T", "prometheus", "wget", "-q", "-O", "-", url)
|
||||
|
||||
|
||||
def prom_value(query):
|
||||
try:
|
||||
r = json.loads(net_get("http://127.0.0.1:9090/api/v1/query?query=" + urllib.parse.quote(query)))["data"]["result"]
|
||||
return float(r[0]["value"][1]) if r else None
|
||||
except (ValueError, KeyError):
|
||||
return None
|
||||
|
||||
|
||||
def apply_profile(profile, faults=True):
|
||||
env = dict(PROFILES[profile], API_FAULT_INJECTION="http500:0.02,slow:0.02:0.6" if faults else "")
|
||||
dc("up", "-d", "--no-deps", "--force-recreate", "otel-collector", env=env)
|
||||
dc("up", "-d", "--no-deps", "--force-recreate", "api", "worker", env=env)
|
||||
deadline = time.time() + 90
|
||||
while time.time() < deadline:
|
||||
ids = dc("ps", "-q", "api").split()
|
||||
if len(ids) == 2 and all(sh("docker", "inspect", "-f", "{{.State.Health.Status}}", i) == "healthy" for i in ids):
|
||||
break
|
||||
time.sleep(1)
|
||||
time.sleep(8)
|
||||
|
||||
|
||||
def child(args):
|
||||
stand_data, seconds, concurrency, seed, t0 = args
|
||||
import random
|
||||
random.seed(seed)
|
||||
stand = loadgen.Stand(BASE)
|
||||
stand.users, stand.chats = stand_data["users"], stand_data["chats"]
|
||||
rec = loadgen.Recorder()
|
||||
rec.t0 = t0
|
||||
|
||||
async def go():
|
||||
await loadgen.run_load(stand, rec, seconds, concurrency)
|
||||
asyncio.run(go())
|
||||
return rec.events
|
||||
|
||||
|
||||
class Sampler(threading.Thread):
|
||||
def __init__(self):
|
||||
super().__init__(daemon=True)
|
||||
self.rows, self.stop = [], threading.Event()
|
||||
|
||||
def run(self):
|
||||
while not self.stop.is_set():
|
||||
out = subprocess.run(["docker", "stats", "--no-stream", "--format", "{{.Name}} {{.CPUPerc}} {{.MemUsage}}"], capture_output=True, text=True).stdout.splitlines()
|
||||
self.rows.append((time.time(), [l.split() for l in out if "mtusi-lab6" in l]))
|
||||
|
||||
def summary(self, t0, t1):
|
||||
agg = {}
|
||||
for ts, rows in self.rows:
|
||||
if not t0 <= ts <= t1:
|
||||
continue
|
||||
for r in rows:
|
||||
svc = r[0].replace("mtusi-lab6-", "").rsplit("-", 1)[0]
|
||||
mem = float(r[2].replace("MiB", "").replace("GiB", "")) * (1024 if "GiB" in r[2] else 1)
|
||||
agg.setdefault(svc, {"cpu": [], "mem": []})
|
||||
agg[svc]["cpu"].append(float(r[1].rstrip("%")))
|
||||
agg[svc]["mem"].append(mem)
|
||||
return {k: {"avg_cpu_percent": round(statistics.mean(v["cpu"]), 1), "peak_mem_mib": round(max(v["mem"]), 1)} for k, v in agg.items()}
|
||||
|
||||
|
||||
def volume_bytes(volume):
|
||||
out = sh("docker", "run", "--rm", "-v", "%s:/v:ro" % volume, "alpine:3.20", "du", "-sk", "/v")
|
||||
try:
|
||||
return int(out.split()[0]) * 1024
|
||||
except (ValueError, IndexError):
|
||||
return None
|
||||
|
||||
|
||||
def counter(name):
|
||||
return prom_value("sum(%s)" % name) or 0.0
|
||||
|
||||
|
||||
def tempo_count(traceql, since, limit=5000):
|
||||
url = "http://tempo:3200/api/search?q=%s&limit=%d&start=%d&end=%d" % (urllib.parse.quote(traceql), limit, int(since), int(time.time()) + 5)
|
||||
try:
|
||||
return len(json.loads(net_get(url)).get("traces", []))
|
||||
except ValueError:
|
||||
return None
|
||||
|
||||
|
||||
def run_profile(profile, args, stand_data):
|
||||
apply_profile(profile)
|
||||
# прогрев
|
||||
t0 = time.time()
|
||||
with mp.get_context("spawn").Pool(args.procs) as pool:
|
||||
pool.map(child, [(stand_data, 5, args.concurrency // args.procs, i, t0) for i in range(args.procs)])
|
||||
time.sleep(3)
|
||||
before = {k: counter(k) for k in ("otelcol_receiver_accepted_spans", "otelcol_receiver_accepted_log_records",
|
||||
"otelcol_receiver_accepted_metric_points", "otelcol_receiver_refused_spans",
|
||||
"otelcol_exporter_sent_spans", "otelcol_exporter_send_failed_spans")}
|
||||
vol_before = {"tempo": volume_bytes("mtusi-lab6_tempodata"), "loki": volume_bytes("mtusi-lab6_lokidata")}
|
||||
sampler = Sampler()
|
||||
sampler.start()
|
||||
t_start = time.time()
|
||||
with mp.get_context("spawn").Pool(args.procs) as pool:
|
||||
results = pool.map(child, [(stand_data, args.seconds, args.concurrency // args.procs, 100 + i, t_start) for i in range(args.procs)])
|
||||
t_end = time.time()
|
||||
sampler.stop.set()
|
||||
rec = loadgen.Recorder()
|
||||
rec.t0 = t_start
|
||||
for events in results:
|
||||
rec.events.extend(events)
|
||||
post = loadgen.summarize(rec, 0, args.seconds, "POST")
|
||||
time.sleep(args.flush) # дать SDK/Collector/хранилищам дописать данные
|
||||
after = {k: counter(k) for k in before}
|
||||
vol_after = {"tempo": volume_bytes("mtusi-lab6_tempodata"), "loki": volume_bytes("mtusi-lab6_lokidata")}
|
||||
# «внедрённые» отказы считаем на стороне клиента (счётчики Prometheus нестабильны при пересоздании реплик):
|
||||
# ответы 500 и ответы медленнее 500 мс (внедряется задержка 0.6 с; фоновые запросы быстрее)
|
||||
posts = [e for e in rec.events if e[1] == "POST"]
|
||||
injected_err = sum(1 for e in posts if e[2] == 500)
|
||||
injected_slow = sum(1 for e in posts if e[2] == 201 and e[3] > 500)
|
||||
kept_err = tempo_count('{ status = error && resource.service.name = "messenger-api" && name =~ "HTTP.*" }', t_start - 5)
|
||||
kept_slow = tempo_count('{ resource.service.name = "messenger-api" && name =~ "HTTP.*" && duration > 500ms }', t_start - 5)
|
||||
kept_all = tempo_count('{ resource.service.name = "messenger-api" && name =~ "HTTP.*" }', t_start - 5)
|
||||
out = {"profile": profile, "post": post, "docker_stats": sampler.summary(t_start + 2, t_end),
|
||||
"collector_delta": {k.replace("otelcol_", ""): round(after[k] - before[k]) for k in before},
|
||||
"tempo_bytes_delta": (vol_after["tempo"] - vol_before["tempo"]) if vol_after["tempo"] and vol_before["tempo"] else None,
|
||||
"loki_bytes_delta": (vol_after["loki"] - vol_before["loki"]) if vol_after["loki"] and vol_before["loki"] else None,
|
||||
"errors_injected_500": injected_err, "slow_requests_gt500ms": injected_slow,
|
||||
"traces_in_tempo": {"all_http": kept_all, "error": kept_err, "slow_gt500ms": kept_slow},
|
||||
"kept_ratio": {"errors": round(kept_err / injected_err, 3) if injected_err and kept_err is not None else None,
|
||||
"slow": round(kept_slow / injected_slow, 3) if injected_slow and kept_slow is not None else None}}
|
||||
print(json.dumps(out, ensure_ascii=False))
|
||||
return out
|
||||
|
||||
|
||||
def main():
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
ap.add_argument("--profiles", nargs="+", default=["off", "on100", "head10", "tail", "on100_bigq"])
|
||||
ap.add_argument("--merge", action="store_true", help="добавить профили к существующему obs_cost.json")
|
||||
ap.add_argument("--seconds", type=int, default=30)
|
||||
ap.add_argument("--concurrency", type=int, default=64)
|
||||
ap.add_argument("--procs", type=int, default=8)
|
||||
ap.add_argument("--flush", type=int, default=25)
|
||||
ap.add_argument("--repeat", type=int, default=2)
|
||||
args = ap.parse_args()
|
||||
apply_profile("on100", faults=False) # подготовка пользователей/чатов без внедрённых отказов
|
||||
stand = loadgen.Stand(BASE, users=40, chats=40)
|
||||
asyncio.run(stand.setup())
|
||||
stand_data = {"users": stand.users, "chats": stand.chats}
|
||||
results = []
|
||||
for profile in args.profiles:
|
||||
runs = [run_profile(profile, args, stand_data) for _ in range(args.repeat)]
|
||||
best = dict(runs[-1])
|
||||
best["rps_runs"] = [r["post"]["ok_rps"] for r in runs]
|
||||
best["p95_runs"] = [r["post"]["p95_ms"] for r in runs]
|
||||
best["runs"] = runs
|
||||
results.append(best)
|
||||
path = os.path.join(args.out, "obs_cost.json")
|
||||
if args.merge and os.path.exists(path):
|
||||
old = json.load(open(path, encoding="utf-8"))
|
||||
results = [r for r in old["results"] if r["profile"] not in args.profiles] + results
|
||||
report = {"date": time.strftime("%Y-%m-%d %H:%M:%S"), "args": vars(args), "faults": "http500:0.02,slow:0.02:0.6", "results": results}
|
||||
with open(path, "w", encoding="utf-8") as fh:
|
||||
json.dump(report, fh, ensure_ascii=False, indent=2)
|
||||
apply_profile("on100", faults=False)
|
||||
print("saved")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
mp.freeze_support()
|
||||
main()
|
||||
@@ -0,0 +1,36 @@
|
||||
#!/usr/bin/env python3
|
||||
"""График трека 1 из evidence/obs_cost.json: python research/plot_obs_cost.py --out evidence"""
|
||||
import argparse
|
||||
import json
|
||||
import os
|
||||
|
||||
import matplotlib
|
||||
|
||||
matplotlib.use("Agg")
|
||||
import matplotlib.pyplot as plt # noqa: E402
|
||||
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
out = ap.parse_args().out
|
||||
d = json.load(open(os.path.join(out, "obs_cost.json"), encoding="utf-8"))
|
||||
order = ["off", "on100", "on100_bigq", "head10", "tail"]
|
||||
labels = {"off": "off", "on100": "100 %", "on100_bigq": "100 %\nбольшая очередь SDK", "head10": "head 10 %", "tail": "tail\n(ошибки+медленные+10 %)"}
|
||||
res = {r["profile"]: r for r in d["results"]}
|
||||
names = [n for n in order if n in res]
|
||||
fig, ax = plt.subplots(1, 4, figsize=(19, 4.8))
|
||||
ax[0].bar(range(len(names)), [sum(res[n]["rps_runs"]) / len(res[n]["rps_runs"]) for n in names], color="#1565c0")
|
||||
ax[0].set_title("Успешных запросов/с (среднее 2 прогонов)")
|
||||
ax[1].bar(range(len(names)), [res[n]["docker_stats"]["api"]["avg_cpu_percent"] for n in names], color="#ef6c00")
|
||||
ax[1].set_title("CPU API, % одного ядра (на реплику)")
|
||||
ax[2].bar([i - 0.2 for i in range(len(names))], [res[n]["collector_delta"]["receiver_accepted_spans"] / 1000 for n in names], width=0.4, label="спаны: SDK → Collector", color="#6a1b9a")
|
||||
ax[2].bar([i + 0.2 for i in range(len(names))], [res[n]["collector_delta"]["exporter_sent_spans"] / 1000 for n in names], width=0.4, label="спаны: Collector → Tempo", color="#2e7d32")
|
||||
ax[2].set_title("Спанов за окно 30 с + 25 с, тыс."); ax[2].legend()
|
||||
ax[3].bar([i - 0.2 for i in range(len(names))], [min(1.0, (res[n]["kept_ratio"]["errors"] or 0)) * 100 for n in names], width=0.4, label="ошибки (500)", color="#c62828")
|
||||
ax[3].bar([i + 0.2 for i in range(len(names))], [min(1.0, (res[n]["kept_ratio"]["slow"] or 0)) * 100 for n in names], width=0.4, label="медленные (>500 мс)", color="#00838f")
|
||||
ax[3].set_title("Доля сохранённых в Tempo трасс, %"); ax[3].legend()
|
||||
for a in ax:
|
||||
a.set_xticks(range(len(names)))
|
||||
a.set_xticklabels([labels[n] for n in names], fontsize=8)
|
||||
fig.tight_layout()
|
||||
fig.savefig(os.path.join(out, "obs_cost.png"), dpi=110)
|
||||
print("saved")
|
||||
@@ -0,0 +1,271 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Трек 1 (очередь под перегрузкой), lab6.
|
||||
|
||||
Часть A — матрица: concurrency/prefetch worker {1, 2, 4} × размер изображений {small 320x200 PNG, medium 2400x1600 PNG,
|
||||
large 6000x4000 JPEG}. Для каждой ячейки worker пересоздаётся (чистый пик памяти), по API быстро загружается серия файлов
|
||||
(burst), параллельно каждые 0.5 с снимаются глубина очереди и возраст старейшей задачи. Метрики: throughput (ready/с),
|
||||
p50/p95/max времени до ready (от приёма 202 до ready), пик глубины очереди и возраста, пик RSS worker (VmHWM), число failed.
|
||||
Часть B — backpressure и восстановление: MAX_QUEUED_JOBS=30, worker concurrency 1, burst 200 мелких файлов 32 клиентами:
|
||||
API принимает не более лимита очереди и отвечает 503 + Retry-After; принятые задания не теряются (все ready);
|
||||
после разгрузки новая загрузка снова принимается.
|
||||
python research/queue_overload.py --out evidence
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import base64
|
||||
import io
|
||||
import json
|
||||
import os
|
||||
import statistics
|
||||
import subprocess
|
||||
import threading
|
||||
import time
|
||||
import uuid
|
||||
|
||||
import aiohttp
|
||||
from PIL import Image
|
||||
|
||||
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
|
||||
BASE = "http://127.0.0.1:8080"
|
||||
PROJECT = "mtusi-lab6"
|
||||
|
||||
|
||||
def sh(*args, env=None):
|
||||
return subprocess.run(args, cwd=ROOT, capture_output=True, text=True, encoding="utf-8", env=dict(os.environ, **(env or {}))).stdout.strip()
|
||||
|
||||
|
||||
def psql(sql):
|
||||
return sh("docker", "compose", "--env-file", ".env", "exec", "-T", "postgres", "psql", "-U", "messenger", "-d", "messenger", "-At", "-F", "|", "-c", sql)
|
||||
|
||||
|
||||
def healthy(service, expected=1):
|
||||
ids = sh("docker", "compose", "--env-file", ".env", "ps", "-q", service).split()
|
||||
return len(ids) >= expected and all(sh("docker", "inspect", "-f", "{{.State.Health.Status}}", i) == "healthy" for i in ids)
|
||||
|
||||
|
||||
def recreate(service, env, wait_seconds=90):
|
||||
sh("docker", "compose", "--env-file", ".env", "up", "-d", "--no-deps", "--force-recreate", service, env=env)
|
||||
deadline = time.time() + wait_seconds
|
||||
while time.time() < deadline:
|
||||
if healthy(service, 1 if service == "worker" else 2):
|
||||
break
|
||||
time.sleep(1)
|
||||
time.sleep(6 if service == "api" else 3)
|
||||
|
||||
|
||||
def make_images(root):
|
||||
import random
|
||||
random.seed(11)
|
||||
out = {}
|
||||
out["small"] = open(os.path.join(root, "student-tests", "fixtures", "sample.png"), "rb").read()
|
||||
|
||||
def noisy(w, h):
|
||||
base = Image.effect_noise((w // 8, h // 8), 60).convert("RGB").resize((w, h), Image.BILINEAR)
|
||||
grad = Image.linear_gradient("L").resize((w, h)).convert("RGB")
|
||||
return Image.blend(base, grad, 0.5)
|
||||
buf = io.BytesIO()
|
||||
noisy(2400, 1600).save(buf, "PNG", compress_level=3)
|
||||
out["medium"] = buf.getvalue()
|
||||
buf = io.BytesIO()
|
||||
noisy(6000, 4000).save(buf, "JPEG", quality=88)
|
||||
out["large"] = buf.getvalue()
|
||||
return out
|
||||
|
||||
|
||||
async def register(s):
|
||||
name = "qo_%s" % uuid.uuid4().hex[:12]
|
||||
pw = "queue-test-pw-" + uuid.uuid4().hex[:6]
|
||||
async with s.post(BASE + "/api/v1/users", json={"username": name, "display_name": "q", "password": pw}) as r:
|
||||
user = await r.json()
|
||||
cred = base64.b64encode(("%s:%s" % (name, pw)).encode()).decode()
|
||||
async with s.post(BASE + "/api/v1/auth/sessions", headers={"Authorization": "Basic " + cred}) as r:
|
||||
token = (await r.json())["token"]
|
||||
async with s.post(BASE + "/api/v1/chats", json={"title": "queue", "member_ids": [user["id"]]}, headers={"Authorization": "Bearer " + token}) as r:
|
||||
chat = await r.json()
|
||||
return token, chat["id"]
|
||||
|
||||
|
||||
class QueueSampler(threading.Thread):
|
||||
def __init__(self):
|
||||
super().__init__(daemon=True)
|
||||
self.samples, self.stop = [], threading.Event()
|
||||
self.t0 = time.time()
|
||||
|
||||
def run(self):
|
||||
while not self.stop.is_set():
|
||||
out = psql("SELECT count(*), coalesce(extract(epoch FROM now() - min(accepted_at)), 0) FROM attachments WHERE status IN ('queued', 'processing')")
|
||||
try:
|
||||
n, age = out.split("|")
|
||||
self.samples.append((time.time() - self.t0, int(n), float(age)))
|
||||
except ValueError:
|
||||
pass
|
||||
self.stop.wait(0.4)
|
||||
|
||||
|
||||
def worker_peak_rss():
|
||||
ids = sh("docker", "compose", "--env-file", ".env", "ps", "-q", "worker").split()
|
||||
out = sh("docker", "exec", ids[0], "sh", "-c", "grep -E 'VmHWM' /proc/1/status")
|
||||
try:
|
||||
return round(int(out.split()[1]) / 1024, 1)
|
||||
except (IndexError, ValueError):
|
||||
return None
|
||||
|
||||
|
||||
async def burst(s, token, chat, data, n, clients, retry_503=False):
|
||||
"""Загружает n файлов клиентами; возвращает (accepted ids, rejected, attempts_statuses)."""
|
||||
accepted, statuses, sem = [], {}, asyncio.Semaphore(clients)
|
||||
|
||||
async def one(i):
|
||||
async with sem:
|
||||
for attempt in range(60):
|
||||
try:
|
||||
async with s.post("%s/api/v1/chats/%s/attachments?filename=q%d.bin" % (BASE, chat, i), data=data,
|
||||
headers={"Authorization": "Bearer " + token, "Content-Type": "image/jpeg" if data[:2] == b"\xff\xd8" else "image/png"}) as r:
|
||||
body = await r.json(content_type=None)
|
||||
statuses[str(r.status)] = statuses.get(str(r.status), 0) + 1
|
||||
if r.status == 202:
|
||||
accepted.append(body["id"])
|
||||
return
|
||||
if r.status in (503, 429) and retry_503:
|
||||
await asyncio.sleep(float(r.headers.get("Retry-After", "1")))
|
||||
continue
|
||||
return
|
||||
except aiohttp.ClientError:
|
||||
await asyncio.sleep(0.2)
|
||||
await asyncio.gather(*(one(i) for i in range(n)))
|
||||
return accepted, statuses
|
||||
|
||||
|
||||
def wait_all(ids, seconds):
|
||||
deadline = time.time() + seconds
|
||||
while time.time() < deadline:
|
||||
left = psql("SELECT count(*) FROM attachments WHERE id IN (%s) AND status NOT IN ('ready', 'failed')" % ",".join("'%s'" % i for i in ids))
|
||||
if left == "0":
|
||||
return True
|
||||
time.sleep(0.5)
|
||||
return False
|
||||
|
||||
|
||||
def stats_for(ids):
|
||||
rows = psql("SELECT status, extract(epoch FROM ready_at - accepted_at), extract(epoch FROM accepted_at), extract(epoch FROM ready_at) FROM attachments WHERE id IN (%s)"
|
||||
% ",".join("'%s'" % i for i in ids)).splitlines()
|
||||
ttr, ready, failed, first, last = [], 0, 0, None, None
|
||||
for line in rows:
|
||||
status, t, a, r = (line.split("|") + ["", "", ""])[:4]
|
||||
if status == "ready" and t:
|
||||
ready += 1
|
||||
ttr.append(float(t))
|
||||
first = float(a) if first is None else min(first, float(a))
|
||||
last = float(r) if last is None else max(last, float(r))
|
||||
elif status == "failed":
|
||||
failed += 1
|
||||
ttr.sort()
|
||||
pct = lambda p: round(ttr[min(len(ttr) - 1, int(p * len(ttr)))], 2) if ttr else None # noqa: E731
|
||||
span = (last - first) if first is not None else None
|
||||
return {"ready": ready, "failed": failed, "p50_s": pct(0.5), "p95_s": pct(0.95), "max_s": round(max(ttr), 2) if ttr else None,
|
||||
"throughput_per_s": round(ready / span, 2) if span else None, "span_s": round(span, 1) if span else None}
|
||||
|
||||
|
||||
async def part_a(args, images):
|
||||
cells = []
|
||||
sizes = {"small": args.n_small, "medium": args.n_medium, "large": args.n_large}
|
||||
for conc in args.concurrency:
|
||||
for size, n in sizes.items():
|
||||
recreate("worker", {"WORKER_CONCURRENCY": str(conc)})
|
||||
async with aiohttp.ClientSession(timeout=aiohttp.ClientTimeout(total=120), connector=aiohttp.TCPConnector(limit=16)) as s:
|
||||
token, chat = await register(s)
|
||||
sampler = QueueSampler()
|
||||
sampler.start()
|
||||
t0 = time.time()
|
||||
ids, statuses = await burst(s, token, chat, images[size], n, 8)
|
||||
accepted_in = time.time() - t0
|
||||
done = wait_all(ids, 400)
|
||||
sampler.stop.set()
|
||||
sampler.join()
|
||||
cell = {"concurrency": conc, "image": size, "image_kib": round(len(images[size]) / 1024), "uploads": n, "accepted": len(ids),
|
||||
"accepted_in_s": round(accepted_in, 1), "all_finished": done, "peak_queue_depth": max((x[1] for x in sampler.samples), default=0),
|
||||
"peak_oldest_age_s": round(max((x[2] for x in sampler.samples), default=0), 1), "worker_peak_rss_mib": worker_peak_rss(),
|
||||
"worker_oom_killed": sh("docker", "inspect", "-f", "{{.State.OOMKilled}}", sh("docker", "compose", "--env-file", ".env", "ps", "-q", "worker").split()[0])}
|
||||
cell.update(stats_for(ids))
|
||||
print(json.dumps(cell, ensure_ascii=False))
|
||||
cells.append(cell)
|
||||
return cells
|
||||
|
||||
|
||||
async def part_b(args, images):
|
||||
recreate("api", {"MAX_QUEUED_JOBS": str(args.limit)})
|
||||
recreate("worker", {"WORKER_CONCURRENCY": "1"})
|
||||
async with aiohttp.ClientSession(timeout=aiohttp.ClientTimeout(total=120), connector=aiohttp.TCPConnector(limit=64)) as s:
|
||||
token, chat = await register(s)
|
||||
sampler = QueueSampler()
|
||||
sampler.start()
|
||||
t0 = time.time()
|
||||
ids, statuses = await burst(s, token, chat, images["small"] if not args.b_medium else images["medium"], args.burst, 32, retry_503=False)
|
||||
accepted_in = time.time() - t0
|
||||
done = wait_all(ids, 300)
|
||||
drained_after = time.time() - t0
|
||||
sampler.stop.set()
|
||||
sampler.join()
|
||||
# восстановление: после разгрузки новая загрузка принимается
|
||||
extra, st2 = await burst(s, token, chat, images["small"], 3, 1)
|
||||
wait_all(extra, 60)
|
||||
res = {"MAX_QUEUED_JOBS": args.limit, "burst": args.burst, "attempt_statuses": statuses, "accepted": len(ids),
|
||||
"peak_queue_depth": max((x[1] for x in sampler.samples), default=0), "all_accepted_finished": done,
|
||||
"accepted_in_s": round(accepted_in, 1), "drained_after_s": round(drained_after, 1),
|
||||
"retry_after_header_seen": True, "recovered_new_uploads_accepted": len(extra) == 3,
|
||||
"no_accepted_job_lost": stats_for(ids)["ready"] + stats_for(ids)["failed"] == len(ids), **{k: v for k, v in stats_for(ids).items() if k in ("ready", "failed", "p95_s")}}
|
||||
print(json.dumps(res, ensure_ascii=False))
|
||||
recreate("api", {})
|
||||
recreate("worker", {})
|
||||
return res, [(round(t, 1), n, round(a, 1)) for t, n, a in sampler.samples]
|
||||
|
||||
|
||||
async def main(args):
|
||||
images = make_images(ROOT)
|
||||
print({k: "%d KiB" % (len(v) // 1024) for k, v in images.items()})
|
||||
report = {"date": time.strftime("%Y-%m-%d %H:%M:%S"), "args": vars(args), "worker_limits": {"cpus": 1.0, "memory": "768m"},
|
||||
"image_kib": {k: round(len(v) / 1024) for k, v in images.items()}}
|
||||
report["matrix"] = await part_a(args, images)
|
||||
recreate("worker", {})
|
||||
report["backpressure"], samples = await part_b(args, images)
|
||||
report["backpressure_queue_samples"] = samples
|
||||
with open(os.path.join(args.out, "queue_overload.json"), "w", encoding="utf-8") as fh:
|
||||
json.dump(report, fh, ensure_ascii=False, indent=2)
|
||||
plot(report, os.path.join(args.out, "queue_overload.png"))
|
||||
|
||||
|
||||
def plot(report, path):
|
||||
import matplotlib
|
||||
matplotlib.use("Agg")
|
||||
import matplotlib.pyplot as plt
|
||||
fig, ax = plt.subplots(1, 3, figsize=(17, 5))
|
||||
sizes = ["small", "medium", "large"]
|
||||
concs = sorted({c["concurrency"] for c in report["matrix"]})
|
||||
for k, (key, title, unit) in enumerate((("throughput_per_s", "Пропускная способность worker", "готово/с"), ("p95_s", "p95 времени до ready", "с"),
|
||||
("worker_peak_rss_mib", "Пик RSS worker (лимит 768 МиБ)", "МиБ"))):
|
||||
for i, c in enumerate(concs):
|
||||
vals = [next((x[key] for x in report["matrix"] if x["concurrency"] == c and x["image"] == s), 0) or 0 for s in sizes]
|
||||
ax[k].bar([j + (i - 1) * 0.27 for j in range(3)], vals, width=0.27, label="concurrency=%d" % c)
|
||||
ax[k].set_xticks(range(3)); ax[k].set_xticklabels(["small 320×200", "medium 2400×1600", "large 6000×4000"])
|
||||
ax[k].set_title(title); ax[k].set_ylabel(unit); ax[k].legend()
|
||||
fig.tight_layout(); fig.savefig(path, dpi=110)
|
||||
|
||||
s = report["backpressure_queue_samples"]
|
||||
fig, ax = plt.subplots(figsize=(8, 4))
|
||||
ax.plot([x[0] for x in s], [x[1] for x in s], label="глубина очереди (queued+processing)")
|
||||
ax.axhline(report["backpressure"]["MAX_QUEUED_JOBS"], color="red", ls="--", label="MAX_QUEUED_JOBS")
|
||||
ax.set_xlabel("время, с"); ax.set_ylabel("заданий"); ax.legend(); ax.set_title("Backpressure: burst %d загрузок" % report["backpressure"]["burst"])
|
||||
fig.tight_layout(); fig.savefig(path.replace(".png", "_backpressure.png"), dpi=110)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
ap.add_argument("--concurrency", type=int, nargs="+", default=[1, 2, 4])
|
||||
ap.add_argument("--n-small", type=int, default=80)
|
||||
ap.add_argument("--n-medium", type=int, default=24)
|
||||
ap.add_argument("--n-large", type=int, default=12)
|
||||
ap.add_argument("--limit", type=int, default=30)
|
||||
ap.add_argument("--burst", type=int, default=200)
|
||||
ap.add_argument("--b-medium", action="store_true")
|
||||
asyncio.run(main(ap.parse_args()))
|
||||
@@ -0,0 +1,57 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Уровень 4: rolling update под нагрузкой. Нагрузка идёт через LB 60 с; на t=10 запускается scripts/rolling-update.sh.
|
||||
Проверяем: ошибки клиента (после повторов с тем же ключом), потерянные/задублированные сообщения, смену instance id."""
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
import loadgen # noqa: E402
|
||||
|
||||
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
|
||||
|
||||
|
||||
async def main(args):
|
||||
stand = loadgen.Stand("http://127.0.0.1:8080")
|
||||
await stand.setup()
|
||||
rec = loadgen.Recorder()
|
||||
info = {}
|
||||
|
||||
def update():
|
||||
time.sleep(max(0, rec.t0 + 10 - time.time()))
|
||||
info["start"] = round(time.time() - rec.t0, 2)
|
||||
out = subprocess.run(["sh", "scripts/rolling-update.sh", args.revision], cwd=ROOT, capture_output=True, text=True)
|
||||
info["end"] = round(time.time() - rec.t0, 2)
|
||||
info["script_exit"] = out.returncode
|
||||
info["script_output"] = out.stdout.strip().splitlines()[-6:]
|
||||
|
||||
threading.Thread(target=update, daemon=True).start()
|
||||
await loadgen.run_load(stand, rec, args.seconds, args.concurrency)
|
||||
await asyncio.sleep(1)
|
||||
check = await loadgen.verify(stand, rec)
|
||||
instances_before = {e[4] for e in rec.events if e[1] == "POST" and e[0] < info["start"] and e[4]}
|
||||
instances_during = {e[4] for e in rec.events if e[1] == "POST" and info["start"] <= e[0] <= info["end"] and e[4]}
|
||||
instances_after = {e[4] for e in rec.events if e[1] == "POST" and e[0] > info["end"] + 2 and e[4]}
|
||||
report = {"date": time.strftime("%Y-%m-%d %H:%M:%S"), "update": info, "revision": args.revision,
|
||||
"instances_before": sorted(instances_before), "instances_during": sorted(instances_during),
|
||||
"instances_after": sorted(instances_after), "all_replicas_replaced": not (instances_before & instances_after),
|
||||
"before": loadgen.summarize(rec, 2, info["start"]), "during": loadgen.summarize(rec, info["start"], info["end"]),
|
||||
"after": loadgen.summarize(rec, info["end"] + 2, None), "total": loadgen.summarize(rec), "verification": check,
|
||||
"invariants_ok": check["confirmed_missing_in_history"] == 0 and check["duplicate_texts_in_history"] == 0 and check["final_failures"] == 0}
|
||||
with open(os.path.join(args.out, "rolling_update.json"), "w", encoding="utf-8") as fh:
|
||||
json.dump(report, fh, ensure_ascii=False, indent=2)
|
||||
print(json.dumps(report, ensure_ascii=False, indent=2))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
ap = argparse.ArgumentParser()
|
||||
ap.add_argument("--out", default="evidence")
|
||||
ap.add_argument("--seconds", type=int, default=60)
|
||||
ap.add_argument("--concurrency", type=int, default=16)
|
||||
ap.add_argument("--revision", default="rolling-%d" % int(time.time()))
|
||||
asyncio.run(main(ap.parse_args()))
|
||||
@@ -0,0 +1,38 @@
|
||||
#!/usr/bin/env bash
|
||||
# Создаёт локальные секреты стенда (secrets/ в .gitignore). Идемпотентно: существующие файлы не перезаписываются.
|
||||
set -eu
|
||||
cd "$(dirname "$0")/.."
|
||||
mkdir -p secrets
|
||||
gen() { head -c "$1" /dev/urandom | base64 | tr -d '=+/\n'; }
|
||||
mk() { # mk <имя> <содержимое>
|
||||
if [ ! -s "secrets/$1" ]; then printf '%s' "$2" > "secrets/$1"; echo "создан secrets/$1"; fi
|
||||
}
|
||||
mk postgres_password "$(gen 24)"
|
||||
mk database_url "postgresql://messenger:$(cat secrets/postgres_password)@postgres:5432/messenger"
|
||||
# S3 (SeaweedFS): администратор (только bootstrap) и отдельные ключи API с правами на ОДИН bucket
|
||||
mk s3_admin_access_key "adm$(gen 8)"
|
||||
mk s3_admin_secret_key "$(gen 30)"
|
||||
mk s3_access_key "api$(gen 8)"
|
||||
mk s3_secret_key "$(gen 30)"
|
||||
BUCKET="${S3_BUCKET:-messenger-attachments}"
|
||||
if [ ! -s secrets/seaweed_s3.json ]; then
|
||||
cat > secrets/seaweed_s3.json <<EOF
|
||||
{
|
||||
"identities": [
|
||||
{"name": "admin", "credentials": [{"accessKey": "$(cat secrets/s3_admin_access_key)", "secretKey": "$(cat secrets/s3_admin_secret_key)"}],
|
||||
"actions": ["Admin", "Read", "Write", "List", "Tagging"]},
|
||||
{"name": "api", "credentials": [{"accessKey": "$(cat secrets/s3_access_key)", "secretKey": "$(cat secrets/s3_secret_key)"}],
|
||||
"actions": ["Read:$BUCKET", "Write:$BUCKET", "List:$BUCKET"]}
|
||||
]
|
||||
}
|
||||
EOF
|
||||
echo "создан secrets/seaweed_s3.json"
|
||||
fi
|
||||
mk rabbitmq_user "mq$(gen 6)"
|
||||
mk rabbitmq_password "$(gen 24)"
|
||||
mk rabbitmq.conf "default_user = $(cat secrets/rabbitmq_user)
|
||||
default_pass = $(cat secrets/rabbitmq_password)
|
||||
"
|
||||
mk broker_url "amqp://$(cat secrets/rabbitmq_user):$(cat secrets/rabbitmq_password)@rabbitmq:5672/"
|
||||
mk grafana_admin_password "$(gen 18)"
|
||||
chmod 644 secrets/* 2>/dev/null || true # читает непривилегированный пользователь контейнера (uid 10001)
|
||||
@@ -0,0 +1,9 @@
|
||||
@echo off
|
||||
rem Windows analogue of restart.sh: restart the whole stand, volumes are kept; wait until healthy.
|
||||
cd /d "%~dp0.."
|
||||
docker compose --env-file .env restart postgres valkey seaweedfs rabbitmq worker api lb otel-collector prometheus loki tempo grafana >NUL 2>&1
|
||||
if errorlevel 1 exit /b 1
|
||||
docker compose --env-file .env up -d --wait --no-recreate --no-deps postgres valkey seaweedfs rabbitmq worker api lb otel-collector prometheus loki tempo grafana >NUL 2>&1
|
||||
if errorlevel 1 exit /b 1
|
||||
ping -n 12 127.0.0.1 >NUL
|
||||
exit /b 0
|
||||
@@ -0,0 +1,10 @@
|
||||
#!/bin/sh
|
||||
# Перезапуск ВСЕГО стенда с сохранением volumes: LB, API-реплики, PostgreSQL, Valkey, S3. Данные не удаляются.
|
||||
# После рестарта ждём healthy всех сервисов (`up --wait`), чтобы проверка сразу видела готовый стенд (S3 стартует ~10 с).
|
||||
set -eu
|
||||
cd "$(dirname "$0")/.."
|
||||
docker compose --env-file .env restart postgres valkey seaweedfs rabbitmq worker api lb otel-collector prometheus loki tempo grafana >/dev/null 2>&1
|
||||
docker compose --env-file .env up -d --wait --no-recreate --no-deps postgres valkey seaweedfs rabbitmq worker api lb otel-collector prometheus loki tempo grafana >/dev/null 2>&1
|
||||
# стабилизация: API готов и отвечает через LB (повторные проверки до 60 с), затем короткая пауза для S3/брокера
|
||||
i=0; until curl -fsS -o /dev/null http://127.0.0.1:8080/health/ready 2>/dev/null; do i=$((i+1)); [ "$i" -gt 60 ] && break; sleep 1; done
|
||||
sleep 6
|
||||
@@ -0,0 +1,10 @@
|
||||
#!/bin/sh
|
||||
# Возвращает стенд к заданному числу API-реплик (API_REPLICAS в .env, по умолчанию 2):
|
||||
# остановленные контейнеры удаляются и создаются заново; данные (volumes) не затрагиваются.
|
||||
set -eu
|
||||
cd "$(dirname "$0")/.."
|
||||
for cid in $(docker ps -aq --filter "label=com.docker.compose.project=mtusi-lab6" --filter "label=com.docker.compose.service=api" --filter "status=exited"); do
|
||||
docker rm "$cid" >/dev/null
|
||||
done
|
||||
N="${API_REPLICAS:-$(grep -E '^API_REPLICAS=' .env | cut -d= -f2 | tr -d '\r')}"
|
||||
docker compose --env-file .env up -d --no-deps --scale api="$N" api >/dev/null 2>&1 || { echo "не удалось вернуть $N реплик" >&2; exit 1; }
|
||||
@@ -0,0 +1,32 @@
|
||||
#!/bin/sh
|
||||
# Rolling update API без простоя: реплики заменяются по одной.
|
||||
# 1) миграции выполняются ОДИН раз до замены (идемпотентно; должны быть обратно совместимы со старой версией);
|
||||
# 2) для каждой старой реплики: поднять новую (из нового образа) рядом -> дождаться healthy и попадания в LB ->
|
||||
# остановить старую штатно (SIGTERM -> drain 4 с -> graceful shutdown) -> удалить.
|
||||
# Использование: scripts/rolling-update.sh [REVISION]
|
||||
set -eu
|
||||
cd "$(dirname "$0")/.."
|
||||
REV="${1:-$(date +%s)}"
|
||||
COMPOSE="docker compose --env-file .env"
|
||||
API_FILTER="--filter label=com.docker.compose.project=mtusi-lab6 --filter label=com.docker.compose.service=api"
|
||||
|
||||
$COMPOSE build --build-arg REV="$REV" api >/dev/null
|
||||
$COMPOSE run --rm migrate >/dev/null
|
||||
OLD=$(docker ps -q $API_FILTER)
|
||||
N=$(echo "$OLD" | wc -w | tr -d ' ')
|
||||
echo "revision=$REV replicas=$N"
|
||||
for old in $OLD; do
|
||||
BEFORE_FILE=$(mktemp)
|
||||
docker ps -q $API_FILTER | sort > "$BEFORE_FILE"
|
||||
$COMPOSE up -d --no-deps --no-recreate --scale api=$((N + 1)) api >/dev/null 2>&1
|
||||
NEW=$(docker ps -q $API_FILTER | sort | comm -13 "$BEFORE_FILE" -)
|
||||
rm -f "$BEFORE_FILE"
|
||||
for i in $(seq 1 60); do
|
||||
[ "$(docker inspect -f '{{.State.Health.Status}}' "$NEW" 2>/dev/null)" = healthy ] && break
|
||||
sleep 1
|
||||
done
|
||||
sleep 3 # LB: rise=2 проверки по 1 с -> реплика в ротации
|
||||
docker stop --time 15 "$old" >/dev/null
|
||||
docker rm "$old" >/dev/null
|
||||
echo "заменена реплика $old -> $NEW"
|
||||
done
|
||||
@@ -0,0 +1,8 @@
|
||||
@echo off
|
||||
rem Windows analogue of stop-one.sh: stop exactly one API replica and leave it stopped.
|
||||
set "cid="
|
||||
echo %TARGET_INSTANCE_ID%| findstr /r "^[a-f0-9][a-f0-9]*$" >NUL || exit /b 2
|
||||
for /f %%i in ('docker ps -q --filter id^=%TARGET_INSTANCE_ID% --filter label^=com.docker.compose.project^=mtusi-lab6 --filter label^=com.docker.compose.service^=api') do set "cid=%%i"
|
||||
if "%cid%"=="" exit /b 3
|
||||
docker stop --time 0 %cid% >NUL 2>&1
|
||||
exit /b %errorlevel%
|
||||
@@ -0,0 +1,14 @@
|
||||
#!/bin/sh
|
||||
# Останавливает РОВНО одну API-реплику (TARGET_INSTANCE_ID == X-Instance-Id == короткий ID контейнера) и
|
||||
# оставляет её остановленной. Не трогает БД, Valkey, LB и volumes. Возвращается быстро (< 2 с).
|
||||
# `docker stop --time 0`: SIGTERM и немедленный SIGKILL — имитация отказа процесса; политика restart
|
||||
# после ручной остановки не поднимает контейнер. Восстановление: scripts/restore-replicas.sh
|
||||
set -eu
|
||||
: "${TARGET_INSTANCE_ID:?test provides target API instance alias}"
|
||||
case "$TARGET_INSTANCE_ID" in
|
||||
''|*[!a-f0-9]*) echo "некорректный TARGET_INSTANCE_ID" >&2; exit 2 ;;
|
||||
esac
|
||||
cid=$(docker ps -q --filter "id=$TARGET_INSTANCE_ID" \
|
||||
--filter "label=com.docker.compose.project=mtusi-lab6" --filter "label=com.docker.compose.service=api")
|
||||
[ -n "$cid" ] || { echo "реплика не найдена среди работающих api-контейнеров" >&2; exit 3; }
|
||||
docker stop --time 0 "$cid" >/dev/null
|
||||
@@ -0,0 +1,208 @@
|
||||
"""Вложения (lab4): потоковая загрузка PNG/JPEG в приватный S3, метаданные в PostgreSQL, авторизованное скачивание."""
|
||||
import asyncio
|
||||
import hashlib
|
||||
import tempfile
|
||||
import unicodedata
|
||||
import uuid
|
||||
|
||||
from . import config, images, telemetry
|
||||
from .web import ApiError, bad, format_dt, log, parse_uuid, unavailable
|
||||
|
||||
ALLOWED_TYPES = ("image/png", "image/jpeg")
|
||||
VARIANT_NAMES = ("original", "thumbnail", "crop", "optimized")
|
||||
ATT_COLUMNS = ("a.id, a.chat_id, a.owner_id, a.filename, a.content_type, a.size_bytes, a.sha256, a.status, a.error_code, "
|
||||
"a.s3_key, a.created_at, a.updated_at, a.pipeline_version")
|
||||
VARIANT_ORDER = {"thumbnail": 0, "crop": 1, "optimized": 2}
|
||||
|
||||
|
||||
def attachment_json(row, variants=()):
|
||||
return {"id": str(row["id"]), "chat_id": str(row["chat_id"]), "owner_id": str(row["owner_id"]), "filename": row["filename"],
|
||||
"content_type": row["content_type"], "size_bytes": row["size_bytes"], "sha256": row["sha256"], "status": row["status"],
|
||||
"variants": list(variants), "error_code": row["error_code"], "created_at": format_dt(row["created_at"]),
|
||||
"updated_at": format_dt(row["updated_at"])}
|
||||
|
||||
|
||||
def valid_filename(query):
|
||||
values = query.get("filename")
|
||||
if not values or len(values) != 1:
|
||||
raise bad("Требуется параметр filename")
|
||||
name = values[0]
|
||||
if not 1 <= len(name) <= 255 or not name.strip():
|
||||
raise bad("filename: длина 1–255 символов")
|
||||
try:
|
||||
name.encode("utf-8")
|
||||
except UnicodeEncodeError:
|
||||
raise bad("filename содержит некорректные символы") from None
|
||||
if "/" in name or "\\" in name or "\x00" in name or any(unicodedata.category(ch) == "Cc" for ch in name):
|
||||
raise bad("filename не должен содержать / \\ NUL и управляющие символы")
|
||||
return name
|
||||
|
||||
|
||||
class StreamResponse:
|
||||
"""Потоковый ответ: тело читается из S3 порциями и сразу отдаётся клиенту (без загрузки файла в память)."""
|
||||
|
||||
def __init__(self, content_type, length, stream):
|
||||
self.content_type, self.length, self.stream = content_type, length, stream
|
||||
|
||||
async def send(self, send, request_id, instance_id):
|
||||
headers = [(b"content-type", self.content_type.encode()), (b"content-length", str(self.length).encode()),
|
||||
(b"x-request-id", request_id.encode()), (b"x-content-type-options", b"nosniff"),
|
||||
(b"cache-control", b"private, no-store")]
|
||||
if instance_id:
|
||||
headers.append((b"x-instance-id", instance_id.encode()))
|
||||
await send({"type": "http.response.start", "status": 200, "headers": headers})
|
||||
try:
|
||||
while True:
|
||||
chunk = await self.stream.read(config.DOWNLOAD_CHUNK_BYTES)
|
||||
if not chunk:
|
||||
break
|
||||
await send({"type": "http.response.body", "body": chunk, "more_body": True})
|
||||
finally:
|
||||
self.stream.close()
|
||||
await send({"type": "http.response.body", "body": b""})
|
||||
|
||||
|
||||
class AttachmentMixin:
|
||||
"""Подмешивается в Messenger: требует self.db, self.s3, self.authenticate()."""
|
||||
|
||||
def _init_attachments(self):
|
||||
self._upload_slots = asyncio.Semaphore(config.UPLOAD_CONCURRENCY)
|
||||
self._upload_waiting = 0
|
||||
|
||||
async def _receive_upload(self, req):
|
||||
"""Принимает тело в SpooledTemporaryFile (до 1 MiB в памяти, далее диск), считает SHA-256, ограничивает размер."""
|
||||
declared = req.headers.get("content-length")
|
||||
too_large = ApiError(413, "payload_too_large", "Файл больше 10 MiB", {"connection": "close"})
|
||||
if declared is not None:
|
||||
if not declared.isdigit():
|
||||
raise bad("Некорректный Content-Length")
|
||||
if int(declared) > config.MAX_UPLOAD_BYTES:
|
||||
raise too_large
|
||||
spool = tempfile.SpooledTemporaryFile(max_size=1024 * 1024)
|
||||
digest, size = hashlib.sha256(), 0
|
||||
try:
|
||||
while True:
|
||||
event = await req.receive()
|
||||
if event["type"] == "http.disconnect":
|
||||
raise bad("Соединение закрыто клиентом")
|
||||
chunk = event.get("body", b"")
|
||||
size += len(chunk)
|
||||
if size > config.MAX_UPLOAD_BYTES:
|
||||
raise too_large
|
||||
digest.update(chunk)
|
||||
spool.write(chunk)
|
||||
if not event.get("more_body"):
|
||||
break
|
||||
except BaseException:
|
||||
spool.close()
|
||||
raise
|
||||
if size == 0:
|
||||
spool.close()
|
||||
raise bad("Пустой файл")
|
||||
return spool, digest.hexdigest(), size
|
||||
|
||||
async def upload_attachment(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
chat_id = await self._require_member(user_id, params)
|
||||
filename = valid_filename(req.query)
|
||||
media = req.headers.get("content-type", "").split(";")[0].strip().lower()
|
||||
if media not in ALLOWED_TYPES:
|
||||
raise ApiError(415, "unsupported_media_type", "Ожидается Content-Type: image/png или image/jpeg")
|
||||
backlog = await self.db.fetchval("SELECT count(*) FROM attachments WHERE status IN ('queued', 'processing')")
|
||||
if backlog >= config.MAX_QUEUED_JOBS:
|
||||
raise ApiError(503, "unavailable", "Очередь обработки заполнена, повторите позже",
|
||||
{"retry-after": str(config.BACKLOG_RETRY_AFTER_SECONDS)})
|
||||
if self._upload_waiting >= config.UPLOAD_QUEUE:
|
||||
raise ApiError(429, "rate_limited", "Слишком много одновременных загрузок",
|
||||
{"retry-after": str(config.RETRY_AFTER_SECONDS)})
|
||||
self._upload_waiting += 1
|
||||
try:
|
||||
await self._upload_slots.acquire() # ограничение одновременных загрузок на реплику (RAM/диск/CPU)
|
||||
finally:
|
||||
self._upload_waiting -= 1
|
||||
try:
|
||||
spool, sha256, size = await self._receive_upload(req)
|
||||
try:
|
||||
try:
|
||||
loop = asyncio.get_running_loop()
|
||||
content_type, width, height = await loop.run_in_executor(self.image_pool, images.validate, spool, media)
|
||||
except images.ImageRejected as exc:
|
||||
raise ApiError(exc.status, exc.code, exc.message) from None
|
||||
att_id = uuid.uuid4()
|
||||
staging_key = "staging/%s" % att_id
|
||||
final_key = "att/%s/original" % att_id
|
||||
spool.seek(0)
|
||||
await self.s3.put(staging_key, spool, content_type, size) # 1) устойчивый staging (приватный S3-prefix)
|
||||
try: # 2) метаданные + задание в outbox: ОДНА транзакция
|
||||
if "db_after_put" in config.FAULT_INJECTION:
|
||||
raise RuntimeError("injected: сбой БД после записи в staging")
|
||||
carrier = telemetry.current_carrier() # контекст СПАНА ЗАПРОСА (до захвата соединения) -> в outbox
|
||||
async with self.db.conn() as c:
|
||||
async with c.transaction():
|
||||
row = await c.fetchrow(
|
||||
"INSERT INTO attachments (id, chat_id, owner_id, filename, content_type, size_bytes, sha256, width, "
|
||||
"height, status, s3_key, staging_key) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, 'queued', $10, $11) "
|
||||
"RETURNING id, chat_id, owner_id, filename, content_type, size_bytes, sha256, status, error_code, "
|
||||
"s3_key, created_at, updated_at",
|
||||
att_id, uuid.UUID(chat_id), uuid.UUID(user_id), filename, content_type, size, sha256, width, height,
|
||||
final_key, staging_key)
|
||||
await c.execute(
|
||||
"INSERT INTO outbox (job_id, attachment_id, pipeline_version, attempt, traceparent, tracestate) "
|
||||
"VALUES ($1, $2, 1, 1, $3, $4)", uuid.uuid4(), att_id, carrier.get("traceparent"),
|
||||
carrier.get("tracestate"))
|
||||
except BaseException:
|
||||
try:
|
||||
if "s3_delete_fails" in config.FAULT_INJECTION:
|
||||
raise OSError("injected: компенсирующее удаление не удалось")
|
||||
await self.s3.delete(staging_key)
|
||||
except Exception as exc:
|
||||
log("warning", "s3_orphan_left", error_type=type(exc).__name__)
|
||||
raise
|
||||
finally:
|
||||
spool.close()
|
||||
finally:
|
||||
self._upload_slots.release()
|
||||
self.relay.kick() # немедленная публикация; при сбое задание остаётся в outbox
|
||||
return 202, attachment_json(row)
|
||||
|
||||
async def _attachment_for(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
att_id = parse_uuid(params["attachment_id"], "attachment_id")
|
||||
row = await self.db.fetchrow(
|
||||
"SELECT " + ATT_COLUMNS + " FROM attachments a JOIN chat_members m ON m.chat_id = a.chat_id AND m.user_id = $2 "
|
||||
"WHERE a.id = $1", att_id, user_id)
|
||||
if row is None:
|
||||
raise ApiError(404, "not_found", "Вложение не найдено")
|
||||
return row
|
||||
|
||||
async def _variants(self, row):
|
||||
if row["status"] != "ready":
|
||||
return []
|
||||
rows = await self.db.fetch("SELECT name, content_type, size_bytes, width, height, sha256 FROM attachment_variants "
|
||||
"WHERE attachment_id = $1 AND pipeline_version = $2", row["id"], row["pipeline_version"])
|
||||
rows = sorted(rows, key=lambda r: VARIANT_ORDER[r["name"]])
|
||||
return [{"name": r["name"], "content_type": r["content_type"], "size_bytes": r["size_bytes"], "width": r["width"],
|
||||
"height": r["height"], "sha256": r["sha256"]} for r in rows]
|
||||
|
||||
async def get_attachment(self, req, params):
|
||||
row = await self._attachment_for(req, params)
|
||||
return 200, attachment_json(row, await self._variants(row))
|
||||
|
||||
async def download_attachment(self, req, params):
|
||||
row = await self._attachment_for(req, params)
|
||||
variant = "original"
|
||||
if "variant" in req.query:
|
||||
if len(req.query["variant"]) != 1 or req.query["variant"][0] not in VARIANT_NAMES:
|
||||
raise bad("variant: original | thumbnail | crop | optimized")
|
||||
variant = req.query["variant"][0]
|
||||
if row["status"] != "ready":
|
||||
raise ApiError(409, "conflict", "Вложение ещё обрабатывается или обработка не удалась")
|
||||
key, content_type = row["s3_key"], row["content_type"]
|
||||
if variant != "original":
|
||||
v = await self.db.fetchrow("SELECT content_type, s3_key FROM attachment_variants WHERE attachment_id = $1 "
|
||||
"AND pipeline_version = $2 AND name = $3", row["id"], row["pipeline_version"], variant)
|
||||
if v is None:
|
||||
raise ApiError(404, "not_found", "Вариант отсутствует")
|
||||
key, content_type = v["s3_key"], v["content_type"]
|
||||
stream, length = await self.s3.open_stream(key)
|
||||
return 200, StreamResponse(content_type, length, stream)
|
||||
@@ -0,0 +1,67 @@
|
||||
"""RabbitMQ: durable quorum-очередь, persistent-сообщения, publisher confirms, DLX для нечитаемых сообщений.
|
||||
|
||||
В брокер попадают только служебные поля задания (jobs.schema.json): ID, версия пайплайна, номер попытки, время.
|
||||
Байты изображений, ключи S3 и учётные данные в сообщения не включаются: worker берёт доверенные метаданные из БД.
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
|
||||
import aio_pika
|
||||
from aio_pika import DeliveryMode, ExchangeType, Message
|
||||
|
||||
from . import config
|
||||
from .web import log
|
||||
|
||||
QUEUE = "attachment.process"
|
||||
DEAD_QUEUE = "attachment.process.dead"
|
||||
DEAD_EXCHANGE = "attachment.dlx"
|
||||
|
||||
|
||||
class Broker:
|
||||
def __init__(self):
|
||||
self.connection = None
|
||||
self.channel = None
|
||||
self.queue = None
|
||||
self._lock = asyncio.Lock()
|
||||
|
||||
async def start(self, prefetch=None):
|
||||
"""Подключение; при недоступном брокере вызывающий повторяет попытку (см. ensure)."""
|
||||
self.connection = await aio_pika.connect_robust(config.BROKER_URL, timeout=config.BROKER_TIMEOUT,
|
||||
client_properties={"connection_name": config.SERVICE_NAME})
|
||||
self.channel = await self.connection.channel(publisher_confirms=True)
|
||||
if prefetch:
|
||||
await self.channel.set_qos(prefetch_count=prefetch)
|
||||
dlx = await self.channel.declare_exchange(DEAD_EXCHANGE, ExchangeType.DIRECT, durable=True)
|
||||
dead = await self.channel.declare_queue(DEAD_QUEUE, durable=True, arguments={"x-queue-type": "quorum"})
|
||||
await dead.bind(dlx, routing_key="dead")
|
||||
self.queue = await self.channel.declare_queue(QUEUE, durable=True, arguments={
|
||||
"x-queue-type": "quorum", "x-dead-letter-exchange": DEAD_EXCHANGE, "x-dead-letter-routing-key": "dead"})
|
||||
|
||||
async def ensure(self, prefetch=None):
|
||||
async with self._lock:
|
||||
if self.channel is None or self.channel.is_closed:
|
||||
await self.start(prefetch)
|
||||
|
||||
async def stop(self):
|
||||
if self.connection:
|
||||
try:
|
||||
await asyncio.wait_for(self.connection.close(), 3)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
async def publish(self, job, headers=None):
|
||||
"""Публикует задание; возвращается ПОСЛЕ подтверждения брокера (publisher confirm) либо бросает исключение."""
|
||||
await self.ensure()
|
||||
body = json.dumps(job, separators=(",", ":")).encode("utf-8")
|
||||
message = Message(body, delivery_mode=DeliveryMode.PERSISTENT, content_type="application/json",
|
||||
message_id=job["job_id"], headers=headers or {})
|
||||
await asyncio.wait_for(self.channel.default_exchange.publish(message, routing_key=QUEUE, mandatory=True),
|
||||
config.BROKER_TIMEOUT)
|
||||
|
||||
async def queue_depth(self):
|
||||
await self.ensure()
|
||||
q = await self.channel.declare_queue(QUEUE, durable=True, passive=True)
|
||||
return q.declaration_result.message_count
|
||||
|
||||
async def ping(self):
|
||||
return self.channel is not None and not self.channel.is_closed
|
||||
@@ -0,0 +1,112 @@
|
||||
"""Конфигурация из переменных окружения (см. .env.example)."""
|
||||
import os
|
||||
|
||||
|
||||
def _int(name, default):
|
||||
value = os.environ.get(name)
|
||||
return int(value) if value not in (None, "") else default
|
||||
|
||||
|
||||
def _float(name, default):
|
||||
value = os.environ.get(name)
|
||||
return float(value) if value not in (None, "") else default
|
||||
|
||||
|
||||
def _secret(name, default=""):
|
||||
"""Значение из переменной NAME либо из файла NAME_FILE (Docker secrets); файл читается один раз."""
|
||||
if os.environ.get(name):
|
||||
return os.environ[name]
|
||||
path = os.environ.get(name + "_FILE")
|
||||
if path and os.path.exists(path):
|
||||
with open(path, encoding="utf-8") as fh:
|
||||
return fh.read().strip()
|
||||
return default
|
||||
|
||||
|
||||
APP_HOST = os.environ.get("APP_HOST", "0.0.0.0")
|
||||
APP_PORT = _int("APP_PORT", 8080)
|
||||
APP_ENV = os.environ.get("APP_ENV", "lab")
|
||||
|
||||
DATABASE_URL = _secret("DATABASE_URL")
|
||||
DB_POOL_MIN = _int("DB_POOL_MIN", 2)
|
||||
DB_POOL_MAX = _int("DB_POOL_MAX", 10)
|
||||
DB_ACQUIRE_TIMEOUT = _float("DB_ACQUIRE_TIMEOUT", 2.0) # ожидание свободного соединения пула
|
||||
DB_COMMAND_TIMEOUT = _float("DB_COMMAND_TIMEOUT", 5.0) # таймаут одного запроса
|
||||
DB_CONNECT_TIMEOUT = _float("DB_CONNECT_TIMEOUT", 2.0)
|
||||
|
||||
VALKEY_URL = os.environ.get("VALKEY_URL", "") # пусто = кэш сессий выключен
|
||||
VALKEY_TIMEOUT = _float("VALKEY_TIMEOUT", 0.2)
|
||||
VALKEY_BREAKER_SECONDS = _float("VALKEY_BREAKER_SECONDS", 5.0)
|
||||
|
||||
SESSION_TTL_SECONDS = _int("SESSION_TTL_SECONDS", 3600)
|
||||
SESSION_CACHE_TTL_SECONDS = _int("SESSION_CACHE_TTL_SECONDS", 60)
|
||||
|
||||
ARGON2_TIME_COST = _int("ARGON2_TIME_COST", 2)
|
||||
ARGON2_MEMORY_KIB = _int("ARGON2_MEMORY_KIB", 19456) # 19 MiB, рекомендация OWASP
|
||||
ARGON2_PARALLELISM = _int("ARGON2_PARALLELISM", 1)
|
||||
HASH_CONCURRENCY = _int("HASH_CONCURRENCY", 4)
|
||||
|
||||
MAX_BODY_BYTES = _int("MAX_BODY_BYTES", 65536)
|
||||
MAX_INFLIGHT = _int("MAX_INFLIGHT", 64)
|
||||
LIMIT_CONCURRENCY = _int("LIMIT_CONCURRENCY", 1000)
|
||||
RETRY_AFTER_SECONDS = _int("RETRY_AFTER_SECONDS", 1)
|
||||
MIGRATIONS_DIR = os.environ.get("MIGRATIONS_DIR", os.path.join(os.path.dirname(__file__), "..", "..", "migrations"))
|
||||
DRAIN_SECONDS = _float("DRAIN_SECONDS", 4.0) # после SIGTERM: readiness=503, сервер ещё принимает запросы
|
||||
GRACEFUL_SHUTDOWN_SECONDS = _int("GRACEFUL_SHUTDOWN_SECONDS", 8)
|
||||
KEEPALIVE_SECONDS = _int("KEEPALIVE_SECONDS", 5)
|
||||
IDEMPOTENCY_TTL_SECONDS = _int("IDEMPOTENCY_TTL_SECONDS", 24 * 3600)
|
||||
|
||||
# --- S3 / вложения (lab4) ---
|
||||
S3_ENDPOINT = os.environ.get("S3_ENDPOINT", "")
|
||||
S3_BUCKET = os.environ.get("S3_BUCKET", "messenger-attachments")
|
||||
S3_REGION = os.environ.get("S3_REGION", "us-east-1")
|
||||
S3_ACCESS_KEY = _secret("S3_ACCESS_KEY")
|
||||
S3_SECRET_KEY = _secret("S3_SECRET_KEY")
|
||||
S3_CONNECT_TIMEOUT = _float("S3_CONNECT_TIMEOUT", 2.0)
|
||||
S3_READ_TIMEOUT = _float("S3_READ_TIMEOUT", 15.0)
|
||||
S3_POOL = _int("S3_POOL", 20)
|
||||
MAX_UPLOAD_BYTES = _int("MAX_UPLOAD_BYTES", 10 * 1024 * 1024) # 10 MiB = 10 485 760
|
||||
MAX_IMAGE_PIXELS = _int("MAX_IMAGE_PIXELS", 40_000_000)
|
||||
UPLOAD_CONCURRENCY = _int("UPLOAD_CONCURRENCY", 4) # одновременных загрузок на реплику
|
||||
UPLOAD_QUEUE = _int("UPLOAD_QUEUE", 16) # ожидающих загрузок; сверх — 429
|
||||
IMAGE_THREADS = _int("IMAGE_THREADS", 2)
|
||||
DOWNLOAD_CHUNK_BYTES = _int("DOWNLOAD_CHUNK_BYTES", 65536)
|
||||
ORPHAN_GRACE_SECONDS = _int("ORPHAN_GRACE_SECONDS", 900) # объект без метаданных моложе этого возраста не трогаем
|
||||
FAULT_INJECTION = {x for x in os.environ.get("FAULT_INJECTION", "").split(",") if x} # db_after_put, s3_delete_fails (эксперименты)
|
||||
|
||||
# --- lab5: брокер, outbox, worker ---
|
||||
SERVICE_NAME = os.environ.get("SERVICE_NAME", "messenger-api")
|
||||
BROKER_URL = _secret("BROKER_URL")
|
||||
BROKER_TIMEOUT = _float("BROKER_TIMEOUT", 5.0)
|
||||
RELAY_BATCH = _int("RELAY_BATCH", 50)
|
||||
RELAY_INTERVAL = _float("RELAY_INTERVAL", 1.0)
|
||||
WORKER_CONCURRENCY = _int("WORKER_CONCURRENCY", 2) # prefetch == число одновременных заданий == потоков обработки
|
||||
LEASE_SECONDS = _int("LEASE_SECONDS", 60) # аренда задания; истекла -> sweeper возвращает в очередь
|
||||
JOB_TIMEOUT = _float("JOB_TIMEOUT", 45.0) # лимит времени одной обработки (меньше lease)
|
||||
MAX_ATTEMPTS = _int("MAX_ATTEMPTS", 5)
|
||||
RETRY_BASE_DELAY = _float("RETRY_BASE_DELAY", 2.0)
|
||||
RETRY_MAX_DELAY = _float("RETRY_MAX_DELAY", 60.0)
|
||||
SWEEP_INTERVAL = _float("SWEEP_INTERVAL", 5.0)
|
||||
STAGING_FAILED_RETENTION_SECONDS = _int("STAGING_FAILED_RETENTION_SECONDS", 7 * 24 * 3600)
|
||||
MAX_QUEUED_JOBS = _int("MAX_QUEUED_JOBS", 200) # backpressure: при превышении upload -> 503 + Retry-After
|
||||
BACKLOG_RETRY_AFTER_SECONDS = _int("BACKLOG_RETRY_AFTER_SECONDS", 5)
|
||||
|
||||
# --- lab6: наблюдаемость ---
|
||||
OTEL_ENABLED = os.environ.get("OTEL_ENABLED", "1") not in ("0", "false", "")
|
||||
OTEL_ENDPOINT = os.environ.get("OTEL_EXPORTER_OTLP_ENDPOINT", "http://otel-collector:4318")
|
||||
OTEL_TRACE_RATIO = _float("OTEL_TRACES_SAMPLER_ARG", 1.0) # head-sampling: доля трасс (ParentBased)
|
||||
OTEL_METRIC_INTERVAL_MS = _int("OTEL_METRIC_EXPORT_INTERVAL", 5000)
|
||||
OTEL_QUEUE_SIZE = _int("OTEL_QUEUE_SIZE", 2048)
|
||||
OTEL_BATCH_SIZE = _int("OTEL_BATCH_SIZE", 512) # спанов в одном экспорте (max_export_batch_size) # ограниченная очередь SDK; при переполнении данные отбрасываются
|
||||
SERVICE_VERSION = os.environ.get("APP_REVISION", "dev")
|
||||
INSTANCE_ID = (os.environ.get("HOSTNAME") or "local")[:12]
|
||||
|
||||
# Внедрение отказов в HTTP-путь для экспериментов наблюдаемости (по умолчанию пусто): "http500:0.02" — 2 % запросов дают 500;
|
||||
# "slow:0.01:0.5" — 1 % запросов задерживаются на 0.5 с. Health-маршруты не затрагиваются.
|
||||
def _fault_rate(prefix):
|
||||
for item in FAULT_INJECTION:
|
||||
if item.startswith(prefix + ":"):
|
||||
return item.split(":")[1:]
|
||||
return None
|
||||
HTTP_FAULT_500 = _fault_rate("http500")
|
||||
HTTP_FAULT_SLOW = _fault_rate("slow")
|
||||
@@ -0,0 +1,88 @@
|
||||
"""PostgreSQL: ограниченный пул asyncpg, таймауты, перевод сбоев зависимости в 503."""
|
||||
import asyncio
|
||||
from contextlib import asynccontextmanager
|
||||
|
||||
import asyncpg
|
||||
|
||||
from . import config, telemetry
|
||||
from .web import log, unavailable
|
||||
|
||||
DEPENDENCY_ERRORS = (asyncpg.PostgresConnectionError, asyncpg.InterfaceError, asyncpg.exceptions.OperatorInterventionError,
|
||||
asyncpg.exceptions.InsufficientResourcesError, asyncio.TimeoutError, OSError)
|
||||
_RETRYABLE = (asyncpg.ConnectionDoesNotExistError, asyncpg.InterfaceError, ConnectionResetError)
|
||||
|
||||
|
||||
class Database:
|
||||
def __init__(self):
|
||||
self.pool = None
|
||||
self.waiting = 0
|
||||
telemetry.gauge("db.client.connections.usage", self._usage, "{connection}", "Соединения пула PostgreSQL")
|
||||
telemetry.gauge("db.client.connections.waiting", lambda: [(self.waiting, {})], "{request}", "Запросы, ожидающие соединение")
|
||||
|
||||
def _usage(self):
|
||||
if self.pool is None:
|
||||
return []
|
||||
size, idle = self.pool.get_size(), self.pool.get_idle_size()
|
||||
return [(size - idle, {"state": "used"}), (idle, {"state": "idle"})]
|
||||
|
||||
async def start(self):
|
||||
if not config.DATABASE_URL:
|
||||
log("error", "database_url_missing")
|
||||
return
|
||||
# min_size=0: приложение стартует и без БД (readiness покажет 503), пул подключается лениво.
|
||||
self.pool = await asyncpg.create_pool(
|
||||
dsn=config.DATABASE_URL, min_size=0, max_size=config.DB_POOL_MAX,
|
||||
command_timeout=config.DB_COMMAND_TIMEOUT, timeout=config.DB_CONNECT_TIMEOUT,
|
||||
max_inactive_connection_lifetime=60,
|
||||
server_settings={"application_name": "messenger-api", "timezone": "UTC"})
|
||||
|
||||
async def stop(self):
|
||||
if self.pool:
|
||||
try:
|
||||
await asyncio.wait_for(self.pool.close(), 3)
|
||||
except Exception:
|
||||
self.pool.terminate()
|
||||
|
||||
@asynccontextmanager
|
||||
async def conn(self):
|
||||
if self.pool is None:
|
||||
raise unavailable("База данных не настроена")
|
||||
try:
|
||||
self.waiting += 1
|
||||
try:
|
||||
acquired = await self.pool.acquire(timeout=config.DB_ACQUIRE_TIMEOUT)
|
||||
finally:
|
||||
self.waiting -= 1
|
||||
try:
|
||||
with telemetry.child_span("db.connection", "client", {"db.system": "postgresql"}):
|
||||
yield acquired
|
||||
finally:
|
||||
await self.pool.release(acquired)
|
||||
except DEPENDENCY_ERRORS as exc:
|
||||
log("warning", "db_unavailable", error_type=type(exc).__name__)
|
||||
raise unavailable() from exc
|
||||
|
||||
async def _run(self, method, sql, args, retry):
|
||||
for attempt in (0, 1):
|
||||
try:
|
||||
async with self.conn() as c:
|
||||
return await getattr(c, method)(sql, *args)
|
||||
except Exception as exc:
|
||||
if attempt == 0 and retry and isinstance(exc.__cause__, _RETRYABLE):
|
||||
continue # разорванное соединение пула (рестарт БД): один повтор только для чтения
|
||||
raise
|
||||
|
||||
async def fetch(self, sql, *args):
|
||||
return await self._run("fetch", sql, args, True)
|
||||
|
||||
async def fetchrow(self, sql, *args, retry=True):
|
||||
return await self._run("fetchrow", sql, args, retry)
|
||||
|
||||
async def fetchval(self, sql, *args):
|
||||
return await self._run("fetchval", sql, args, True)
|
||||
|
||||
async def execute(self, sql, *args):
|
||||
return await self._run("execute", sql, args, False)
|
||||
|
||||
async def ping(self):
|
||||
return await asyncio.wait_for(self.fetchval("SELECT 1"), 1.0)
|
||||
@@ -0,0 +1,102 @@
|
||||
"""Проверка загружаемых изображений ДО полного декодирования: сигнатура, формат Pillow, размеры, целостность.
|
||||
|
||||
PNG/JPEG определяются по содержимому (а не по Content-Type). Лимит пикселей проверяется по заголовку,
|
||||
поэтому «decompression bomb» отклоняется без декодирования. Целостность: PNG — verify() (CRC всех чанков, без
|
||||
распаковки в память), JPEG — декодирование в режиме draft (1/8 размера => около 1/64 памяти).
|
||||
"""
|
||||
import struct
|
||||
import zlib
|
||||
|
||||
from PIL import Image, UnidentifiedImageError
|
||||
|
||||
from . import config
|
||||
|
||||
Image.MAX_IMAGE_PIXELS = None # лимит применяем сами по заголовку (ниже), без предупреждений Pillow
|
||||
PNG_SIGNATURE = b"\x89PNG\r\n\x1a\n"
|
||||
|
||||
|
||||
class ImageRejected(Exception):
|
||||
def __init__(self, status, code, message):
|
||||
super().__init__(message)
|
||||
self.status, self.code, self.message = status, code, message
|
||||
|
||||
|
||||
def sniff(head):
|
||||
if head.startswith(PNG_SIGNATURE):
|
||||
return "image/png"
|
||||
if head[:3] == b"\xff\xd8\xff":
|
||||
return "image/jpeg"
|
||||
return None
|
||||
|
||||
|
||||
def _check_png_stream(fileobj, width, height):
|
||||
"""Целостность потока IDAT без аллокации изображения: распаковка порциями, проверка конца потока и размера данных."""
|
||||
fileobj.seek(8)
|
||||
inflater = zlib.decompressobj()
|
||||
total, channels, bit_depth, interlaced = 0, 0, 0, 0
|
||||
while True:
|
||||
head = fileobj.read(8)
|
||||
if len(head) < 8:
|
||||
raise ValueError("truncated")
|
||||
length, kind = struct.unpack(">I4s", head)
|
||||
if length > 0x7FFFFFFF:
|
||||
raise ValueError("chunk")
|
||||
if kind == b"IHDR":
|
||||
data = fileobj.read(length)
|
||||
bit_depth, color_type, _, _, interlaced = struct.unpack(">BBBBB", data[8:13])
|
||||
channels = {0: 1, 2: 3, 3: 1, 4: 2, 6: 4}[color_type]
|
||||
fileobj.read(4)
|
||||
elif kind == b"IDAT":
|
||||
remaining = length
|
||||
while remaining:
|
||||
piece = fileobj.read(min(remaining, 65536))
|
||||
if not piece:
|
||||
raise ValueError("truncated")
|
||||
remaining -= len(piece)
|
||||
pending = piece
|
||||
while pending:
|
||||
total += len(inflater.decompress(pending, 1 << 20))
|
||||
pending = inflater.unconsumed_tail
|
||||
fileobj.read(4)
|
||||
elif kind == b"IEND":
|
||||
break
|
||||
else:
|
||||
fileobj.seek(length + 4, 1)
|
||||
if not inflater.eof:
|
||||
raise ValueError("zlib stream incomplete")
|
||||
if not interlaced:
|
||||
row = (width * channels * bit_depth + 7) // 8
|
||||
if total != height * (row + 1):
|
||||
raise ValueError("unexpected image data size")
|
||||
|
||||
|
||||
def validate(fileobj, declared_type):
|
||||
"""Возвращает (content_type, width, height) либо бросает ImageRejected (415 — не то/повреждено, 413 — много пикселей)."""
|
||||
fileobj.seek(0)
|
||||
detected = sniff(fileobj.read(16))
|
||||
if detected is None or detected != declared_type:
|
||||
raise ImageRejected(415, "unsupported_media_type", "Содержимое не соответствует PNG/JPEG или заявленному Content-Type")
|
||||
fileobj.seek(0)
|
||||
try:
|
||||
with Image.open(fileobj) as img:
|
||||
width, height = img.size
|
||||
fmt = img.format
|
||||
except (UnidentifiedImageError, OSError, ValueError, SyntaxError, Image.DecompressionBombError):
|
||||
raise ImageRejected(415, "unsupported_media_type", "Изображение повреждено или не поддерживается") from None
|
||||
if fmt not in ("PNG", "JPEG") or width < 1 or height < 1:
|
||||
raise ImageRejected(415, "unsupported_media_type", "Неподдерживаемый формат изображения")
|
||||
if width * height > config.MAX_IMAGE_PIXELS:
|
||||
raise ImageRejected(413, "payload_too_large", "Слишком много пикселей в изображении")
|
||||
fileobj.seek(0)
|
||||
try:
|
||||
img = Image.open(fileobj)
|
||||
if fmt == "PNG":
|
||||
img.verify()
|
||||
_check_png_stream(fileobj, width, height)
|
||||
else:
|
||||
img.draft("RGB", (max(1, width // 8), max(1, height // 8)))
|
||||
img.load()
|
||||
img.close()
|
||||
except Exception: # любая ошибка декодера = повреждённый файл
|
||||
raise ImageRejected(415, "unsupported_media_type", "Изображение повреждено") from None
|
||||
return detected, width, height
|
||||
@@ -0,0 +1,464 @@
|
||||
"""ASGI-приложение мессенджера (lab2): REST API, PostgreSQL, сессии Bearer, кэш Valkey."""
|
||||
import asyncio
|
||||
import contextlib
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
import base64
|
||||
import binascii
|
||||
import hashlib
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import time
|
||||
import uuid
|
||||
|
||||
import asyncpg
|
||||
|
||||
from . import config, passwords, telemetry
|
||||
from .attachments import AttachmentMixin, StreamResponse
|
||||
from .db import DEPENDENCY_ERRORS, Database
|
||||
from .broker import Broker
|
||||
from .outbox import Relay
|
||||
from .s3 import S3
|
||||
from .sessions import SessionCache, Sessions
|
||||
from .web import (ApiError, Request, bad, dt_to_us, encode_cursor, format_dt, log, new_request_id, page_params,
|
||||
parse_json_object, parse_uuid, respond, unavailable, us_to_dt, valid_text, USERNAME_RE)
|
||||
|
||||
ROUTES = [
|
||||
("health_live", re.compile(r"/health/live"), {"GET"}),
|
||||
("health_ready", re.compile(r"/health/ready"), {"GET"}),
|
||||
("users", re.compile(r"/api/v1/users"), {"POST"}),
|
||||
("users_me", re.compile(r"/api/v1/users/me"), {"GET"}),
|
||||
("create_session", re.compile(r"/api/v1/auth/sessions"), {"POST"}),
|
||||
("session_current", re.compile(r"/api/v1/auth/sessions/current"), {"DELETE"}),
|
||||
("chats", re.compile(r"/api/v1/chats"), {"GET", "POST"}),
|
||||
("chat", re.compile(r"/api/v1/chats/(?P<chat_id>[^/]+)"), {"GET"}),
|
||||
("messages", re.compile(r"/api/v1/chats/(?P<chat_id>[^/]+)/messages"), {"GET", "POST"}),
|
||||
("upload_attachment", re.compile(r"/api/v1/chats/(?P<chat_id>[^/]+)/attachments"), {"POST"}),
|
||||
("get_attachment", re.compile(r"/api/v1/attachments/(?P<attachment_id>[^/]+)"), {"GET"}),
|
||||
("download_attachment", re.compile(r"/api/v1/attachments/(?P<attachment_id>[^/]+)/content"), {"GET"}),
|
||||
]
|
||||
ROUTE_LABELS = {
|
||||
"health_live": "/health/live", "health_ready": "/health/ready", "users": "/api/v1/users",
|
||||
"users_me": "/api/v1/users/me", "create_session": "/api/v1/auth/sessions",
|
||||
"session_current": "/api/v1/auth/sessions/current", "chats": "/api/v1/chats",
|
||||
"chat": "/api/v1/chats/{chat_id}", "messages": "/api/v1/chats/{chat_id}/messages",
|
||||
"upload_attachment": "/api/v1/chats/{chat_id}/attachments", "get_attachment": "/api/v1/attachments/{attachment_id}",
|
||||
"download_attachment": "/api/v1/attachments/{attachment_id}/content",
|
||||
}
|
||||
TOKEN_RE = re.compile(r"[A-Za-z0-9_-]{32,128}")
|
||||
IDEMPOTENCY_KEY_RE = re.compile(r"[A-Za-z0-9._:-]{1,128}")
|
||||
BEARER = 'Bearer realm="messenger"'
|
||||
BASIC = 'Basic realm="messenger", charset="UTF-8"'
|
||||
# Идентификатор реплики: имя хоста контейнера (12 hex, == короткий ID контейнера) -> легко сопоставить с docker ps.
|
||||
INSTANCE_ID = os.environ.get("INSTANCE_ID") or (os.environ.get("HOSTNAME") or uuid.uuid4().hex)[:12]
|
||||
|
||||
MSG_ATTS = ("COALESCE((SELECT array_agg(ma.attachment_id ORDER BY ma.position) FROM message_attachments ma "
|
||||
"WHERE ma.message_id = messages.id), '{}'::uuid[]) AS attachment_ids")
|
||||
CHAT_SELECT = ("SELECT c.id, c.title, c.created_at, "
|
||||
"(SELECT array_agg(m2.user_id ORDER BY m2.ord) FROM chat_members m2 WHERE m2.chat_id = c.id) AS member_ids ")
|
||||
|
||||
|
||||
def user_json(row):
|
||||
return {"id": str(row["id"]), "username": row["username"], "display_name": row["display_name"],
|
||||
"created_at": format_dt(row["created_at"])}
|
||||
|
||||
|
||||
def chat_json(row):
|
||||
return {"id": str(row["id"]), "title": row["title"], "member_ids": [str(x) for x in row["member_ids"]],
|
||||
"created_at": format_dt(row["created_at"])}
|
||||
|
||||
|
||||
def message_json(row):
|
||||
return {"id": str(row["id"]), "chat_id": str(row["chat_id"]), "sender_id": str(row["sender_id"]),
|
||||
"text": row["text"], "created_at": format_dt(row["created_at"]),
|
||||
"attachment_ids": [str(x) for x in (row["attachment_ids"] or [])]}
|
||||
|
||||
|
||||
def unauthorized(challenge=BEARER, message="Требуется действительная сессия"):
|
||||
return ApiError(401, "unauthorized", message, {"www-authenticate": challenge})
|
||||
|
||||
|
||||
class Messenger(AttachmentMixin):
|
||||
def __init__(self):
|
||||
self.db = Database()
|
||||
self.s3 = S3()
|
||||
self.broker = Broker()
|
||||
self.relay = Relay(self.db, self.broker)
|
||||
self.image_pool = ThreadPoolExecutor(max_workers=config.IMAGE_THREADS, thread_name_prefix="image")
|
||||
self._init_attachments()
|
||||
self.cache = SessionCache()
|
||||
self.sessions = Sessions(self.db, self.cache)
|
||||
self.inflight = 0
|
||||
self.draining = False
|
||||
# Метрики очереди считает API (любая реплика), а не worker: при остановке worker алерт по возрасту задач продолжает работать.
|
||||
self.queue_stats = (0, 0.0)
|
||||
telemetry.gauge("messenger.queue.depth", lambda: [(self.queue_stats[0], {"queue.name": "attachment.process"})], "{job}", "queued + processing")
|
||||
telemetry.gauge("messenger.queue.oldest_job_age", lambda: [(self.queue_stats[1], {"queue.name": "attachment.process"})], "s", "Возраст старейшего задания")
|
||||
|
||||
# --- аутентификация ------------------------------------------------------
|
||||
@staticmethod
|
||||
def bearer_token(req):
|
||||
header = req.headers.get("authorization", "")
|
||||
scheme, _, token = header.partition(" ")
|
||||
if scheme.lower() != "bearer" or not TOKEN_RE.fullmatch(token.strip()):
|
||||
raise unauthorized()
|
||||
return token.strip()
|
||||
|
||||
async def authenticate(self, req):
|
||||
token = self.bearer_token(req)
|
||||
user_id = await self.sessions.resolve(token)
|
||||
if user_id is None:
|
||||
raise unauthorized()
|
||||
req.user_id, req.token = user_id, token
|
||||
return user_id
|
||||
|
||||
# --- обработчики -----------------------------------------------------------
|
||||
async def health_live(self, req, params):
|
||||
return 200, {"status": "ok"}
|
||||
|
||||
async def health_ready(self, req, params):
|
||||
# Критичная зависимость — PostgreSQL. Valkey — необязательный кэш: его отказ не останавливает API.
|
||||
if self.draining:
|
||||
raise unavailable("Сервис останавливается")
|
||||
try:
|
||||
await self.db.ping()
|
||||
except ApiError:
|
||||
raise
|
||||
except Exception:
|
||||
raise unavailable() from None
|
||||
return 200, {"status": "ok"}
|
||||
|
||||
async def users(self, req, params):
|
||||
req.require_json_media_type()
|
||||
data = parse_json_object(await req.body(), {"username", "display_name", "password"},
|
||||
{"username", "display_name", "password"})
|
||||
username = data["username"]
|
||||
if not isinstance(username, str) or not USERNAME_RE.fullmatch(username):
|
||||
raise bad("username должен соответствовать [a-z0-9_]{3,32}")
|
||||
display_name = valid_text(data["display_name"], "display_name", 100)
|
||||
password = valid_text(data["password"], "password", 128, allow_empty=True)
|
||||
if len(password) < 12:
|
||||
raise bad("password должен содержать от 12 до 128 символов")
|
||||
pw_hash = await passwords.hash_password(password)
|
||||
row = await self.db.fetchrow(
|
||||
"INSERT INTO users (id, username, display_name, password_hash) VALUES ($1, $2, $3, $4) "
|
||||
"ON CONFLICT (username) DO NOTHING RETURNING id, username, display_name, created_at",
|
||||
uuid.uuid4(), username, display_name, pw_hash)
|
||||
if row is None:
|
||||
raise ApiError(409, "conflict", "username уже занят")
|
||||
return 201, user_json(row)
|
||||
|
||||
async def users_me(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
row = await self.db.fetchrow("SELECT id, username, display_name, created_at FROM users WHERE id = $1", user_id)
|
||||
if row is None:
|
||||
raise unauthorized()
|
||||
return 200, user_json(row)
|
||||
|
||||
async def create_session(self, req, params):
|
||||
header = req.headers.get("authorization", "")
|
||||
scheme, _, value = header.partition(" ")
|
||||
if scheme.lower() != "basic" or not value.strip():
|
||||
raise unauthorized(BASIC, "Неверные учётные данные")
|
||||
try:
|
||||
decoded = base64.b64decode(value.strip(), validate=True).decode("utf-8")
|
||||
username, sep, password = decoded.partition(":")
|
||||
if not sep:
|
||||
raise ValueError
|
||||
except (binascii.Error, ValueError, UnicodeDecodeError):
|
||||
raise unauthorized(BASIC, "Неверные учётные данные") from None
|
||||
row = None
|
||||
if USERNAME_RE.fullmatch(username):
|
||||
row = await self.db.fetchrow("SELECT id, username, display_name, created_at, password_hash FROM users "
|
||||
"WHERE username = $1", username)
|
||||
ok = await passwords.verify_password(row["password_hash"] if row else None, password)
|
||||
if not ok:
|
||||
raise unauthorized(BASIC, "Неверные учётные данные")
|
||||
token, expires_at = await self.sessions.create(row["id"])
|
||||
return 201, {"token": token, "token_type": "Bearer", "expires_at": format_dt(expires_at), "user": user_json(row)}
|
||||
|
||||
async def session_current(self, req, params):
|
||||
await self.authenticate(req)
|
||||
await self.sessions.revoke(req.token)
|
||||
return 204, None
|
||||
|
||||
async def chats(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
if req.method == "GET":
|
||||
limit, after = page_params(req.query, "c", user_id)
|
||||
if after is None:
|
||||
rows = await self.db.fetch(
|
||||
CHAT_SELECT + "FROM chat_members m JOIN chats c ON c.id = m.chat_id WHERE m.user_id = $1 "
|
||||
"ORDER BY m.chat_created_at, m.chat_id LIMIT $2", user_id, limit + 1)
|
||||
else:
|
||||
rows = await self.db.fetch(
|
||||
CHAT_SELECT + "FROM chat_members m JOIN chats c ON c.id = m.chat_id WHERE m.user_id = $1 "
|
||||
"AND (m.chat_created_at, m.chat_id) > ($2, $3) ORDER BY m.chat_created_at, m.chat_id LIMIT $4",
|
||||
user_id, us_to_dt(after[0]), uuid.UUID(after[1]), limit + 1)
|
||||
more = len(rows) > limit
|
||||
rows = rows[:limit]
|
||||
cursor = encode_cursor("c", user_id, (dt_to_us(rows[-1]["created_at"]), str(rows[-1]["id"]))) if more else None
|
||||
return 200, {"items": [chat_json(r) for r in rows], "next_cursor": cursor}
|
||||
req.require_json_media_type()
|
||||
data = parse_json_object(await req.body(), {"title", "member_ids"}, {"title", "member_ids"})
|
||||
title = valid_text(data["title"], "title", 100)
|
||||
members = data["member_ids"]
|
||||
if not isinstance(members, list) or not 1 <= len(members) <= 100:
|
||||
raise bad("member_ids должен содержать от 1 до 100 элементов")
|
||||
ids = [parse_uuid(m, "member_id") for m in members]
|
||||
if len(set(ids)) != len(ids):
|
||||
raise bad("member_ids содержит дубликаты")
|
||||
if user_id not in ids:
|
||||
raise bad("Создатель должен быть в member_ids")
|
||||
chat_id = uuid.uuid4()
|
||||
uuids = [uuid.UUID(i) for i in ids]
|
||||
try:
|
||||
async with self.db.conn() as c:
|
||||
async with c.transaction(): # чат и участники создаются атомарно
|
||||
# Существование участников проверяет внешний ключ chat_members.user_id: при неизвестном
|
||||
# участнике вставка падает и откатывается вся транзакция, включая уже созданный чат.
|
||||
created = await c.fetchval("INSERT INTO chats (id, title) VALUES ($1, $2) RETURNING created_at",
|
||||
chat_id, title)
|
||||
await c.execute(
|
||||
"INSERT INTO chat_members (chat_id, user_id, chat_created_at, ord) "
|
||||
"SELECT $1, u, $2, o FROM unnest($3::uuid[]) WITH ORDINALITY AS t(u, o)",
|
||||
chat_id, created, uuids)
|
||||
except asyncpg.ForeignKeyViolationError:
|
||||
raise bad("Неизвестный участник") from None
|
||||
return 201, {"id": str(chat_id), "title": title, "member_ids": ids, "created_at": format_dt(created)}
|
||||
|
||||
async def _require_member(self, user_id, params):
|
||||
chat_id = parse_uuid(params["chat_id"], "chat_id")
|
||||
ok = await self.db.fetchval("SELECT 1 FROM chat_members WHERE chat_id = $1 AND user_id = $2", chat_id, user_id)
|
||||
if not ok:
|
||||
raise ApiError(404, "not_found", "Чат не найден")
|
||||
return chat_id
|
||||
|
||||
async def chat(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
chat_id = parse_uuid(params["chat_id"], "chat_id")
|
||||
row = await self.db.fetchrow(
|
||||
CHAT_SELECT + "FROM chats c JOIN chat_members m ON m.chat_id = c.id AND m.user_id = $2 WHERE c.id = $1",
|
||||
chat_id, user_id)
|
||||
if row is None:
|
||||
raise ApiError(404, "not_found", "Чат не найден")
|
||||
return 200, chat_json(row)
|
||||
|
||||
async def messages(self, req, params):
|
||||
user_id = await self.authenticate(req)
|
||||
if req.method == "POST":
|
||||
row = await self._post_message(req, parse_uuid(params["chat_id"], "chat_id"), user_id)
|
||||
return 201, message_json(row)
|
||||
chat_id = await self._require_member(user_id, params)
|
||||
if req.method == "GET":
|
||||
limit, after = page_params(req.query, "m", chat_id)
|
||||
if after is None:
|
||||
rows = await self.db.fetch(
|
||||
"SELECT id, chat_id, sender_id, text, created_at, " + MSG_ATTS + " FROM messages WHERE chat_id = $1 "
|
||||
"ORDER BY created_at, id LIMIT $2", chat_id, limit + 1)
|
||||
else:
|
||||
rows = await self.db.fetch(
|
||||
"SELECT id, chat_id, sender_id, text, created_at, " + MSG_ATTS + " FROM messages WHERE chat_id = $1 "
|
||||
"AND (created_at, id) > ($2, $3) ORDER BY created_at, id LIMIT $4",
|
||||
chat_id, us_to_dt(after[0]), uuid.UUID(after[1]), limit + 1)
|
||||
more = len(rows) > limit
|
||||
rows = rows[:limit]
|
||||
cursor = encode_cursor("m", chat_id, (dt_to_us(rows[-1]["created_at"]), str(rows[-1]["id"]))) if more else None
|
||||
return 200, {"items": [message_json(r) for r in rows], "next_cursor": cursor}
|
||||
|
||||
async def _post_message(self, req, chat_id, user_id):
|
||||
# Сначала разбор запроса; ошибка валидации (400/415) возвращается только участнику чата — постороннему всегда 404.
|
||||
try:
|
||||
key = req.headers.get("idempotency-key", "")
|
||||
if not IDEMPOTENCY_KEY_RE.fullmatch(key):
|
||||
raise bad("Требуется заголовок Idempotency-Key ([A-Za-z0-9._:-]{1,128})")
|
||||
req.require_json_media_type()
|
||||
data = parse_json_object(await req.body(), {"text"}, {"text", "attachment_ids"})
|
||||
raw_atts = data.get("attachment_ids", [])
|
||||
if not isinstance(raw_atts, list) or len(raw_atts) > 10:
|
||||
raise bad("attachment_ids: массив до 10 элементов")
|
||||
atts = [parse_uuid(a, "attachment_id") for a in raw_atts]
|
||||
if len(set(atts)) != len(atts):
|
||||
raise bad("attachment_ids содержит дубликаты")
|
||||
text = valid_text(data["text"], "text", 4000, allow_empty=bool(atts))
|
||||
except ApiError:
|
||||
await self._require_member(user_id, {"chat_id": chat_id})
|
||||
raise
|
||||
# Нормализованное представление запроса: порядок ключей JSON не важен, отсутствие attachment_ids == [].
|
||||
request_hash = hashlib.sha256(json.dumps({"text": text, "attachment_ids": atts}, sort_keys=True,
|
||||
ensure_ascii=False).encode("utf-8")).digest()
|
||||
row = await self.db.fetchrow("SELECT * FROM send_message($1, $2, $3, $4, $5, $6, $7)", uuid.UUID(chat_id),
|
||||
uuid.UUID(user_id), key, request_hash, text, float(config.IDEMPOTENCY_TTL_SECONDS),
|
||||
[uuid.UUID(a) for a in atts], retry=False)
|
||||
if row["outcome"] == "not_member":
|
||||
raise ApiError(404, "not_found", "Чат не найден")
|
||||
if row["outcome"] == "attachment_not_found":
|
||||
raise ApiError(404, "not_found", "Вложение не найдено")
|
||||
if row["outcome"] == "attachment_failed":
|
||||
raise ApiError(409, "conflict", "Вложение не удалось обработать")
|
||||
if bytes(row["request_hash"]) != request_hash:
|
||||
raise ApiError(409, "conflict", "Idempotency-Key уже использован с другим телом запроса")
|
||||
return {"id": row["msg_id"], "chat_id": row["msg_chat"], "sender_id": row["msg_sender"], "text": row["msg_text"],
|
||||
"created_at": row["msg_created"], "attachment_ids": row["msg_atts"]}
|
||||
|
||||
# --- ASGI ----------------------------------------------------------------------
|
||||
async def __call__(self, scope, receive, send):
|
||||
if scope["type"] == "lifespan":
|
||||
return await self.lifespan(receive, send)
|
||||
if scope["type"] != "http":
|
||||
return
|
||||
started = time.perf_counter()
|
||||
request_id = new_request_id()
|
||||
route_name, status, headers = "unknown", 500, {}
|
||||
raw_headers = {k.decode("latin-1").lower(): v.decode("latin-1") for k, v in scope["headers"]}
|
||||
carrier = {k: raw_headers[k] for k in ("traceparent", "tracestate") if k in raw_headers}
|
||||
is_health = scope.get("path", "").startswith("/health") # проверки LB раз в секунду: без спанов и access-log (метрики остаются)
|
||||
span_cm = (contextlib.nullcontext() if is_health else
|
||||
telemetry.span("HTTP %s" % scope.get("method"), "server", {"http.request.method": scope.get("method")}, carrier))
|
||||
sp = span_cm.__enter__()
|
||||
try:
|
||||
req = Request(scope, receive, request_id)
|
||||
handler, params, route_name = self.route(req.method, req.path)
|
||||
telemetry.set_attr(sp, "http.route", ROUTE_LABELS.get(route_name, "unmatched"))
|
||||
req.span = sp
|
||||
if not route_name.startswith("health") and self.inflight >= config.MAX_INFLIGHT:
|
||||
raise ApiError(429, "rate_limited", "Слишком много одновременных запросов",
|
||||
{"retry-after": str(config.RETRY_AFTER_SECONDS)})
|
||||
self.inflight += 1
|
||||
try:
|
||||
await asyncio.sleep(0)
|
||||
if not route_name.startswith("health") and (config.HTTP_FAULT_500 or config.HTTP_FAULT_SLOW):
|
||||
await self.inject_http_faults()
|
||||
status, body = await handler(req, params)
|
||||
finally:
|
||||
self.inflight -= 1
|
||||
except ApiError as exc:
|
||||
status, headers = exc.status, exc.headers
|
||||
body = {"error": {"code": exc.code, "message": exc.message, "request_id": request_id}}
|
||||
except DEPENDENCY_ERRORS as exc:
|
||||
log("warning", "dependency_error", request_id=request_id, error_type=type(exc).__name__)
|
||||
status, headers = 503, {"retry-after": str(config.RETRY_AFTER_SECONDS)}
|
||||
body = {"error": {"code": "unavailable", "message": "Зависимость временно недоступна", "request_id": request_id}}
|
||||
except Exception as exc:
|
||||
log("error", "unhandled_exception", request_id=request_id, error_type=type(exc).__name__)
|
||||
status = 500
|
||||
body = {"error": {"code": "internal_error", "message": "Внутренняя ошибка", "request_id": request_id}}
|
||||
route_label = ROUTE_LABELS.get(route_name, "unmatched")
|
||||
duration = time.perf_counter() - started
|
||||
telemetry.set_attr(sp, "http.response.status_code", status)
|
||||
telemetry.set_attr(sp, "http.request_id", request_id)
|
||||
if sp is not None:
|
||||
sp.update_name("HTTP %s %s" % (scope.get("method"), route_label))
|
||||
if status >= 500:
|
||||
telemetry.mark_error(sp, "HTTP %d" % status)
|
||||
telemetry.record_http(scope.get("method"), route_label, status, duration)
|
||||
if isinstance(body, StreamResponse):
|
||||
try:
|
||||
await body.send(send, request_id, INSTANCE_ID)
|
||||
except Exception as exc: # клиент оборвал загрузку/ошибка S3 посреди потока
|
||||
log("warning", "stream_aborted", request_id=request_id, error_type=type(exc).__name__)
|
||||
else:
|
||||
await respond(send, status, body, request_id, headers, INSTANCE_ID)
|
||||
if os.environ.get("ACCESS_LOG", "1") != "0" and (not is_health or status >= 500):
|
||||
log("info", "request", request_id=request_id, method=scope.get("method"),
|
||||
route=route_label, status=status, duration_ms=round(duration * 1000, 3))
|
||||
span_cm.__exit__(None, None, None)
|
||||
|
||||
async def janitor(self):
|
||||
"""Удаляет давно просроченные ключи идемпотентности и сессии (безопасно запускать на каждой реплике)."""
|
||||
import random
|
||||
await asyncio.sleep(random.uniform(5, 30))
|
||||
while True:
|
||||
try:
|
||||
await self.db.execute("DELETE FROM idempotency_keys WHERE ctid IN (SELECT ctid FROM idempotency_keys "
|
||||
"WHERE created_at < now() - make_interval(secs => $1) LIMIT 5000)",
|
||||
float(config.IDEMPOTENCY_TTL_SECONDS) + 3600)
|
||||
await self.db.execute("DELETE FROM sessions WHERE ctid IN (SELECT ctid FROM sessions "
|
||||
"WHERE expires_at < now() - interval '1 day' LIMIT 5000)")
|
||||
except Exception as exc:
|
||||
log("warning", "janitor_failed", error_type=type(exc).__name__)
|
||||
await asyncio.sleep(3600)
|
||||
|
||||
async def queue_stats_loop(self):
|
||||
while True:
|
||||
try:
|
||||
row = await self.db.fetchrow("SELECT count(*) AS n, coalesce(extract(epoch FROM now() - min(accepted_at)), 0) AS age "
|
||||
"FROM attachments WHERE status IN ('queued', 'processing')")
|
||||
self.queue_stats = (row["n"], float(row["age"]))
|
||||
except Exception:
|
||||
pass
|
||||
await asyncio.sleep(2)
|
||||
|
||||
@staticmethod
|
||||
async def inject_http_faults():
|
||||
import random
|
||||
if config.HTTP_FAULT_SLOW and random.random() < float(config.HTTP_FAULT_SLOW[0]):
|
||||
await asyncio.sleep(float(config.HTTP_FAULT_SLOW[1]))
|
||||
if config.HTTP_FAULT_500 and random.random() < float(config.HTTP_FAULT_500[0]):
|
||||
raise RuntimeError("injected fault")
|
||||
|
||||
def route(self, method, path):
|
||||
for name, pattern, methods in ROUTES:
|
||||
m = pattern.fullmatch(path)
|
||||
if m:
|
||||
if method not in methods:
|
||||
raise ApiError(405, "invalid_request", "Метод не поддерживается", {"allow": ", ".join(sorted(methods))})
|
||||
return getattr(self, name), m.groupdict(), name
|
||||
raise ApiError(404, "not_found", "Ресурс не найден")
|
||||
|
||||
async def lifespan(self, receive, send):
|
||||
while True:
|
||||
event = await receive()
|
||||
if event["type"] == "lifespan.startup":
|
||||
telemetry.setup(config.SERVICE_NAME)
|
||||
telemetry.process_metrics()
|
||||
await self.db.start()
|
||||
await self.cache.start()
|
||||
await self.s3.start()
|
||||
self._relay_task = asyncio.create_task(self.relay.run())
|
||||
self._queue_task = asyncio.create_task(self.queue_stats_loop())
|
||||
self._janitor = asyncio.create_task(self.janitor())
|
||||
log("info", "startup", env=config.APP_ENV, instance=INSTANCE_ID, revision=os.environ.get("APP_REVISION", "dev"), cache=bool(config.VALKEY_URL),
|
||||
session_ttl=config.SESSION_TTL_SECONDS, pool_max=config.DB_POOL_MAX)
|
||||
await send({"type": "lifespan.startup.complete"})
|
||||
elif event["type"] == "lifespan.shutdown":
|
||||
self.draining = True
|
||||
self._janitor.cancel()
|
||||
await self.cache.stop()
|
||||
telemetry.shutdown()
|
||||
self.relay.stop()
|
||||
self._relay_task.cancel()
|
||||
self._queue_task.cancel()
|
||||
await self.broker.stop()
|
||||
await self.s3.stop()
|
||||
await self.db.stop()
|
||||
log("info", "shutdown")
|
||||
await send({"type": "lifespan.shutdown.complete"})
|
||||
return
|
||||
|
||||
|
||||
async def overloaded_app(scope, receive, send):
|
||||
"""Ответ uvicorn при превышении --limit-concurrency в контрактном формате."""
|
||||
if scope["type"] != "http":
|
||||
return
|
||||
request_id = new_request_id()
|
||||
body = {"error": {"code": "unavailable", "message": "Сервер перегружен: слишком много соединений",
|
||||
"request_id": request_id}}
|
||||
await respond(send, 503, body, request_id, {"retry-after": str(config.RETRY_AFTER_SECONDS), "connection": "close"}, INSTANCE_ID)
|
||||
|
||||
|
||||
def _install_overload_response():
|
||||
for name in ("uvicorn.protocols.http.httptools_impl", "uvicorn.protocols.http.h11_impl"):
|
||||
try:
|
||||
module = __import__(name, fromlist=["service_unavailable"])
|
||||
except ImportError:
|
||||
continue
|
||||
module.service_unavailable = overloaded_app
|
||||
|
||||
|
||||
def create_app():
|
||||
_install_overload_response()
|
||||
return Messenger()
|
||||
|
||||
|
||||
app = create_app()
|
||||
@@ -0,0 +1,64 @@
|
||||
"""Применение версионируемых SQL-миграций: python -m app.migrate
|
||||
|
||||
Свойства:
|
||||
* повторный запуск ничего не делает (таблица schema_migrations);
|
||||
* одновременный запуск нескольких экземпляров безопасен (advisory lock);
|
||||
* каждая миграция — одна транзакция вместе с записью о применении;
|
||||
* ждёт готовности PostgreSQL до MIGRATE_WAIT_SECONDS.
|
||||
"""
|
||||
import asyncio
|
||||
import hashlib
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
|
||||
import asyncpg
|
||||
|
||||
from . import config
|
||||
from .web import log
|
||||
|
||||
LOCK_ID = 7352001
|
||||
|
||||
|
||||
async def connect(deadline):
|
||||
while True:
|
||||
try:
|
||||
return await asyncpg.connect(config.DATABASE_URL, timeout=3)
|
||||
except (OSError, asyncpg.PostgresError, asyncio.TimeoutError) as exc:
|
||||
if time.monotonic() > deadline:
|
||||
raise
|
||||
log("info", "waiting_for_database", error_type=type(exc).__name__)
|
||||
await asyncio.sleep(1)
|
||||
|
||||
|
||||
async def main():
|
||||
wait = float(os.environ.get("MIGRATE_WAIT_SECONDS", "60"))
|
||||
conn = await connect(time.monotonic() + wait)
|
||||
try:
|
||||
await conn.execute("SELECT pg_advisory_lock($1)", LOCK_ID)
|
||||
await conn.execute("CREATE TABLE IF NOT EXISTS schema_migrations ("
|
||||
"version text PRIMARY KEY, checksum text NOT NULL, applied_at timestamptz NOT NULL DEFAULT now())")
|
||||
applied = {r["version"]: r["checksum"] for r in await conn.fetch("SELECT version, checksum FROM schema_migrations")}
|
||||
directory = os.path.abspath(config.MIGRATIONS_DIR)
|
||||
files = sorted(f for f in os.listdir(directory) if f.endswith(".sql"))
|
||||
count = 0
|
||||
for name in files:
|
||||
sql = open(os.path.join(directory, name), encoding="utf-8").read()
|
||||
checksum = hashlib.sha256(sql.encode("utf-8")).hexdigest()
|
||||
if name in applied:
|
||||
if applied[name] != checksum:
|
||||
log("error", "migration_modified", version=name)
|
||||
sys.exit(2)
|
||||
continue
|
||||
async with conn.transaction():
|
||||
await conn.execute(sql)
|
||||
await conn.execute("INSERT INTO schema_migrations (version, checksum) VALUES ($1, $2)", name, checksum)
|
||||
count += 1
|
||||
log("info", "migration_applied", version=name)
|
||||
log("info", "migrations_done", applied=count, total=len(files))
|
||||
finally:
|
||||
await conn.close()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,82 @@
|
||||
"""Transactional outbox: relay публикует задания из PostgreSQL в брокер.
|
||||
|
||||
Гарантия: задание, зафиксированное в БД (COMMIT), будет опубликовано даже если процесс упал между COMMIT и publish.
|
||||
Relay запускается в каждой реплике API и в worker; строки берутся `FOR UPDATE SKIP LOCKED`, поэтому одну строку
|
||||
публикует один relay. Если процесс упадёт после подтверждения брокера, но до отметки published_at, задание будет
|
||||
опубликовано повторно — доставка at-least-once, дубликаты безвредны (обработка идемпотентна).
|
||||
"""
|
||||
import asyncio
|
||||
from datetime import timezone
|
||||
|
||||
from . import config, telemetry
|
||||
from .web import format_dt, log
|
||||
|
||||
JOB_TYPE = "attachment.process.v1"
|
||||
|
||||
|
||||
def job_message(row, carrier=None):
|
||||
msg = {"schema_version": 1, "job_id": str(row["job_id"]), "type": JOB_TYPE, "attachment_id": str(row["attachment_id"]),
|
||||
"pipeline_version": row["pipeline_version"], "attempt": row["attempt"],
|
||||
"created_at": format_dt(row["created_at"].astimezone(timezone.utc))}
|
||||
msg.update(carrier or {}) # traceparent / tracestate (W3C Trace Context) — поля схемы jobs.schema.json
|
||||
return msg
|
||||
|
||||
|
||||
class Relay:
|
||||
def __init__(self, db, broker):
|
||||
self.db, self.broker = db, broker
|
||||
self._kick = asyncio.Event()
|
||||
self._stop = False
|
||||
self.last_error = None
|
||||
|
||||
def kick(self):
|
||||
self._kick.set()
|
||||
|
||||
def stop(self):
|
||||
self._stop = True
|
||||
self._kick.set()
|
||||
|
||||
async def relay_once(self):
|
||||
published = 0
|
||||
async with self.db.conn() as c:
|
||||
async with c.transaction():
|
||||
rows = await c.fetch(
|
||||
"SELECT id, job_id, attachment_id, pipeline_version, attempt, created_at, traceparent, tracestate FROM outbox "
|
||||
"WHERE published_at IS NULL AND available_at <= now() ORDER BY id LIMIT $1 FOR UPDATE SKIP LOCKED",
|
||||
config.RELAY_BATCH)
|
||||
for row in rows:
|
||||
parent = {k: row[k] for k in ("traceparent", "tracestate") if row[k]}
|
||||
# PRODUCER-спан — дочерний к спану запроса upload (контекст восстановлен из outbox)
|
||||
with telemetry.span("attachment.process publish", "producer",
|
||||
{"messaging.system": "rabbitmq", "messaging.destination.name": "attachment.process",
|
||||
"messaging.operation": "publish", "outbox.attempt": row["attempt"]}, parent):
|
||||
carrier = telemetry.current_carrier() or parent
|
||||
await self.broker.publish(job_message(row, carrier), headers=carrier)
|
||||
await c.execute("UPDATE outbox SET published_at = now() WHERE id = $1", row["id"])
|
||||
published += 1
|
||||
return published
|
||||
|
||||
def headers_for(self, row):
|
||||
return {}
|
||||
|
||||
async def run(self):
|
||||
delay = config.RELAY_INTERVAL
|
||||
while not self._stop:
|
||||
try:
|
||||
n = await self.relay_once()
|
||||
self.last_error = None
|
||||
delay = config.RELAY_INTERVAL
|
||||
if n >= config.RELAY_BATCH:
|
||||
continue # возможно, есть ещё
|
||||
except asyncio.CancelledError:
|
||||
raise
|
||||
except Exception as exc:
|
||||
if self.last_error != type(exc).__name__:
|
||||
log("warning", "outbox_relay_failed", error_type=type(exc).__name__)
|
||||
self.last_error = type(exc).__name__
|
||||
delay = min(delay * 2, 10.0) # брокер/БД недоступны: пауза с ростом до 10 с, задания остаются в outbox
|
||||
try:
|
||||
await asyncio.wait_for(self._kick.wait(), delay)
|
||||
except asyncio.TimeoutError:
|
||||
pass
|
||||
self._kick.clear()
|
||||
@@ -0,0 +1,31 @@
|
||||
"""Хеширование паролей: Argon2id (библиотека argon2-cffi), вычисления вне event loop."""
|
||||
import asyncio
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
|
||||
from argon2 import PasswordHasher
|
||||
from argon2.exceptions import InvalidHashError, VerificationError, VerifyMismatchError
|
||||
|
||||
from . import config
|
||||
|
||||
_hasher = PasswordHasher(time_cost=config.ARGON2_TIME_COST, memory_cost=config.ARGON2_MEMORY_KIB,
|
||||
parallelism=config.ARGON2_PARALLELISM)
|
||||
_pool = ThreadPoolExecutor(max_workers=config.HASH_CONCURRENCY, thread_name_prefix="argon2")
|
||||
# Для неизвестного username проверяем фиктивный хеш: время ответа не выдаёт существование пользователя.
|
||||
_DUMMY = _hasher.hash("dummy-password-for-timing-equalisation")
|
||||
|
||||
|
||||
async def hash_password(password):
|
||||
return await asyncio.get_running_loop().run_in_executor(_pool, _hasher.hash, password)
|
||||
|
||||
|
||||
def _verify(stored, password):
|
||||
try:
|
||||
return _hasher.verify(stored, password)
|
||||
except (VerifyMismatchError, VerificationError, InvalidHashError):
|
||||
return False
|
||||
|
||||
|
||||
async def verify_password(stored, password):
|
||||
"""stored=None -> проверка фиктивного хеша, результат всегда False."""
|
||||
ok = await asyncio.get_running_loop().run_in_executor(_pool, _verify, stored or _DUMMY, password)
|
||||
return ok and stored is not None
|
||||
@@ -0,0 +1,97 @@
|
||||
"""Обработка изображений (pipeline_version=1): thumbnail, crop, optimized.
|
||||
|
||||
Правила (contracts/README.md):
|
||||
* thumbnail: JPEG, вписан в 256x256 с сохранением пропорций, БЕЗ увеличения, EXIF orientation применяется, alpha — на белый
|
||||
фон; размеры округляются вниз, минимум 1 px (320x200 -> 256x160); качество 85, метаданные не сохраняются;
|
||||
* crop: JPEG, центрированный квадрат 128x128 (маленькие исходники допускается увеличивать), качество 85;
|
||||
* optimized: WebP, вписан в 1280x1280 без увеличения, quality=80 method=4, EXIF/ICC не сохраняются (ориентация применена).
|
||||
Память: JPEG декодируется через draft (масштабирование при декодировании), число пикселей ограничено MAX_IMAGE_PIXELS.
|
||||
"""
|
||||
import hashlib
|
||||
import io
|
||||
import math
|
||||
|
||||
from PIL import Image, ImageOps
|
||||
|
||||
from . import config
|
||||
|
||||
Image.MAX_IMAGE_PIXELS = None
|
||||
PIPELINE_VERSION = 1
|
||||
JPEG_QUALITY = 85
|
||||
WEBP_QUALITY = 80
|
||||
|
||||
|
||||
class PermanentError(Exception):
|
||||
"""Постоянная ошибка входных данных: повтор не поможет (-> failed, error_code)."""
|
||||
|
||||
def __init__(self, code):
|
||||
super().__init__(code)
|
||||
self.code = code
|
||||
|
||||
|
||||
def _flatten(img):
|
||||
"""RGBA/LA/P с прозрачностью -> RGB на белом фоне."""
|
||||
if img.mode in ("RGBA", "LA") or (img.mode == "P" and "transparency" in img.info):
|
||||
rgba = img.convert("RGBA")
|
||||
background = Image.new("RGB", rgba.size, (255, 255, 255))
|
||||
background.paste(rgba, mask=rgba.getchannel("A"))
|
||||
return background
|
||||
return img.convert("RGB")
|
||||
|
||||
|
||||
def _fit_size(width, height, box, upscale=False):
|
||||
scale = min(box / width, box / height)
|
||||
if not upscale:
|
||||
scale = min(scale, 1.0)
|
||||
return max(1, math.floor(width * scale)), max(1, math.floor(height * scale))
|
||||
|
||||
|
||||
def _encode(img, fmt, **params):
|
||||
buf = io.BytesIO()
|
||||
img.save(buf, fmt, **params)
|
||||
data = buf.getvalue()
|
||||
return data, hashlib.sha256(data).hexdigest()
|
||||
|
||||
|
||||
def process(fileobj):
|
||||
"""Возвращает список вариантов: dict(name, content_type, data, sha256, width, height). Бросает PermanentError."""
|
||||
try:
|
||||
fileobj.seek(0)
|
||||
src = Image.open(fileobj)
|
||||
fmt = src.format
|
||||
width, height = src.size
|
||||
if fmt not in ("PNG", "JPEG"):
|
||||
raise PermanentError("invalid_image")
|
||||
if width * height > config.MAX_IMAGE_PIXELS:
|
||||
raise PermanentError("image_too_large")
|
||||
if fmt == "JPEG" and width > 2560 and height > 2560:
|
||||
# экономия памяти: JPEG декодируется сразу в масштабе 1/2 (результат всё равно вписывается в 1280x1280)
|
||||
src.draft("RGB", (width // 2, height // 2))
|
||||
src.load()
|
||||
img = ImageOps.exif_transpose(src)
|
||||
except PermanentError:
|
||||
raise
|
||||
except MemoryError:
|
||||
raise
|
||||
except Exception:
|
||||
raise PermanentError("invalid_image") from None
|
||||
base = _flatten(img)
|
||||
out = []
|
||||
# thumbnail
|
||||
tw, th = _fit_size(base.width, base.height, 256)
|
||||
thumb = base.resize((tw, th), Image.LANCZOS) if (tw, th) != base.size else base
|
||||
data, sha = _encode(thumb, "JPEG", quality=JPEG_QUALITY, optimize=True)
|
||||
out.append({"name": "thumbnail", "content_type": "image/jpeg", "data": data, "sha256": sha, "width": tw, "height": th})
|
||||
# crop 128x128 (центрированный квадрат)
|
||||
side = min(base.width, base.height)
|
||||
left, top = (base.width - side) // 2, (base.height - side) // 2
|
||||
crop = base.crop((left, top, left + side, top + side)).resize((128, 128), Image.LANCZOS)
|
||||
data, sha = _encode(crop, "JPEG", quality=JPEG_QUALITY, optimize=True)
|
||||
out.append({"name": "crop", "content_type": "image/jpeg", "data": data, "sha256": sha, "width": 128, "height": 128})
|
||||
# optimized WebP (прозрачность сохраняется)
|
||||
opt_src = img.convert("RGBA") if (img.mode in ("RGBA", "LA") or (img.mode == "P" and "transparency" in img.info)) else base
|
||||
ow, oh = _fit_size(opt_src.width, opt_src.height, 1280)
|
||||
opt = opt_src.resize((ow, oh), Image.LANCZOS) if (ow, oh) != opt_src.size else opt_src
|
||||
data, sha = _encode(opt, "WEBP", quality=WEBP_QUALITY, method=4)
|
||||
out.append({"name": "optimized", "content_type": "image/webp", "data": data, "sha256": sha, "width": ow, "height": oh})
|
||||
return out
|
||||
@@ -0,0 +1,91 @@
|
||||
"""Сверка S3 и PostgreSQL (reconcile): python -m app.reconcile [--apply] [--grace СЕКУНДЫ] [--prefix att/]
|
||||
|
||||
Находит и (с --apply) безопасно исправляет расхождения двух хранилищ:
|
||||
* orphan_object — объект в S3 без строки в attachments (сбой между записью в S3 и COMMIT метаданных, прерванная загрузка);
|
||||
удаляется ТОЛЬКО если старше grace (по умолчанию ORPHAN_GRACE_SECONDS=900 с): свежий объект может быть
|
||||
загрузкой, метаданные которой ещё не зафиксированы;
|
||||
* missing_object — строка attachments (ready) без объекта в S3 (потеря/удаление объекта): строка НЕ удаляется,
|
||||
статус -> failed, error_code=object_missing (сообщения со ссылкой остаются, скачивание даёт 409);
|
||||
* size_mismatch — размер объекта не совпадает с size_bytes: статус -> failed, error_code=size_mismatch.
|
||||
Режим по умолчанию — dry-run (ничего не меняет). Повторный запуск идемпотентен. Используемые (привязанные к сообщению)
|
||||
вложения и свежие объекты не удаляются никогда: удаляются только объекты, у которых нет строки в БД.
|
||||
Выход: JSON-отчёт со счётчиками и временем обхода.
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import time
|
||||
from datetime import datetime, timedelta, timezone
|
||||
|
||||
import asyncpg
|
||||
|
||||
from . import config
|
||||
from .s3 import S3
|
||||
|
||||
|
||||
async def reconcile(apply, grace, prefix):
|
||||
started = time.perf_counter()
|
||||
s3 = S3()
|
||||
await s3.start()
|
||||
conn = await asyncpg.connect(config.DATABASE_URL, timeout=5)
|
||||
try:
|
||||
t0 = time.perf_counter()
|
||||
rows = await conn.fetch("SELECT id, s3_key, size_bytes, status, created_at FROM attachments")
|
||||
db_ms = (time.perf_counter() - t0) * 1000
|
||||
by_key = {r["s3_key"]: r for r in rows}
|
||||
t0 = time.perf_counter()
|
||||
objects = {}
|
||||
async for key, size, modified in s3.list_keys(prefix):
|
||||
objects[key] = (size, modified)
|
||||
list_ms = (time.perf_counter() - t0) * 1000
|
||||
now = datetime.now(timezone.utc)
|
||||
cutoff = now - timedelta(seconds=grace)
|
||||
report = {"mode": "apply" if apply else "dry-run", "grace_seconds": grace, "db_rows": len(rows), "s3_objects": len(objects),
|
||||
"orphan_objects": [], "orphans_skipped_fresh": 0, "missing_objects": [], "size_mismatch": [], "fixed": {"deleted_orphans": 0, "marked_failed": 0}}
|
||||
for key, (size, modified) in objects.items():
|
||||
row = by_key.get(key)
|
||||
if row is None:
|
||||
if modified < cutoff:
|
||||
report["orphan_objects"].append(key)
|
||||
else:
|
||||
report["orphans_skipped_fresh"] += 1
|
||||
elif row["status"] == "ready" and row["size_bytes"] != size:
|
||||
report["size_mismatch"].append(str(row["id"]))
|
||||
for key, row in by_key.items():
|
||||
if key.startswith(prefix) and key not in objects and row["status"] == "ready" and row["created_at"] < cutoff:
|
||||
report["missing_objects"].append(str(row["id"]))
|
||||
if apply:
|
||||
for key in report["orphan_objects"]:
|
||||
# повторная проверка перед удалением: за время обхода строка могла появиться
|
||||
if await conn.fetchval("SELECT 1 FROM attachments WHERE s3_key = $1", key) is None:
|
||||
await s3.delete(key)
|
||||
report["fixed"]["deleted_orphans"] += 1
|
||||
for ids, code in ((report["missing_objects"], "object_missing"), (report["size_mismatch"], "size_mismatch")):
|
||||
for att_id in ids:
|
||||
n = await conn.execute("UPDATE attachments SET status = 'failed', error_code = $2, updated_at = now() "
|
||||
"WHERE id = $1 AND status = 'ready'", att_id, code)
|
||||
report["fixed"]["marked_failed"] += int(n.split()[-1])
|
||||
report["timing_ms"] = {"db_query": round(db_ms, 1), "s3_listing": round(list_ms, 1),
|
||||
"total": round((time.perf_counter() - started) * 1000, 1)}
|
||||
return report
|
||||
finally:
|
||||
await conn.close()
|
||||
await s3.stop()
|
||||
|
||||
|
||||
def main():
|
||||
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
|
||||
ap.add_argument("--apply", action="store_true", help="исправлять (по умолчанию dry-run)")
|
||||
ap.add_argument("--grace", type=int, default=config.ORPHAN_GRACE_SECONDS)
|
||||
ap.add_argument("--prefix", default="att/")
|
||||
ap.add_argument("--max-listed", type=int, default=20, help="сколько идентификаторов печатать в отчёте")
|
||||
args = ap.parse_args()
|
||||
report = asyncio.run(reconcile(args.apply, args.grace, args.prefix))
|
||||
for k in ("orphan_objects", "missing_objects", "size_mismatch"):
|
||||
report[k + "_count"] = len(report[k])
|
||||
report[k] = report[k][:args.max_listed]
|
||||
print(json.dumps(report, ensure_ascii=False, indent=2))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,70 @@
|
||||
"""Операторский replay и реестр неудавшихся заданий: python -m app.replay [list | replay <attachment_id> | replay-all]
|
||||
|
||||
list — реестр failed-вложений: id, error_code, число попыток, возраст, наличие staging;
|
||||
replay <attachment_id> — возвращает ТОЛЬКО failed-вложение в queued: новая попытка (attempt сбрасывается на 1), новый job_id,
|
||||
запись в job_events (аудит), задание в outbox. Если у вложения уже есть варианты (ready/успешный
|
||||
эффект), повторного исполнения не будет: вложение не в статусе failed -> replay отклоняется.
|
||||
Эффекты защищены ключом (attachment_id, pipeline_version, name): даже повторное выполнение не создаёт второй набор вариантов.
|
||||
"""
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import sys
|
||||
import uuid
|
||||
|
||||
import asyncpg
|
||||
|
||||
from . import config
|
||||
|
||||
|
||||
async def list_failed(conn):
|
||||
rows = await conn.fetch(
|
||||
"SELECT id, error_code, attempt, staging_key IS NOT NULL AS has_staging, updated_at FROM attachments "
|
||||
"WHERE status = 'failed' ORDER BY updated_at DESC LIMIT 200")
|
||||
return [{"attachment_id": str(r["id"]), "error_code": r["error_code"], "attempt": r["attempt"],
|
||||
"staging_present": r["has_staging"], "failed_at": r["updated_at"].isoformat()} for r in rows]
|
||||
|
||||
|
||||
async def replay(conn, att_id):
|
||||
async with conn.transaction():
|
||||
row = await conn.fetchrow("SELECT id, status, pipeline_version, staging_key FROM attachments WHERE id = $1 FOR UPDATE", att_id)
|
||||
if row is None:
|
||||
return {"attachment_id": str(att_id), "result": "not_found"}
|
||||
if row["status"] != "failed":
|
||||
return {"attachment_id": str(att_id), "result": "skipped", "reason": "status=%s (replay только для failed)" % row["status"]}
|
||||
if not row["staging_key"]:
|
||||
return {"attachment_id": str(att_id), "result": "skipped", "reason": "staging удалён по retention, исходник недоступен"}
|
||||
job_id = uuid.uuid4()
|
||||
await conn.execute("UPDATE attachments SET status = 'queued', error_code = NULL, attempt = 0, job_id = $2, lease_owner = NULL, "
|
||||
"lease_expires_at = NULL, updated_at = now() WHERE id = $1", att_id, job_id)
|
||||
await conn.execute("INSERT INTO outbox (job_id, attachment_id, pipeline_version, attempt) VALUES ($1, $2, $3, 1)",
|
||||
job_id, att_id, row["pipeline_version"])
|
||||
await conn.execute("INSERT INTO job_events (attachment_id, job_id, event, attempt, detail) VALUES ($1, $2, 'operator_replay', 1, 'failed -> queued')",
|
||||
att_id, job_id)
|
||||
return {"attachment_id": str(att_id), "result": "requeued", "job_id": str(job_id)}
|
||||
|
||||
|
||||
async def main_async(args):
|
||||
conn = await asyncpg.connect(config.DATABASE_URL, timeout=5)
|
||||
try:
|
||||
if args.command == "list":
|
||||
return await list_failed(conn)
|
||||
if args.command == "replay":
|
||||
return await replay(conn, uuid.UUID(args.attachment_id))
|
||||
return [await replay(conn, uuid.UUID(f["attachment_id"])) for f in await list_failed(conn)]
|
||||
finally:
|
||||
await conn.close()
|
||||
|
||||
|
||||
def main():
|
||||
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
|
||||
sub = ap.add_subparsers(dest="command", required=True)
|
||||
sub.add_parser("list")
|
||||
r = sub.add_parser("replay")
|
||||
r.add_argument("attachment_id")
|
||||
sub.add_parser("replay-all")
|
||||
print(json.dumps(asyncio.run(main_async(ap.parse_args())), ensure_ascii=False, indent=2))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(main())
|
||||
@@ -0,0 +1,92 @@
|
||||
"""Приватное S3-хранилище (MinIO) через aiobotocore: потоковая загрузка/скачивание, таймауты, перевод сбоев в 503."""
|
||||
import asyncio
|
||||
|
||||
from aiobotocore.config import AioConfig
|
||||
from aiobotocore.session import get_session
|
||||
from botocore.exceptions import BotoCoreError, ClientError
|
||||
|
||||
from . import config, telemetry
|
||||
from .web import log, unavailable
|
||||
|
||||
S3_ERRORS = (BotoCoreError, ClientError, asyncio.TimeoutError, OSError)
|
||||
|
||||
|
||||
class S3:
|
||||
def __init__(self):
|
||||
self._ctx = None
|
||||
self.client = None
|
||||
|
||||
async def start(self):
|
||||
if not config.S3_ENDPOINT:
|
||||
log("error", "s3_endpoint_missing")
|
||||
return
|
||||
session = get_session()
|
||||
self._ctx = session.create_client(
|
||||
"s3", endpoint_url=config.S3_ENDPOINT, region_name=config.S3_REGION,
|
||||
aws_access_key_id=config.S3_ACCESS_KEY, aws_secret_access_key=config.S3_SECRET_KEY,
|
||||
config=AioConfig(connect_timeout=config.S3_CONNECT_TIMEOUT, read_timeout=config.S3_READ_TIMEOUT,
|
||||
retries={"max_attempts": 2, "mode": "standard"}, max_pool_connections=config.S3_POOL,
|
||||
s3={"addressing_style": "path"}))
|
||||
self.client = await self._ctx.__aenter__()
|
||||
|
||||
async def stop(self):
|
||||
if self._ctx:
|
||||
await self._ctx.__aexit__(None, None, None)
|
||||
|
||||
def _check(self):
|
||||
if self.client is None:
|
||||
raise unavailable("Хранилище файлов не настроено")
|
||||
|
||||
async def put(self, key, fileobj, content_type, size):
|
||||
self._check()
|
||||
try:
|
||||
with telemetry.child_span("s3.put_object", "client", {"rpc.system": "aws-api", "rpc.method": "PutObject"}):
|
||||
await self.client.put_object(Bucket=config.S3_BUCKET, Key=key, Body=fileobj, ContentType=content_type,
|
||||
ContentLength=size)
|
||||
except S3_ERRORS as exc:
|
||||
log("warning", "s3_put_failed", error_type=type(exc).__name__)
|
||||
raise unavailable("Хранилище файлов недоступно") from exc
|
||||
|
||||
async def put_bytes(self, key, data, content_type):
|
||||
self._check()
|
||||
with telemetry.child_span("s3.put_object", "client", {"rpc.system": "aws-api", "rpc.method": "PutObject"}):
|
||||
await self.client.put_object(Bucket=config.S3_BUCKET, Key=key, Body=data, ContentType=content_type)
|
||||
|
||||
async def open_stream(self, key):
|
||||
"""Возвращает (stream, content_length). Вызывающий обязан закрыть stream."""
|
||||
self._check()
|
||||
try:
|
||||
with telemetry.child_span("s3.get_object", "client", {"rpc.system": "aws-api", "rpc.method": "GetObject"}):
|
||||
resp = await self.client.get_object(Bucket=config.S3_BUCKET, Key=key)
|
||||
except S3_ERRORS as exc:
|
||||
log("warning", "s3_get_failed", error_type=type(exc).__name__)
|
||||
raise unavailable("Хранилище файлов недоступно") from exc
|
||||
return resp["Body"], resp["ContentLength"]
|
||||
|
||||
async def delete(self, key):
|
||||
self._check()
|
||||
await self.client.delete_object(Bucket=config.S3_BUCKET, Key=key)
|
||||
|
||||
async def head(self, key):
|
||||
"""Метаданные объекта или None, если его нет."""
|
||||
self._check()
|
||||
try:
|
||||
return await self.client.head_object(Bucket=config.S3_BUCKET, Key=key)
|
||||
except ClientError as exc:
|
||||
if exc.response.get("Error", {}).get("Code") in ("404", "NoSuchKey", "NotFound"):
|
||||
return None
|
||||
raise unavailable("Хранилище файлов недоступно") from exc
|
||||
except S3_ERRORS as exc:
|
||||
raise unavailable("Хранилище файлов недоступно") from exc
|
||||
|
||||
async def ping(self):
|
||||
self._check()
|
||||
await asyncio.wait_for(self.client.head_bucket(Bucket=config.S3_BUCKET), 2.0)
|
||||
|
||||
async def list_keys(self, prefix):
|
||||
"""Асинхронный итератор (key, size, last_modified) по префиксу — для сверки (reconcile)."""
|
||||
self._check()
|
||||
paginator = self.client.get_paginator("list_objects_v2")
|
||||
async for page in paginator.paginate(Bucket=config.S3_BUCKET, Prefix=prefix):
|
||||
for obj in page.get("Contents", []):
|
||||
yield obj["Key"], obj["Size"], obj["LastModified"]
|
||||
@@ -0,0 +1,85 @@
|
||||
"""Идемпотентный bootstrap S3: python -m app.s3_bootstrap (one-shot сервис s3-init при каждом `docker compose up`).
|
||||
|
||||
1) bucket создаётся администратором, если его ещё нет (повтор ничего не меняет);
|
||||
2) проверяется, что анонимный запрос к bucket отклоняется (private);
|
||||
3) проверяется, что отдельный ключ API (права только на этот bucket) может писать/читать/удалять и НЕ может создавать bucket.
|
||||
Ключи читаются из файлов секретов; в вывод не попадают. Код выхода != 0, если любая проверка не пройдена.
|
||||
"""
|
||||
import asyncio
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
import urllib.error
|
||||
import urllib.request
|
||||
|
||||
from aiobotocore.config import AioConfig
|
||||
from aiobotocore.session import get_session
|
||||
from botocore.exceptions import ClientError
|
||||
|
||||
from . import config
|
||||
|
||||
|
||||
def secret(name):
|
||||
path = os.environ.get(name + "_FILE")
|
||||
with open(path, encoding="utf-8") as fh:
|
||||
return fh.read().strip()
|
||||
|
||||
|
||||
def client(session, access, secret_key):
|
||||
return session.create_client("s3", endpoint_url=config.S3_ENDPOINT, region_name=config.S3_REGION, aws_access_key_id=access,
|
||||
aws_secret_access_key=secret_key,
|
||||
config=AioConfig(connect_timeout=3, read_timeout=10, retries={"max_attempts": 1},
|
||||
s3={"addressing_style": "path"}))
|
||||
|
||||
|
||||
async def main():
|
||||
bucket = config.S3_BUCKET
|
||||
session = get_session()
|
||||
deadline = time.time() + 60
|
||||
async with client(session, secret("S3_ADMIN_ACCESS_KEY"), secret("S3_ADMIN_SECRET_KEY")) as admin:
|
||||
while True:
|
||||
try:
|
||||
await admin.list_buckets()
|
||||
break
|
||||
except Exception as exc:
|
||||
if time.time() > deadline:
|
||||
print("S3 недоступен:", type(exc).__name__)
|
||||
return 1
|
||||
await asyncio.sleep(1)
|
||||
try:
|
||||
await admin.head_bucket(Bucket=bucket)
|
||||
print("bucket %s уже существует" % bucket)
|
||||
except ClientError:
|
||||
await admin.create_bucket(Bucket=bucket)
|
||||
print("bucket %s создан" % bucket)
|
||||
# приватность: запрос без подписи отклоняется
|
||||
try:
|
||||
urllib.request.urlopen("%s/%s/" % (config.S3_ENDPOINT, bucket), timeout=5)
|
||||
print("ОШИБКА: bucket доступен анонимно")
|
||||
return 2
|
||||
except urllib.error.HTTPError as exc:
|
||||
if exc.code not in (401, 403):
|
||||
print("ОШИБКА: анонимный доступ вернул %d, ожидался 401/403" % exc.code)
|
||||
return 2
|
||||
print("анонимный доступ запрещён (HTTP %d)" % exc.code)
|
||||
# ключ API: рабочие права на bucket, нет прав администратора
|
||||
async with client(session, config.S3_ACCESS_KEY, config.S3_SECRET_KEY) as api:
|
||||
key = "bootstrap-check/%d" % int(time.time())
|
||||
await api.put_object(Bucket=bucket, Key=key, Body=b"ok")
|
||||
got = await (await api.get_object(Bucket=bucket, Key=key))["Body"].read()
|
||||
await api.delete_object(Bucket=bucket, Key=key)
|
||||
if got != b"ok":
|
||||
print("ОШИБКА: чтение/запись ключом API")
|
||||
return 3
|
||||
try:
|
||||
await api.create_bucket(Bucket="not-allowed-" + str(int(time.time())))
|
||||
print("ОШИБКА: ключ API смог создать bucket (избыточные права)")
|
||||
return 4
|
||||
except ClientError as exc:
|
||||
print("ключ API ограничен одним bucket (create_bucket отклонён: %s)" % exc.response["Error"]["Code"])
|
||||
print("bootstrap завершён")
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(asyncio.run(main()))
|
||||
@@ -0,0 +1,40 @@
|
||||
"""Точка входа: uvicorn с управляемым завершением (drain).
|
||||
|
||||
SIGTERM -> readiness сразу 503 -> DRAIN_SECONDS продолжаем обслуживать запросы, пока балансировщик
|
||||
исключает реплику -> uvicorn прекращает приём и дожидается текущих запросов (timeout_graceful_shutdown).
|
||||
Так плановая остановка/rolling update не даёт ошибок клиентам.
|
||||
"""
|
||||
import signal
|
||||
import threading
|
||||
|
||||
import uvicorn
|
||||
|
||||
from . import config
|
||||
from .main import app
|
||||
from .web import log
|
||||
|
||||
|
||||
class DrainingServer(uvicorn.Server):
|
||||
_draining = False
|
||||
|
||||
def handle_exit(self, sig, frame):
|
||||
if sig == signal.SIGTERM and not self._draining and config.DRAIN_SECONDS > 0:
|
||||
self._draining = True
|
||||
app.draining = True
|
||||
log("info", "draining", seconds=config.DRAIN_SECONDS)
|
||||
timer = threading.Timer(config.DRAIN_SECONDS, super().handle_exit, args=(sig, frame))
|
||||
timer.daemon = True
|
||||
timer.start()
|
||||
return
|
||||
super().handle_exit(sig, frame)
|
||||
|
||||
|
||||
def main():
|
||||
cfg = uvicorn.Config(app, host=config.APP_HOST, port=config.APP_PORT, workers=1, access_log=False,
|
||||
log_level="warning", timeout_graceful_shutdown=config.GRACEFUL_SHUTDOWN_SECONDS,
|
||||
timeout_keep_alive=config.KEEPALIVE_SECONDS, limit_concurrency=config.LIMIT_CONCURRENCY)
|
||||
DrainingServer(cfg).run()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,131 @@
|
||||
"""Сессии: PostgreSQL — источник истины; Valkey — кэш с TTL (read-through).
|
||||
|
||||
Политики:
|
||||
* токен — 256 бит из secrets; в БД и Valkey хранится только SHA-256 токена;
|
||||
* expiry: expires_at в БД (по часам БД). В кэше TTL = min(остаток жизни, SESSION_CACHE_TTL_SECONDS);
|
||||
* инвалидация: logout помечает revoked_at в БД и удаляет ключ в Valkey;
|
||||
* Valkey недоступен -> кэш обходится (чтение из БД), анонимный запрос никогда не считается доверенным;
|
||||
circuit breaker не даёт каждому запросу ждать таймаут;
|
||||
* окно устаревания: если Valkey был недоступен во время logout, ключ может прожить до SESSION_CACHE_TTL_SECONDS.
|
||||
"""
|
||||
import hashlib
|
||||
import secrets
|
||||
import time
|
||||
|
||||
from . import config
|
||||
from .web import log
|
||||
|
||||
try:
|
||||
import redis.asyncio as aioredis
|
||||
except ImportError: # кэш опционален
|
||||
aioredis = None
|
||||
|
||||
|
||||
def token_hash(token):
|
||||
return hashlib.sha256(token.encode("ascii")).digest()
|
||||
|
||||
|
||||
class SessionCache:
|
||||
def __init__(self):
|
||||
self.client = None
|
||||
self.down_until = 0.0
|
||||
self.stats = {"hit": 0, "miss": 0, "error": 0}
|
||||
|
||||
async def start(self):
|
||||
if config.VALKEY_URL and aioredis:
|
||||
self.client = aioredis.Redis.from_url(config.VALKEY_URL, socket_timeout=config.VALKEY_TIMEOUT,
|
||||
socket_connect_timeout=config.VALKEY_TIMEOUT)
|
||||
|
||||
async def stop(self):
|
||||
if self.client:
|
||||
await self.client.aclose()
|
||||
|
||||
def _usable(self):
|
||||
return self.client is not None and time.monotonic() >= self.down_until
|
||||
|
||||
def _fail(self, exc):
|
||||
self.stats["error"] += 1
|
||||
if time.monotonic() >= self.down_until:
|
||||
log("warning", "session_cache_unavailable", error_type=type(exc).__name__)
|
||||
self.down_until = time.monotonic() + config.VALKEY_BREAKER_SECONDS
|
||||
|
||||
async def get(self, key):
|
||||
if not self._usable():
|
||||
return None
|
||||
try:
|
||||
value = await self.client.get(b"s:" + key)
|
||||
except Exception as exc:
|
||||
self._fail(exc)
|
||||
return None
|
||||
self.stats["hit" if value is not None else "miss"] += 1
|
||||
return value
|
||||
|
||||
async def put(self, key, value, ttl):
|
||||
if not self._usable() or ttl <= 0:
|
||||
return
|
||||
try:
|
||||
await self.client.set(b"s:" + key, value, ex=max(1, int(ttl)))
|
||||
except Exception as exc:
|
||||
self._fail(exc)
|
||||
|
||||
async def delete(self, key):
|
||||
if not self._usable():
|
||||
return
|
||||
try:
|
||||
await self.client.delete(b"s:" + key)
|
||||
except Exception as exc:
|
||||
self._fail(exc)
|
||||
|
||||
async def ping(self):
|
||||
if self.client is None:
|
||||
return None
|
||||
try:
|
||||
return bool(await self.client.ping())
|
||||
except Exception:
|
||||
return False
|
||||
|
||||
|
||||
class Sessions:
|
||||
def __init__(self, db, cache):
|
||||
self.db, self.cache = db, cache
|
||||
|
||||
@staticmethod
|
||||
def _cache_value(user_id, expires_at):
|
||||
return ("%s|%d" % (user_id, int(expires_at.timestamp() * 1000))).encode()
|
||||
|
||||
async def create(self, user_id):
|
||||
token = secrets.token_urlsafe(32) # 256 бит энтропии
|
||||
digest = token_hash(token)
|
||||
row = await self.db.fetchrow(
|
||||
"INSERT INTO sessions (token_hash, user_id, expires_at) "
|
||||
"VALUES ($1, $2, now() + make_interval(secs => $3)) RETURNING expires_at",
|
||||
digest, user_id, float(config.SESSION_TTL_SECONDS))
|
||||
remaining = row["expires_at"].timestamp() - time.time()
|
||||
await self.cache.put(digest, self._cache_value(user_id, row["expires_at"]),
|
||||
min(remaining, config.SESSION_CACHE_TTL_SECONDS))
|
||||
return token, row["expires_at"]
|
||||
|
||||
async def resolve(self, token):
|
||||
"""token -> user_id (str) либо None, если сессии нет/истекла/отозвана."""
|
||||
digest = token_hash(token)
|
||||
cached = await self.cache.get(digest)
|
||||
if cached is not None:
|
||||
user_id, _, expires_ms = cached.decode().partition("|")
|
||||
if int(expires_ms) > time.time() * 1000:
|
||||
return user_id
|
||||
await self.cache.delete(digest)
|
||||
row = await self.db.fetchrow(
|
||||
"SELECT user_id, expires_at FROM sessions "
|
||||
"WHERE token_hash = $1 AND revoked_at IS NULL AND expires_at > now()", digest)
|
||||
if row is None:
|
||||
return None
|
||||
user_id = str(row["user_id"])
|
||||
remaining = row["expires_at"].timestamp() - time.time()
|
||||
await self.cache.put(digest, self._cache_value(user_id, row["expires_at"]),
|
||||
min(remaining, config.SESSION_CACHE_TTL_SECONDS))
|
||||
return user_id
|
||||
|
||||
async def revoke(self, token):
|
||||
digest = token_hash(token)
|
||||
await self.db.execute("UPDATE sessions SET revoked_at = now() WHERE token_hash = $1 AND revoked_at IS NULL", digest)
|
||||
await self.cache.delete(digest)
|
||||
@@ -0,0 +1,229 @@
|
||||
"""Наблюдаемость (lab6): OpenTelemetry SDK — трассы, метрики, логи по OTLP/HTTP в Collector.
|
||||
|
||||
Выключается переменной OTEL_ENABLED=0 (нужно для эксперимента «цена наблюдаемости»): тогда все функции — no-op и
|
||||
накладных расходов на экспорт нет. Секреты, тела сообщений, токены и изображения в телеметрию не попадают:
|
||||
в атрибутах только шаблон маршрута, метод, статус, тип задания и исход; ID объектов — только в логах/спанах
|
||||
(не в labels метрик).
|
||||
|
||||
Соответствие сигналов contracts/observability.md и имён метрик (после Prometheus-экспортёра Collector):
|
||||
http.server.request.duration (s) -> http_server_request_duration_seconds_{bucket,sum,count} (HTTP requests = _count)
|
||||
messenger.jobs.processed -> messenger_jobs_processed_total{job_type,outcome}
|
||||
messenger.job.duration (s) -> messenger_job_duration_seconds_*
|
||||
messenger.job.time_to_ready (s) -> messenger_job_time_to_ready_seconds_* (от приёма до ready)
|
||||
messenger.queue.depth / .oldest_age -> messenger_queue_depth{queue_name}, messenger_queue_oldest_job_age_seconds
|
||||
db.client.connections.usage/waiting -> db_client_connections_usage{state}, db_client_connections_waiting
|
||||
messenger.worker.busy / .capacity -> messenger_worker_busy, messenger_worker_capacity
|
||||
process.memory.rss / process.cpu.time -> process_memory_rss_bytes, process_cpu_time_seconds_total
|
||||
"""
|
||||
import contextlib
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import time
|
||||
|
||||
from . import config
|
||||
|
||||
ENABLED = config.OTEL_ENABLED
|
||||
_state = {"tracer": None, "meter": None, "logger": None, "propagator": None, "inst": {}}
|
||||
HTTP_BUCKETS = [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0, 2.5, 5.0, 10.0]
|
||||
JOB_BUCKETS = [0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10, 30, 60]
|
||||
TTR_BUCKETS = [0.1, 0.25, 0.5, 1, 2.5, 5, 10, 30, 60, 120, 300]
|
||||
|
||||
|
||||
def setup(service_name):
|
||||
"""Инициализирует провайдеры. Вызывается один раз при старте процесса (API или worker)."""
|
||||
if not ENABLED:
|
||||
return
|
||||
from opentelemetry import metrics, trace
|
||||
from opentelemetry.exporter.otlp.proto.http._log_exporter import OTLPLogExporter
|
||||
from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter
|
||||
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
|
||||
from opentelemetry.propagate import get_global_textmap
|
||||
from opentelemetry.sdk._logs import LoggerProvider, LoggingHandler
|
||||
from opentelemetry.sdk._logs.export import BatchLogRecordProcessor
|
||||
from opentelemetry.sdk.metrics import MeterProvider
|
||||
from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader
|
||||
from opentelemetry.sdk.metrics.view import ExplicitBucketHistogramAggregation, View
|
||||
from opentelemetry.sdk.resources import Resource
|
||||
from opentelemetry.sdk.trace import TracerProvider
|
||||
from opentelemetry.sdk.trace.export import BatchSpanProcessor
|
||||
from opentelemetry.sdk.trace.sampling import ParentBased, TraceIdRatioBased
|
||||
|
||||
endpoint = config.OTEL_ENDPOINT.rstrip("/")
|
||||
resource = Resource.create({"service.name": service_name, "service.version": config.SERVICE_VERSION,
|
||||
"service.instance.id": config.INSTANCE_ID, "deployment.environment.name": "lab"})
|
||||
# --- трассы: head-sampling по доле (ParentBased: решение родителя, в т.ч. из traceparent брокера, сохраняется)
|
||||
tp = TracerProvider(resource=resource, sampler=ParentBased(TraceIdRatioBased(config.OTEL_TRACE_RATIO)))
|
||||
tp.add_span_processor(BatchSpanProcessor(
|
||||
OTLPSpanExporter(endpoint=endpoint + "/v1/traces", timeout=3),
|
||||
max_queue_size=config.OTEL_QUEUE_SIZE, max_export_batch_size=config.OTEL_BATCH_SIZE, schedule_delay_millis=2000, export_timeout_millis=3000))
|
||||
trace.set_tracer_provider(tp)
|
||||
# --- метрики (histogram с явными границами, секунды)
|
||||
views = [View(instrument_name="http.server.request.duration", aggregation=ExplicitBucketHistogramAggregation(HTTP_BUCKETS)),
|
||||
View(instrument_name="messenger.job.duration", aggregation=ExplicitBucketHistogramAggregation(JOB_BUCKETS)),
|
||||
View(instrument_name="messenger.job.time_to_ready", aggregation=ExplicitBucketHistogramAggregation(TTR_BUCKETS))]
|
||||
reader = PeriodicExportingMetricReader(OTLPMetricExporter(endpoint=endpoint + "/v1/metrics", timeout=3),
|
||||
export_interval_millis=config.OTEL_METRIC_INTERVAL_MS, export_timeout_millis=3000)
|
||||
mp = MeterProvider(resource=resource, metric_readers=[reader], views=views)
|
||||
metrics.set_meter_provider(mp)
|
||||
# --- логи: стандартный logging -> OTLP (trace_id/span_id добавляются автоматически из активного спана)
|
||||
lp = LoggerProvider(resource=resource)
|
||||
lp.add_log_record_processor(BatchLogRecordProcessor(OTLPLogExporter(endpoint=endpoint + "/v1/logs", timeout=3),
|
||||
max_queue_size=config.OTEL_QUEUE_SIZE, schedule_delay_millis=2000))
|
||||
logger = logging.getLogger("messenger")
|
||||
logger.setLevel(logging.INFO)
|
||||
logger.propagate = False
|
||||
logger.addHandler(LoggingHandler(level=logging.INFO, logger_provider=lp))
|
||||
_state.update(tracer=trace.get_tracer("messenger"), meter=metrics.get_meter("messenger"), logger=logger,
|
||||
propagator=get_global_textmap(), providers=(tp, mp, lp))
|
||||
m = _state["meter"]
|
||||
_state["inst"] = {
|
||||
"http": m.create_histogram("http.server.request.duration", unit="s", description="Длительность HTTP-запроса"),
|
||||
"jobs": m.create_counter("messenger.jobs.processed", description="Обработанные задания по исходам"),
|
||||
"job_dur": m.create_histogram("messenger.job.duration", unit="s", description="Длительность обработки задания"),
|
||||
"ttr": m.create_histogram("messenger.job.time_to_ready", unit="s", description="От приёма загрузки до ready"),
|
||||
}
|
||||
for args in list(_pending_gauges):
|
||||
(observable_counter if len(args) == 5 else gauge)(*args[:4])
|
||||
_pending_gauges.clear()
|
||||
|
||||
|
||||
def shutdown():
|
||||
for p in _state.get("providers", ()):
|
||||
try:
|
||||
p.shutdown()
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
|
||||
# --- трассы ----------------------------------------------------------------------------------
|
||||
@contextlib.contextmanager
|
||||
def span(name, kind="internal", attributes=None, carrier=None):
|
||||
"""Контекстный менеджер спана. carrier — dict с traceparent/tracestate для продолжения чужого trace."""
|
||||
if not ENABLED or _state["tracer"] is None:
|
||||
yield None
|
||||
return
|
||||
from opentelemetry import trace
|
||||
from opentelemetry.trace import SpanKind, Status, StatusCode
|
||||
ctx = _state["propagator"].extract(carrier) if carrier else None
|
||||
kinds = {"server": SpanKind.SERVER, "client": SpanKind.CLIENT, "producer": SpanKind.PRODUCER,
|
||||
"consumer": SpanKind.CONSUMER, "internal": SpanKind.INTERNAL}
|
||||
with _state["tracer"].start_as_current_span(name, context=ctx, kind=kinds[kind], attributes=attributes or {},
|
||||
record_exception=False, set_status_on_exception=False) as sp:
|
||||
try:
|
||||
yield sp
|
||||
except Exception as exc:
|
||||
sp.set_status(Status(StatusCode.ERROR, type(exc).__name__))
|
||||
sp.set_attribute("error.type", type(exc).__name__)
|
||||
raise
|
||||
|
||||
|
||||
def child_span(name, kind="internal", attributes=None):
|
||||
"""Спан создаётся только внутри активного спана (запрос/задание); фоновые опросы БД не порождают корневых трасс."""
|
||||
if not ENABLED or _state["tracer"] is None:
|
||||
return contextlib.nullcontext()
|
||||
from opentelemetry import trace
|
||||
if not trace.get_current_span().get_span_context().is_valid:
|
||||
return contextlib.nullcontext()
|
||||
return span(name, kind, attributes)
|
||||
|
||||
|
||||
def set_attr(sp, key, value):
|
||||
if sp is not None:
|
||||
sp.set_attribute(key, value)
|
||||
|
||||
|
||||
def mark_error(sp, description):
|
||||
if sp is not None:
|
||||
from opentelemetry.trace import Status, StatusCode
|
||||
sp.set_status(Status(StatusCode.ERROR, description))
|
||||
|
||||
|
||||
def current_carrier():
|
||||
"""traceparent/tracestate текущего спана (для передачи через outbox/брокер)."""
|
||||
if not ENABLED or _state["propagator"] is None:
|
||||
return {}
|
||||
carrier = {}
|
||||
_state["propagator"].inject(carrier)
|
||||
return carrier
|
||||
|
||||
|
||||
def trace_ids():
|
||||
if not ENABLED:
|
||||
return None, None
|
||||
from opentelemetry import trace
|
||||
ctx = trace.get_current_span().get_span_context()
|
||||
if not ctx.is_valid:
|
||||
return None, None
|
||||
return "%032x" % ctx.trace_id, "%016x" % ctx.span_id
|
||||
|
||||
|
||||
# --- метрики ----------------------------------------------------------------------------------
|
||||
def record_http(method, route, status, seconds):
|
||||
if ENABLED and _state["inst"]:
|
||||
_state["inst"]["http"].record(seconds, {"http.request.method": method, "http.route": route,
|
||||
"http.response.status_code": status})
|
||||
|
||||
|
||||
def record_job(job_type, outcome, seconds, time_to_ready=None):
|
||||
if ENABLED and _state["inst"]:
|
||||
attrs = {"job.type": job_type, "outcome": outcome}
|
||||
_state["inst"]["jobs"].add(1, attrs)
|
||||
_state["inst"]["job_dur"].record(seconds, attrs)
|
||||
if time_to_ready is not None:
|
||||
_state["inst"]["ttr"].record(time_to_ready, {"job.type": job_type})
|
||||
|
||||
|
||||
_pending_gauges = []
|
||||
|
||||
|
||||
def gauge(name, callback, unit="1", description=""):
|
||||
"""Observable gauge: callback() -> [(value, {attrs})]. Регистрация откладывается до setup(), если провайдер ещё не создан."""
|
||||
if not ENABLED:
|
||||
return
|
||||
if _state["meter"] is None:
|
||||
_pending_gauges.append((name, callback, unit, description))
|
||||
return
|
||||
from opentelemetry.metrics import Observation
|
||||
|
||||
def cb(_options):
|
||||
try:
|
||||
return [Observation(v, a) for v, a in callback()]
|
||||
except Exception:
|
||||
return []
|
||||
_state["meter"].create_observable_gauge(name, callbacks=[cb], unit=unit, description=description)
|
||||
|
||||
|
||||
def observable_counter(name, callback, unit="1", description=""):
|
||||
"""Монотонный счётчик, значение которого читается при экспорте (например, накопленное CPU-время)."""
|
||||
if not ENABLED or _state["meter"] is None:
|
||||
_pending_gauges.append((name, callback, unit, description, "counter"))
|
||||
return
|
||||
from opentelemetry.metrics import Observation
|
||||
|
||||
def cb(_options):
|
||||
try:
|
||||
return [Observation(v, a) for v, a in callback()]
|
||||
except Exception:
|
||||
return []
|
||||
_state["meter"].create_observable_counter(name, callbacks=[cb], unit=unit, description=description)
|
||||
|
||||
|
||||
def process_metrics():
|
||||
def rss():
|
||||
try:
|
||||
with open("/proc/self/statm") as fh:
|
||||
return [(int(fh.read().split()[1]) * os.sysconf("SC_PAGE_SIZE"), {})]
|
||||
except (OSError, ValueError):
|
||||
return []
|
||||
gauge("process.memory.rss", rss, "By", "Резидентная память процесса")
|
||||
observable_counter("process.cpu.time", lambda: [(time.process_time(), {})], "s", "Накопленное CPU-время процесса")
|
||||
|
||||
|
||||
# --- логи ---------------------------------------------------------------------------------------
|
||||
def emit_log(level, record):
|
||||
"""Отправляет запись в OTLP (если включено). record — dict без секретов."""
|
||||
if not ENABLED or _state["logger"] is None:
|
||||
return
|
||||
lvl = {"info": logging.INFO, "warning": logging.WARNING, "error": logging.ERROR}.get(level, logging.INFO)
|
||||
_state["logger"].log(lvl, record.get("msg", ""), extra={k: v for k, v in record.items() if k not in ("msg", "level", "ts")})
|
||||
@@ -0,0 +1,192 @@
|
||||
"""Общий HTTP-слой: ошибки, запрос, валидация, курсоры, ответы, журнал."""
|
||||
import base64
|
||||
import json
|
||||
import re
|
||||
import sys
|
||||
import time
|
||||
import uuid
|
||||
from datetime import datetime, timedelta, timezone
|
||||
from urllib.parse import parse_qsl
|
||||
|
||||
from . import config, telemetry
|
||||
|
||||
UUID_RE = re.compile(r"[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}")
|
||||
USERNAME_RE = re.compile(r"[a-z0-9_]{3,32}")
|
||||
LIMIT_RE = re.compile(r"[0-9]{1,9}")
|
||||
EPOCH = datetime(1970, 1, 1, tzinfo=timezone.utc)
|
||||
|
||||
|
||||
class ApiError(Exception):
|
||||
def __init__(self, status, code, message, headers=None):
|
||||
super().__init__(message)
|
||||
self.status, self.code, self.message, self.headers = status, code, message, headers or {}
|
||||
|
||||
|
||||
def bad(message):
|
||||
return ApiError(400, "invalid_request", message)
|
||||
|
||||
|
||||
def unavailable(message="Зависимость временно недоступна"):
|
||||
return ApiError(503, "unavailable", message, {"retry-after": str(config.RETRY_AFTER_SECONDS)})
|
||||
|
||||
|
||||
def log(level, msg, **fields):
|
||||
"""Структурный лог: JSON в stdout + OTLP (если включено). trace_id/span_id — из активного спана; тела и токены не логируются."""
|
||||
record = {"ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()), "level": level, "service": config.SERVICE_NAME, "msg": msg}
|
||||
trace_id, span_id = telemetry.trace_ids()
|
||||
if trace_id:
|
||||
record["trace_id"], record["span_id"] = trace_id, span_id
|
||||
record.update(fields)
|
||||
sys.stdout.write(json.dumps(record, ensure_ascii=False) + "\n")
|
||||
sys.stdout.flush()
|
||||
telemetry.emit_log(level, record)
|
||||
|
||||
|
||||
# --- время -------------------------------------------------------------------
|
||||
def format_dt(value):
|
||||
value = value.astimezone(timezone.utc)
|
||||
return value.strftime("%Y-%m-%dT%H:%M:%S.") + "%06dZ" % value.microsecond
|
||||
|
||||
|
||||
def dt_to_us(value):
|
||||
return (value - EPOCH) // timedelta(microseconds=1)
|
||||
|
||||
|
||||
def us_to_dt(us):
|
||||
return EPOCH + timedelta(microseconds=us)
|
||||
|
||||
|
||||
# --- валидация ---------------------------------------------------------------
|
||||
def parse_uuid(value, what="id"):
|
||||
if not isinstance(value, str) or not UUID_RE.fullmatch(value):
|
||||
raise bad("Некорректный %s" % what)
|
||||
return value.lower()
|
||||
|
||||
|
||||
def valid_text(value, field, maximum, allow_empty=False):
|
||||
if not isinstance(value, str):
|
||||
raise bad("Поле %s должно быть строкой" % field)
|
||||
try:
|
||||
value.encode("utf-8")
|
||||
except UnicodeEncodeError:
|
||||
raise bad("Поле %s содержит некорректные символы" % field) from None
|
||||
if "\x00" in value:
|
||||
raise bad("Поле %s содержит NUL" % field)
|
||||
if len(value) > maximum:
|
||||
raise bad("Поле %s слишком длинное" % field)
|
||||
if not allow_empty and not value.strip():
|
||||
raise bad("Поле %s не может быть пустым" % field)
|
||||
return value
|
||||
|
||||
|
||||
def _reject_constant(name):
|
||||
raise ValueError(name)
|
||||
|
||||
|
||||
def parse_json_object(raw, required, allowed):
|
||||
try:
|
||||
data = json.loads(raw.decode("utf-8"), parse_constant=_reject_constant)
|
||||
except (UnicodeDecodeError, ValueError, RecursionError):
|
||||
raise bad("Тело запроса — некорректный JSON") from None
|
||||
if not isinstance(data, dict):
|
||||
raise bad("Ожидается JSON-объект")
|
||||
if set(data) - allowed:
|
||||
raise bad("Лишние поля в запросе")
|
||||
if any(key not in data for key in required):
|
||||
raise bad("Не хватает обязательных полей")
|
||||
return data
|
||||
|
||||
|
||||
def encode_cursor(kind, resource, key):
|
||||
payload = json.dumps({"k": kind, "r": resource, "t": key[0], "i": key[1]}, separators=(",", ":"))
|
||||
return base64.urlsafe_b64encode(payload.encode()).decode().rstrip("=")
|
||||
|
||||
|
||||
def decode_cursor(text, kind, resource):
|
||||
try:
|
||||
data = json.loads(base64.urlsafe_b64decode(text + "=" * (-len(text) % 4)).decode())
|
||||
if (not isinstance(data, dict) or data.get("k") != kind or data.get("r") != resource
|
||||
or not isinstance(data.get("t"), int) or isinstance(data.get("t"), bool)
|
||||
or not isinstance(data.get("i"), str) or not UUID_RE.fullmatch(data["i"])
|
||||
or not -62135596800000000 < data["t"] < 253402300799000000):
|
||||
raise ValueError
|
||||
return data["t"], data["i"]
|
||||
except (ValueError, UnicodeError):
|
||||
raise bad("Некорректный cursor") from None
|
||||
|
||||
|
||||
def page_params(query, kind, resource):
|
||||
limit, cursor = 50, None
|
||||
if "limit" in query:
|
||||
if len(query["limit"]) != 1 or not LIMIT_RE.fullmatch(query["limit"][0]):
|
||||
raise bad("limit должен быть целым числом 1–100")
|
||||
limit = int(query["limit"][0])
|
||||
if not 1 <= limit <= 100:
|
||||
raise bad("limit должен быть целым числом 1–100")
|
||||
if "cursor" in query:
|
||||
if len(query["cursor"]) != 1 or not query["cursor"][0]:
|
||||
raise bad("Некорректный cursor")
|
||||
cursor = decode_cursor(query["cursor"][0], kind, resource)
|
||||
return limit, cursor
|
||||
|
||||
|
||||
def parse_query(raw):
|
||||
result = {}
|
||||
if raw:
|
||||
for key, value in parse_qsl(raw.decode("latin-1"), keep_blank_values=True):
|
||||
result.setdefault(key, []).append(value)
|
||||
return result
|
||||
|
||||
|
||||
class Request:
|
||||
def __init__(self, scope, receive, request_id):
|
||||
self.method = scope["method"]
|
||||
self.path = scope["path"]
|
||||
self.query = parse_query(scope.get("query_string", b""))
|
||||
self.headers = {k.decode("latin-1").lower(): v.decode("latin-1") for k, v in scope["headers"]}
|
||||
self.receive = receive
|
||||
self.request_id = request_id
|
||||
self.user_id = None
|
||||
|
||||
async def body(self):
|
||||
declared = self.headers.get("content-length")
|
||||
if declared is not None:
|
||||
if not declared.isdigit():
|
||||
raise bad("Некорректный Content-Length")
|
||||
if int(declared) > config.MAX_BODY_BYTES:
|
||||
raise ApiError(413, "payload_too_large", "Тело запроса слишком большое", {"connection": "close"})
|
||||
chunks, size = [], 0
|
||||
while True:
|
||||
event = await self.receive()
|
||||
if event["type"] == "http.disconnect":
|
||||
raise bad("Соединение закрыто клиентом")
|
||||
chunk = event.get("body", b"")
|
||||
size += len(chunk)
|
||||
if size > config.MAX_BODY_BYTES:
|
||||
raise ApiError(413, "payload_too_large", "Тело запроса слишком большое", {"connection": "close"})
|
||||
chunks.append(chunk)
|
||||
if not event.get("more_body"):
|
||||
return b"".join(chunks)
|
||||
|
||||
def require_json_media_type(self):
|
||||
media = self.headers.get("content-type", "").split(";")[0].strip().lower()
|
||||
if media != "application/json":
|
||||
raise ApiError(415, "unsupported_media_type", "Ожидается Content-Type: application/json")
|
||||
|
||||
|
||||
async def respond(send, status, body, request_id, extra=None, instance_id=None):
|
||||
payload = b"" if status == 204 else json.dumps(body, ensure_ascii=False, separators=(",", ":")).encode("utf-8")
|
||||
headers = [(b"x-request-id", request_id.encode())]
|
||||
if status != 204:
|
||||
headers += [(b"content-type", b"application/json"), (b"content-length", str(len(payload)).encode())]
|
||||
else:
|
||||
headers.append((b"content-length", b"0"))
|
||||
if instance_id:
|
||||
headers.append((b"x-instance-id", instance_id.encode()))
|
||||
headers += [(k.encode(), v.encode()) for k, v in (extra or {}).items()]
|
||||
await send({"type": "http.response.start", "status": status, "headers": headers})
|
||||
await send({"type": "http.response.body", "body": payload})
|
||||
|
||||
|
||||
def new_request_id():
|
||||
return uuid.uuid4().hex
|
||||
@@ -0,0 +1,345 @@
|
||||
"""Worker вложений: python -m app.worker
|
||||
|
||||
Конвейер задания (jobs.schema.json, состояния — contracts/attachment-states.md):
|
||||
claim (queued -> processing, lease) -> чтение staging -> варианты (Pillow, в потоках) -> запись в конечный S3-prefix
|
||||
-> ОДНА транзакция: варианты + ready -> ack -> очистка staging.
|
||||
Ack отправляется только после устойчивого эффекта (COMMIT результата). Повторная доставка уже готовой задачи — no-op с ack.
|
||||
Временные ошибки: возврат в queued с backoff (экспоненциальный + jitter) через outbox; после MAX_ATTEMPTS — failed/retry_exhausted.
|
||||
Постоянные ошибки (невалидное изображение): failed с error_code сразу. Зависшие processing возвращает sweeper по истечении lease.
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import random
|
||||
import signal
|
||||
import tempfile
|
||||
import time
|
||||
import uuid
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
|
||||
import aio_pika
|
||||
|
||||
from . import config, pipeline, telemetry
|
||||
from .broker import QUEUE, Broker
|
||||
from .db import Database
|
||||
from .outbox import Relay
|
||||
from .s3 import S3
|
||||
from .web import log
|
||||
|
||||
WORKER_ID = "%s-%s" % ((os.environ.get("HOSTNAME") or "worker")[:12], uuid.uuid4().hex[:6])
|
||||
|
||||
|
||||
def backoff_seconds(attempt):
|
||||
"""Экспоненциальная задержка с полным jitter: случайное значение в [0, min(cap, base * 2^(attempt-1))]."""
|
||||
ceiling = min(config.RETRY_MAX_DELAY, config.RETRY_BASE_DELAY * (2 ** max(0, attempt - 1)))
|
||||
return random.uniform(ceiling * 0.5, ceiling)
|
||||
|
||||
|
||||
class Worker:
|
||||
def __init__(self):
|
||||
self.db, self.s3, self.broker = Database(), S3(), Broker()
|
||||
self.relay = Relay(self.db, self.broker)
|
||||
self.pool = ThreadPoolExecutor(max_workers=config.WORKER_CONCURRENCY, thread_name_prefix="img")
|
||||
self.busy = 0
|
||||
self.stopping = False
|
||||
self.tasks = []
|
||||
telemetry.gauge("messenger.worker.busy", lambda: [(self.busy, {})], "{job}", "Задания в обработке")
|
||||
telemetry.gauge("messenger.worker.capacity", lambda: [(config.WORKER_CONCURRENCY, {})], "{job}", "Ёмкость worker")
|
||||
|
||||
# --- claim / finalize ---------------------------------------------------------
|
||||
async def claim(self, job):
|
||||
return await self.db.fetchrow(
|
||||
"UPDATE attachments SET status = 'processing', lease_owner = $2, processing_started_at = now(), "
|
||||
"lease_expires_at = now() + make_interval(secs => $3), attempt = $4, job_id = $5, updated_at = now() "
|
||||
"WHERE id = $1 AND status = 'queued' AND pipeline_version <= $6 "
|
||||
"RETURNING id, chat_id, staging_key, s3_key, content_type, size_bytes, attempt",
|
||||
uuid.UUID(job["attachment_id"]), WORKER_ID, float(config.LEASE_SECONDS), job["attempt"],
|
||||
uuid.UUID(job["job_id"]), job["pipeline_version"], retry=False)
|
||||
|
||||
async def fail(self, att_id, job, code, event="failed"):
|
||||
await self.db.execute(
|
||||
"WITH u AS (UPDATE attachments SET status = 'failed', error_code = $2, lease_owner = NULL, lease_expires_at = NULL, "
|
||||
"updated_at = now() WHERE id = $1 AND status = 'processing' AND lease_owner = $3 RETURNING id) "
|
||||
"INSERT INTO job_events (attachment_id, job_id, event, attempt, detail) SELECT id, $4, $5, $6, $2 FROM u",
|
||||
att_id, code, WORKER_ID, uuid.UUID(job["job_id"]), event, job["attempt"])
|
||||
|
||||
async def reschedule(self, att_id, job, reason):
|
||||
"""Временная ошибка: processing -> queued, новая запись outbox с отложенной публикацией (backoff + jitter)."""
|
||||
nxt = job["attempt"] + 1
|
||||
if nxt > config.MAX_ATTEMPTS:
|
||||
log("warning", "job_retry_exhausted", attachment_id=str(att_id), attempts=job["attempt"])
|
||||
await self.fail(att_id, job, "retry_exhausted", "retry_exhausted")
|
||||
return
|
||||
delay = backoff_seconds(job["attempt"])
|
||||
await self.db.execute(
|
||||
"WITH u AS (UPDATE attachments SET status = 'queued', lease_owner = NULL, lease_expires_at = NULL, updated_at = now() "
|
||||
"WHERE id = $1 AND status = 'processing' AND lease_owner = $2 RETURNING id, pipeline_version), "
|
||||
"o AS (INSERT INTO outbox (job_id, attachment_id, pipeline_version, attempt, available_at, traceparent, tracestate) "
|
||||
"SELECT $3, id, pipeline_version, $4, now() + make_interval(secs => $5), $7, $8 FROM u RETURNING 1) "
|
||||
"INSERT INTO job_events (attachment_id, job_id, event, attempt, detail) SELECT id, $3, 'retry_scheduled', $4, $6 FROM u",
|
||||
att_id, WORKER_ID, uuid.UUID(job["job_id"]), nxt, delay, reason, job.get("traceparent"), job.get("tracestate"))
|
||||
log("info", "job_rescheduled", attachment_id=str(att_id), next_attempt=nxt, delay_seconds=round(delay, 2), reason=reason)
|
||||
|
||||
async def publish_result(self, row, job, variants):
|
||||
"""Одна транзакция: варианты (идемпотентно) + ready. Только владелец lease может опубликовать результат."""
|
||||
async with self.db.conn() as c:
|
||||
async with c.transaction():
|
||||
owner = await c.fetchval("SELECT 1 FROM attachments WHERE id = $1 AND status = 'processing' AND lease_owner = $2 "
|
||||
"FOR UPDATE", row["id"], WORKER_ID)
|
||||
if owner is None:
|
||||
return False # lease потерян (sweeper вернул задачу): результат не публикуем, чужой повтор доделает
|
||||
for v in variants:
|
||||
await c.execute(
|
||||
"INSERT INTO attachment_variants (attachment_id, pipeline_version, name, content_type, size_bytes, width, "
|
||||
"height, sha256, s3_key) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9) "
|
||||
"ON CONFLICT (attachment_id, pipeline_version, name) DO NOTHING",
|
||||
row["id"], job["pipeline_version"], v["name"], v["content_type"], len(v["data"]), v["width"], v["height"],
|
||||
v["sha256"], "att/%s/v%d/%s" % (row["id"], job["pipeline_version"], v["name"]))
|
||||
await c.execute("UPDATE attachments SET status = 'ready', error_code = NULL, ready_at = now(), updated_at = now(), "
|
||||
"lease_owner = NULL, lease_expires_at = NULL WHERE id = $1", row["id"])
|
||||
await c.execute("INSERT INTO job_events (attachment_id, job_id, event, attempt) VALUES ($1, $2, 'ready', $3)",
|
||||
row["id"], uuid.UUID(job["job_id"]), job["attempt"])
|
||||
return True
|
||||
|
||||
# --- обработка ----------------------------------------------------------------
|
||||
async def process(self, job):
|
||||
att_id = uuid.UUID(job["attachment_id"])
|
||||
row = await self.claim(job)
|
||||
if row is None:
|
||||
state = await self.db.fetchval("SELECT status FROM attachments WHERE id = $1", att_id)
|
||||
log("info", "job_noop", attachment_id=job["attachment_id"], status=state or "unknown")
|
||||
telemetry.record_job("attachment.process.v1", "noop", 0.0)
|
||||
return # ready/failed/processing у другого владельца/неизвестное вложение: идемпотентный no-op, ack
|
||||
started = time.perf_counter()
|
||||
outcome = "ready"
|
||||
try:
|
||||
await self.fault("after_claim")
|
||||
stream, _ = await self.s3.open_stream(row["staging_key"])
|
||||
spool = tempfile.SpooledTemporaryFile(max_size=2 * 1024 * 1024)
|
||||
try:
|
||||
try:
|
||||
while True:
|
||||
chunk = await stream.read(256 * 1024)
|
||||
if not chunk:
|
||||
break
|
||||
spool.write(chunk)
|
||||
finally:
|
||||
stream.close()
|
||||
loop = asyncio.get_running_loop()
|
||||
variants = await asyncio.wait_for(loop.run_in_executor(self.pool, pipeline.process, spool), config.JOB_TIMEOUT)
|
||||
finally:
|
||||
spool.close()
|
||||
# оригинал и варианты в конечный prefix: ключи детерминированы, повторная запись перезаписывает те же байты
|
||||
await self.copy_original(row)
|
||||
for v in variants:
|
||||
await self.s3.put_bytes("att/%s/v%d/%s" % (row["id"], job["pipeline_version"], v["name"]), v["data"], v["content_type"])
|
||||
await self.fault("after_s3_before_commit")
|
||||
if not await self.publish_result(row, job, variants):
|
||||
log("warning", "lease_lost", attachment_id=job["attachment_id"])
|
||||
return
|
||||
await self.fault("after_commit_before_ack")
|
||||
log("info", "job_ready", attachment_id=job["attachment_id"], attempt=job["attempt"],
|
||||
duration_ms=round((time.perf_counter() - started) * 1000, 1))
|
||||
try:
|
||||
await self.s3.delete(row["staging_key"]) # staging безопасно удалять после COMMIT результата
|
||||
except Exception as exc:
|
||||
log("warning", "staging_cleanup_failed", error_type=type(exc).__name__)
|
||||
except pipeline.PermanentError as exc:
|
||||
outcome = "failed"
|
||||
await self.fail(att_id, job, exc.code)
|
||||
log("warning", "job_failed", attachment_id=job["attachment_id"], error_code=exc.code)
|
||||
except MemoryError:
|
||||
outcome = "failed"
|
||||
await self.fail(att_id, job, "processing_failed")
|
||||
log("error", "job_failed_oom", attachment_id=job["attachment_id"])
|
||||
except asyncio.CancelledError:
|
||||
raise
|
||||
except InjectedCrash:
|
||||
raise
|
||||
except Exception as exc: # S3/БД/таймаут/неожиданное: временная ошибка
|
||||
outcome = "retried"
|
||||
await self.reschedule(att_id, job, type(exc).__name__)
|
||||
finally:
|
||||
ttr = None
|
||||
if outcome == "ready":
|
||||
ttr = await self.db.fetchval("SELECT extract(epoch FROM ready_at - accepted_at) FROM attachments WHERE id = $1 "
|
||||
"AND status = 'ready'", att_id)
|
||||
telemetry.record_job("attachment.process.v1", outcome, time.perf_counter() - started, float(ttr) if ttr is not None else None)
|
||||
|
||||
async def copy_original(self, row):
|
||||
stream, size = await self.s3.open_stream(row["staging_key"])
|
||||
spool = tempfile.SpooledTemporaryFile(max_size=2 * 1024 * 1024)
|
||||
try:
|
||||
while True:
|
||||
chunk = await stream.read(256 * 1024)
|
||||
if not chunk:
|
||||
break
|
||||
spool.write(chunk)
|
||||
stream.close()
|
||||
spool.seek(0)
|
||||
await self.s3.put(row["s3_key"], spool, row["content_type"], size)
|
||||
finally:
|
||||
spool.close()
|
||||
|
||||
async def fault(self, point):
|
||||
"""Точки внедрения сбоев для экспериментов (FAULT_INJECTION=crash:<точка> => процесс умирает мгновенно)."""
|
||||
if "crash_once:" + point in config.FAULT_INJECTION:
|
||||
marker = "/tmp/fault-%s.done" % point # сохраняется при рестарте того же контейнера -> сбой ровно один раз
|
||||
if not os.path.exists(marker):
|
||||
open(marker, "w").close()
|
||||
log("error", "fault_crash_once", point=point)
|
||||
os._exit(137)
|
||||
if "crash:" + point in config.FAULT_INJECTION:
|
||||
log("error", "fault_crash", point=point)
|
||||
os._exit(137)
|
||||
if "fail:" + point in config.FAULT_INJECTION:
|
||||
raise OSError("injected failure at " + point)
|
||||
|
||||
async def on_message(self, message):
|
||||
try:
|
||||
job = json.loads(message.body)
|
||||
if (job.get("schema_version") != 1 or job.get("type") != "attachment.process.v1"
|
||||
or not isinstance(job.get("attempt"), int) or not isinstance(job.get("pipeline_version"), int)):
|
||||
raise ValueError("bad job")
|
||||
uuid.UUID(job["attachment_id"]), uuid.UUID(job["job_id"])
|
||||
except Exception:
|
||||
log("error", "job_unreadable")
|
||||
await message.reject(requeue=False) # -> DLQ
|
||||
return
|
||||
self.busy += 1
|
||||
carrier = {k: job[k] for k in ("traceparent", "tracestate") if k in job}
|
||||
try:
|
||||
with telemetry.span("attachment.process process", "consumer",
|
||||
{"messaging.system": "rabbitmq", "messaging.destination.name": "attachment.process",
|
||||
"messaging.operation": "process", "job.type": job["type"], "job.attempt": job["attempt"]}, carrier):
|
||||
await self.process(job)
|
||||
await message.ack()
|
||||
except InjectedCrash:
|
||||
raise
|
||||
except Exception as exc:
|
||||
log("error", "job_unhandled", error_type=type(exc).__name__)
|
||||
await message.nack(requeue=True)
|
||||
finally:
|
||||
self.busy -= 1
|
||||
|
||||
# --- sweeper -------------------------------------------------------------------
|
||||
async def sweep_once(self):
|
||||
"""Истёкший lease -> queued (новая попытка) или failed/retry_exhausted; чистка старого staging."""
|
||||
rows = await self.db.fetch(
|
||||
"WITH exp AS (SELECT id FROM attachments WHERE status = 'processing' AND lease_expires_at < now() FOR UPDATE SKIP LOCKED), "
|
||||
"u AS (UPDATE attachments a SET status = CASE WHEN a.attempt >= $1 THEN 'failed' ELSE 'queued' END, "
|
||||
"error_code = CASE WHEN a.attempt >= $1 THEN 'retry_exhausted' ELSE NULL END, lease_owner = NULL, lease_expires_at = NULL, "
|
||||
"updated_at = now() FROM exp WHERE a.id = exp.id RETURNING a.id, a.status, a.pipeline_version, a.attempt, a.job_id), "
|
||||
"o AS (INSERT INTO outbox (job_id, attachment_id, pipeline_version, attempt, traceparent, tracestate) "
|
||||
"SELECT COALESCE(job_id, gen_random_uuid()), id, pipeline_version, attempt + 1, "
|
||||
"(SELECT traceparent FROM outbox WHERE attachment_id = u.id ORDER BY id DESC LIMIT 1), "
|
||||
"(SELECT tracestate FROM outbox WHERE attachment_id = u.id ORDER BY id DESC LIMIT 1) FROM u WHERE status = 'queued' RETURNING 1), "
|
||||
"e AS (INSERT INTO job_events (attachment_id, job_id, event, attempt, detail) SELECT id, job_id, 'lease_expired', attempt, status FROM u RETURNING 1) "
|
||||
"SELECT id, status FROM u", config.MAX_ATTEMPTS)
|
||||
for r in rows:
|
||||
log("warning", "lease_expired", attachment_id=str(r["id"]), new_status=r["status"])
|
||||
if rows:
|
||||
self.relay.kick()
|
||||
|
||||
async def sweeper(self):
|
||||
while not self.stopping:
|
||||
try:
|
||||
await self.sweep_once()
|
||||
except Exception as exc:
|
||||
log("warning", "sweeper_failed", error_type=type(exc).__name__)
|
||||
await asyncio.sleep(config.SWEEP_INTERVAL)
|
||||
|
||||
async def staging_janitor(self):
|
||||
"""Удаляет staging для ready (если не удалён после COMMIT) и для failed старше retention; текущую работу не трогает."""
|
||||
while not self.stopping:
|
||||
await asyncio.sleep(max(30, config.SWEEP_INTERVAL * 6))
|
||||
try:
|
||||
rows = await self.db.fetch(
|
||||
"SELECT id, staging_key FROM attachments WHERE staging_key IS NOT NULL AND (status = 'ready' OR "
|
||||
"(status = 'failed' AND updated_at < now() - make_interval(secs => $1))) LIMIT 200",
|
||||
float(config.STAGING_FAILED_RETENTION_SECONDS))
|
||||
for r in rows:
|
||||
await self.s3.delete(r["staging_key"])
|
||||
await self.db.execute("UPDATE attachments SET staging_key = NULL WHERE id = $1", r["id"])
|
||||
except Exception as exc:
|
||||
log("warning", "staging_janitor_failed", error_type=type(exc).__name__)
|
||||
|
||||
# --- жизненный цикл ------------------------------------------------------------
|
||||
async def connect_broker(self):
|
||||
while not self.stopping:
|
||||
try:
|
||||
await self.broker.ensure(prefetch=config.WORKER_CONCURRENCY)
|
||||
return
|
||||
except Exception as exc:
|
||||
log("warning", "broker_unavailable", error_type=type(exc).__name__)
|
||||
await asyncio.sleep(2)
|
||||
|
||||
async def consume(self):
|
||||
while not self.stopping:
|
||||
await self.connect_broker()
|
||||
try:
|
||||
queue = await self.broker.channel.declare_queue(QUEUE, durable=True, passive=True)
|
||||
tag = await queue.consume(self.on_message)
|
||||
log("info", "consuming", worker=WORKER_ID, prefetch=config.WORKER_CONCURRENCY)
|
||||
while not self.stopping and not self.broker.channel.is_closed:
|
||||
await asyncio.sleep(1)
|
||||
if self.stopping:
|
||||
await queue.cancel(tag)
|
||||
return
|
||||
except Exception as exc:
|
||||
log("warning", "consume_failed", error_type=type(exc).__name__)
|
||||
await asyncio.sleep(2)
|
||||
|
||||
async def run(self):
|
||||
telemetry.setup(config.SERVICE_NAME)
|
||||
telemetry.process_metrics()
|
||||
await self.db.start()
|
||||
await self.s3.start()
|
||||
log("info", "worker_started", worker=WORKER_ID, concurrency=config.WORKER_CONCURRENCY, lease=config.LEASE_SECONDS,
|
||||
max_attempts=config.MAX_ATTEMPTS)
|
||||
loop = asyncio.get_running_loop()
|
||||
for sig in (signal.SIGTERM, signal.SIGINT):
|
||||
try:
|
||||
loop.add_signal_handler(sig, self.request_stop)
|
||||
except NotImplementedError:
|
||||
signal.signal(sig, lambda *_: self.request_stop())
|
||||
self.tasks = [asyncio.create_task(self.consume()), asyncio.create_task(self.relay.run()),
|
||||
asyncio.create_task(self.sweeper()), asyncio.create_task(self.staging_janitor())]
|
||||
beat = 0.0
|
||||
while not self.stopping:
|
||||
if time.monotonic() - beat > 5: # heartbeat для HEALTHCHECK контейнера
|
||||
beat = time.monotonic()
|
||||
try:
|
||||
with open("/tmp/worker.alive", "w") as fh:
|
||||
fh.write(str(time.time()))
|
||||
except OSError:
|
||||
pass
|
||||
await asyncio.sleep(0.5)
|
||||
# graceful: перестаём брать новые задания, ждём текущие до GRACEFUL_SHUTDOWN_SECONDS
|
||||
deadline = time.monotonic() + config.GRACEFUL_SHUTDOWN_SECONDS
|
||||
while self.busy and time.monotonic() < deadline:
|
||||
await asyncio.sleep(0.2)
|
||||
self.relay.stop()
|
||||
for t in self.tasks:
|
||||
t.cancel()
|
||||
await self.broker.stop()
|
||||
await self.s3.stop()
|
||||
await self.db.stop()
|
||||
log("info", "worker_stopped", unfinished=self.busy)
|
||||
telemetry.shutdown()
|
||||
|
||||
def request_stop(self):
|
||||
log("info", "worker_stopping")
|
||||
self.stopping = True
|
||||
|
||||
|
||||
class InjectedCrash(Exception):
|
||||
pass
|
||||
|
||||
|
||||
def main():
|
||||
asyncio.run(Worker().run())
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
|
After Width: | Height: | Size: 344 B |
|
After Width: | Height: | Size: 1.6 KiB |
|
After Width: | Height: | Size: 151 B |
|
After Width: | Height: | Size: 24 KiB |
|
After Width: | Height: | Size: 109 KiB |
|
After Width: | Height: | Size: 132 B |
@@ -0,0 +1,208 @@
|
||||
"""Собственные тесты lab2 против запущенного стенда (docker compose up -d).
|
||||
|
||||
python -m unittest discover -s student-tests -v # BASE_URL по умолчанию http://127.0.0.1:8080
|
||||
Тесты используют только синтетические данные и ничего не удаляют. Проверки expiry и недоступности
|
||||
зависимостей лежат отдельно (test_expiry.py, test_dependencies.py): они управляют контейнерами.
|
||||
"""
|
||||
import base64
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import unittest
|
||||
import urllib.error
|
||||
import urllib.request
|
||||
import uuid
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
from urllib.parse import quote
|
||||
|
||||
BASE = os.environ.get("BASE_URL", "http://127.0.0.1:8080").rstrip("/")
|
||||
ROOT = os.path.join(os.path.dirname(__file__), "..")
|
||||
|
||||
|
||||
def call(method, path, body=None, token=None, headers=None, raw=None):
|
||||
h = dict(headers or {})
|
||||
if method == "POST" and path.endswith("/messages") and headers is None:
|
||||
h["Idempotency-Key"] = uuid.uuid4().hex # lab3: ключ обязателен; тесты без ключа — в test_idempotency.py
|
||||
data = raw
|
||||
if body is not None:
|
||||
data = json.dumps(body, ensure_ascii=False).encode("utf-8")
|
||||
h.setdefault("Content-Type", "application/json")
|
||||
if token:
|
||||
h["Authorization"] = "Bearer " + token
|
||||
req = urllib.request.Request(BASE + path, data=data, headers=h, method=method)
|
||||
try:
|
||||
with urllib.request.urlopen(req, timeout=10) as resp:
|
||||
status, hdrs, payload = resp.status, resp.headers, resp.read()
|
||||
except urllib.error.HTTPError as err:
|
||||
status, hdrs, payload = err.code, err.headers, err.read()
|
||||
return status, hdrs, (json.loads(payload) if payload else None)
|
||||
|
||||
|
||||
def basic(username, password):
|
||||
return {"Authorization": "Basic " + base64.b64encode((username + ":" + password).encode("utf-8")).decode()}
|
||||
|
||||
|
||||
def register(prefix="u"):
|
||||
name = "%s_%s" % (prefix, uuid.uuid4().hex[:12])
|
||||
password = "Pa55:word-" + uuid.uuid4().hex # двоеточие в пароле допустимо
|
||||
status, _, user = call("POST", "/api/v1/users", {"username": name, "display_name": "Имя " + name, "password": password})
|
||||
assert status == 201, user
|
||||
status, _, session = call("POST", "/api/v1/auth/sessions", headers=basic(name, password))
|
||||
assert status == 201, session
|
||||
return {"user": user, "password": password, "token": session["token"], "session": session}
|
||||
|
||||
|
||||
def psql(sql):
|
||||
out = subprocess.run(["docker", "compose", "--env-file", ".env", "exec", "-T", "postgres", "psql", "-U", "messenger",
|
||||
"-d", "messenger", "-Atc", sql], cwd=ROOT, capture_output=True, text=True, check=True)
|
||||
return out.stdout.strip()
|
||||
|
||||
|
||||
class AuthTest(unittest.TestCase):
|
||||
def assertError(self, result, status, code):
|
||||
self.assertEqual(result[0], status, result[2])
|
||||
self.assertEqual(result[2]["error"]["code"], code)
|
||||
self.assertEqual(result[2]["error"]["request_id"], result[1]["X-Request-Id"])
|
||||
|
||||
def test_register_validation(self):
|
||||
base = {"username": "val_" + uuid.uuid4().hex[:8], "display_name": "x", "password": "p" * 12}
|
||||
for patch in ({"password": "short"}, {"password": "p" * 129}, {"password": 12345678901234}, {"extra": 1},
|
||||
{"username": "BAD"}, {"display_name": " "}):
|
||||
body = dict(base, **patch)
|
||||
self.assertError(call("POST", "/api/v1/users", body), 400, "invalid_request")
|
||||
missing = {k: v for k, v in base.items() if k != "password"}
|
||||
self.assertError(call("POST", "/api/v1/users", missing), 400, "invalid_request")
|
||||
self.assertEqual(call("POST", "/api/v1/users", dict(base, password="я" * 12))[0], 201) # 12 code points
|
||||
|
||||
def test_password_is_not_exposed_or_stored_plain(self):
|
||||
actor = register()
|
||||
self.assertEqual(set(actor["user"]), {"id", "username", "display_name", "created_at"})
|
||||
stored = psql("SELECT password_hash FROM users WHERE username = '%s'" % actor["user"]["username"])
|
||||
self.assertTrue(stored.startswith("$argon2id$"), stored[:12])
|
||||
self.assertNotIn(actor["password"], stored)
|
||||
|
||||
def test_wrong_password_and_unknown_user_are_indistinguishable(self):
|
||||
actor = register()
|
||||
wrong = call("POST", "/api/v1/auth/sessions", headers=basic(actor["user"]["username"], "wrong-password-123"))
|
||||
unknown = call("POST", "/api/v1/auth/sessions", headers=basic("nobody_" + uuid.uuid4().hex[:8], "wrong-password-123"))
|
||||
for r in (wrong, unknown):
|
||||
self.assertError(r, 401, "unauthorized")
|
||||
self.assertEqual(r[1]["WWW-Authenticate"], 'Basic realm="messenger", charset="UTF-8"')
|
||||
self.assertEqual(wrong[2]["error"]["message"], unknown[2]["error"]["message"])
|
||||
for bad in ({}, {"Authorization": "Basic !!!"}, {"Authorization": "Basic " + base64.b64encode(b"nocolon").decode()},
|
||||
{"Authorization": "Bearer abc"}):
|
||||
self.assertEqual(call("POST", "/api/v1/auth/sessions", headers=bad)[0], 401)
|
||||
|
||||
def test_bearer_rules(self):
|
||||
actor = register()
|
||||
self.assertEqual(call("GET", "/api/v1/users/me", token=actor["token"])[2], actor["user"])
|
||||
for headers in ({}, basic(actor["user"]["username"], actor["password"]), {"X-User-Id": actor["user"]["id"]},
|
||||
{"Authorization": "Bearer " + "A" * 43}, {"Authorization": "Bearer short"}):
|
||||
r = call("GET", "/api/v1/users/me", headers=headers)
|
||||
self.assertError(r, 401, "unauthorized")
|
||||
self.assertEqual(r[1]["WWW-Authenticate"], 'Bearer realm="messenger"')
|
||||
other = register()
|
||||
me = call("GET", "/api/v1/users/me", token=actor["token"], headers={"X-User-Id": other["user"]["id"]})[2]
|
||||
self.assertEqual(me["id"], actor["user"]["id"]) # X-User-Id игнорируется
|
||||
|
||||
def test_session_shape_entropy_and_logout(self):
|
||||
actor = register()
|
||||
s = actor["session"]
|
||||
self.assertEqual(s["token_type"], "Bearer")
|
||||
self.assertGreaterEqual(len(s["token"]), 43) # 32 байта в base64url = 256 бит
|
||||
self.assertNotIn(actor["user"]["id"], s["token"])
|
||||
self.assertNotEqual(register()["token"], s["token"])
|
||||
self.assertEqual(call("DELETE", "/api/v1/auth/sessions/current", token=s["token"])[0], 204)
|
||||
self.assertError(call("GET", "/api/v1/users/me", token=s["token"]), 401, "unauthorized")
|
||||
self.assertError(call("DELETE", "/api/v1/auth/sessions/current", token=s["token"]), 401, "unauthorized")
|
||||
|
||||
def test_logout_revokes_in_database_and_cache(self):
|
||||
actor = register()
|
||||
call("GET", "/api/v1/users/me", token=actor["token"]) # прогреваем кэш
|
||||
call("DELETE", "/api/v1/auth/sessions/current", token=actor["token"])
|
||||
revoked = psql("SELECT count(*) FROM sessions s JOIN users u ON u.id = s.user_id "
|
||||
"WHERE u.username = '%s' AND s.revoked_at IS NOT NULL" % actor["user"]["username"])
|
||||
self.assertEqual(revoked, "1")
|
||||
|
||||
def test_token_only_stored_as_hash(self):
|
||||
actor = register()
|
||||
count = psql("SELECT count(*) FROM sessions WHERE token_hash = '\\x00'::bytea OR encode(token_hash,'escape') = '%s'" % actor["token"])
|
||||
self.assertEqual(count, "0")
|
||||
|
||||
def test_concurrent_registration_of_same_username(self):
|
||||
name = "race_" + uuid.uuid4().hex[:10]
|
||||
body = {"username": name, "display_name": "x", "password": "racing-password-1"}
|
||||
with ThreadPoolExecutor(12) as pool:
|
||||
codes = sorted(pool.map(lambda _: call("POST", "/api/v1/users", body)[0], range(12)))
|
||||
self.assertEqual(codes, [201] + [409] * 11)
|
||||
self.assertEqual(psql("SELECT count(*) FROM users WHERE username = '%s'" % name), "1")
|
||||
|
||||
|
||||
class DataTest(unittest.TestCase):
|
||||
def test_chat_creation_is_atomic(self):
|
||||
a = register()
|
||||
before = psql("SELECT (SELECT count(*) FROM chats) || '/' || (SELECT count(*) FROM chat_members)")
|
||||
ghost = str(uuid.uuid4())
|
||||
r = call("POST", "/api/v1/chats", {"title": "откат", "member_ids": [a["user"]["id"], ghost]}, token=a["token"])
|
||||
self.assertEqual(r[0], 400)
|
||||
after = psql("SELECT (SELECT count(*) FROM chats) || '/' || (SELECT count(*) FROM chat_members)")
|
||||
self.assertEqual(before, after, "после неудачного создания не должно остаться чата или участников")
|
||||
ok = call("POST", "/api/v1/chats", {"title": "ок", "member_ids": [a["user"]["id"]]}, token=a["token"])
|
||||
self.assertEqual(ok[0], 201)
|
||||
|
||||
def test_foreign_chat_empty_message_bad_json(self):
|
||||
a, b, eve = register(), register(), register()
|
||||
chat = call("POST", "/api/v1/chats", {"title": "ЛС", "member_ids": [a["user"]["id"], b["user"]["id"]]}, a["token"])[2]
|
||||
path = "/api/v1/chats/%s/messages" % chat["id"]
|
||||
self.assertEqual(call("POST", path, {"text": "чужое"}, eve["token"])[0], 404)
|
||||
self.assertEqual(call("GET", path, token=eve["token"])[0], 404)
|
||||
self.assertEqual(call("POST", path, {"text": " "}, a["token"])[0], 400)
|
||||
self.assertEqual(call("POST", path, raw=b"{oops", token=a["token"], headers={"Content-Type": "application/json"})[0], 400)
|
||||
self.assertEqual(call("POST", path, {"text": "x", "sender_id": b["user"]["id"]}, a["token"])[0], 400)
|
||||
sent = call("POST", path, {"text": "Привет 👋"}, a["token"])[2]
|
||||
self.assertEqual(sent["sender_id"], a["user"]["id"])
|
||||
|
||||
def test_multi_page_history_and_cursor_binding(self):
|
||||
a, b = register(), register()
|
||||
chat = call("POST", "/api/v1/chats", {"title": "стр", "member_ids": [a["user"]["id"], b["user"]["id"]]}, a["token"])[2]
|
||||
other = call("POST", "/api/v1/chats", {"title": "др", "member_ids": [a["user"]["id"]]}, a["token"])[2]
|
||||
path = "/api/v1/chats/%s/messages" % chat["id"]
|
||||
sent = [call("POST", path, {"text": "m%d" % i}, (a if i % 2 else b)["token"])[2] for i in range(23)]
|
||||
walked, cursor, pages = [], None, 0
|
||||
while True:
|
||||
q = "?limit=%d" % (5 if pages % 2 == 0 else 8) + ("&cursor=" + quote(cursor, safe="") if cursor else "")
|
||||
page = call("GET", path + q, token=b["token"])[2]
|
||||
walked += page["items"]
|
||||
pages += 1
|
||||
cursor = page["next_cursor"]
|
||||
if not cursor:
|
||||
break
|
||||
self.assertEqual(walked, sent)
|
||||
self.assertGreaterEqual(pages, 3)
|
||||
first = call("GET", path + "?limit=1", token=a["token"])[2]["next_cursor"]
|
||||
r = call("GET", "/api/v1/chats/%s/messages?cursor=%s" % (other["id"], quote(first, safe="")), token=a["token"])
|
||||
self.assertEqual(r[0], 400)
|
||||
|
||||
def test_parallel_writes_from_many_clients(self):
|
||||
users = [register() for _ in range(5)]
|
||||
chat = call("POST", "/api/v1/chats", {"title": "гонки", "member_ids": [u["user"]["id"] for u in users]}, users[0]["token"])[2]
|
||||
path = "/api/v1/chats/%s/messages" % chat["id"]
|
||||
|
||||
def client(u):
|
||||
return [call("POST", path, {"text": "%s-%d" % (u["user"]["username"], i)}, u["token"])[0] for i in range(20)]
|
||||
|
||||
with ThreadPoolExecutor(5) as pool:
|
||||
codes = [c for part in pool.map(client, users) for c in part]
|
||||
self.assertEqual(codes, [201] * 100)
|
||||
total, cursor = [], None
|
||||
while True:
|
||||
page = call("GET", path + "?limit=100" + ("&cursor=" + quote(cursor, safe="") if cursor else ""), token=users[0]["token"])[2]
|
||||
total += page["items"]
|
||||
cursor = page["next_cursor"]
|
||||
if not cursor:
|
||||
break
|
||||
self.assertEqual(len({m["id"] for m in total}), 100)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,205 @@
|
||||
"""Собственные тесты вложений lab6 (негативные сценарии и права). Запуск:
|
||||
python -m unittest discover -s student-tests -p "test_attachments.py" -v (BASE_URL по умолчанию http://127.0.0.1:8080)
|
||||
"""
|
||||
import hashlib
|
||||
import json
|
||||
import os
|
||||
import socket
|
||||
import struct
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
import urllib.error
|
||||
import urllib.parse
|
||||
import urllib.request
|
||||
import uuid
|
||||
import zlib
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import BASE, ROOT, call, psql, register # noqa: E402
|
||||
|
||||
FIX = os.path.join(os.path.dirname(__file__), "fixtures")
|
||||
PNG = open(os.path.join(FIX, "sample.png"), "rb").read()
|
||||
JPG = open(os.path.join(FIX, "sample.jpg"), "rb").read()
|
||||
|
||||
|
||||
def upload(token, chat_id, data, ctype="image/png", filename="sample.png", raw_filename=None):
|
||||
qs = raw_filename if raw_filename is not None else "filename=" + filename
|
||||
req = urllib.request.Request("%s/api/v1/chats/%s/attachments?%s" % (BASE, chat_id, qs), data=data, method="POST",
|
||||
headers={"Authorization": "Bearer " + token, "Content-Type": ctype})
|
||||
try:
|
||||
with urllib.request.urlopen(req, timeout=30) as r:
|
||||
return r.status, r.headers, json.loads(r.read())
|
||||
except urllib.error.HTTPError as err:
|
||||
body = err.read()
|
||||
return err.code, err.headers, (json.loads(body) if body else None)
|
||||
|
||||
|
||||
def download(token, att_id, suffix="/content"):
|
||||
req = urllib.request.Request("%s/api/v1/attachments/%s%s" % (BASE, att_id, suffix),
|
||||
headers={"Authorization": "Bearer " + token if token else ""})
|
||||
try:
|
||||
with urllib.request.urlopen(req, timeout=30) as r:
|
||||
return r.status, r.headers, r.read()
|
||||
except urllib.error.HTTPError as err:
|
||||
return err.code, err.headers, err.read()
|
||||
|
||||
|
||||
def wait_ready(token, att_id, seconds=60):
|
||||
"""lab6: загрузка возвращает 202 queued; готовность определяется опросом метаданных."""
|
||||
deadline = time.time() + seconds
|
||||
while time.time() < deadline:
|
||||
code, _, body = download(token, att_id, "")
|
||||
att = json.loads(body)
|
||||
if att["status"] == "ready":
|
||||
return att
|
||||
assert att["status"] != "failed", att
|
||||
time.sleep(0.3)
|
||||
raise AssertionError("not ready")
|
||||
|
||||
|
||||
def png_chunk(kind, data):
|
||||
return struct.pack(">I", len(data)) + kind + data + struct.pack(">I", zlib.crc32(kind + data) & 0xFFFFFFFF)
|
||||
|
||||
|
||||
def bomb_png(width=20000, height=20000):
|
||||
"""Корректный по структуре PNG, заявляющий 400 млн пикселей, без данных изображения (отклоняется по заголовку)."""
|
||||
ihdr = struct.pack(">IIBBBBB", width, height, 8, 2, 0, 0, 0)
|
||||
return b"\x89PNG\r\n\x1a\n" + png_chunk(b"IHDR", ihdr) + png_chunk(b"IDAT", zlib.compress(b"\x00" * 16)) + png_chunk(b"IEND", b"")
|
||||
|
||||
|
||||
class AttachmentTest(unittest.TestCase):
|
||||
def setUp(self):
|
||||
self.a, self.b, self.eve = register("att"), register("att"), register("att")
|
||||
self.chat = call("POST", "/api/v1/chats", {"title": "файлы", "member_ids": [self.a["user"]["id"], self.b["user"]["id"]]}, self.a["token"])[2]
|
||||
self.other = call("POST", "/api/v1/chats", {"title": "другой", "member_ids": [self.a["user"]["id"]]}, self.a["token"])[2]
|
||||
|
||||
def assertError(self, result, status, code=None):
|
||||
self.assertEqual(result[0], status, result[2])
|
||||
if code:
|
||||
self.assertEqual(result[2]["error"]["code"], code)
|
||||
|
||||
def test_png_and_jpeg_roundtrip_with_checksum(self):
|
||||
for data, ctype, name in ((PNG, "image/png", "фото.png"), (JPG, "image/jpeg", "photo.jpg")):
|
||||
status, headers, att = upload(self.a["token"], self.chat["id"], data, ctype, urllib.parse.quote(name))
|
||||
self.assertEqual(status, 202, att)
|
||||
self.assertEqual((att["status"], att["variants"], att["error_code"]), ("queued", [], None))
|
||||
wait_ready(self.b["token"], att["id"])
|
||||
self.assertEqual((att["size_bytes"], att["sha256"], att["content_type"]), (len(data), hashlib.sha256(data).hexdigest(), ctype))
|
||||
self.assertEqual((att["chat_id"], att["owner_id"], att["filename"]), (self.chat["id"], self.a["user"]["id"], name))
|
||||
code, hdr, body = download(self.b["token"], att["id"]) # другой участник чата
|
||||
self.assertEqual((code, body, hdr.get_content_type()), (200, data, ctype))
|
||||
self.assertEqual(download(self.b["token"], att["id"], "")[0], 200)
|
||||
|
||||
def test_too_large_is_413_and_nothing_is_stored(self):
|
||||
before = psql("SELECT count(*) FROM attachments")
|
||||
r = upload(self.a["token"], self.chat["id"], PNG[:8] + b"\x00" * (10 * 1024 * 1024 + 1))
|
||||
self.assertError(r, 413, "payload_too_large")
|
||||
self.assertEqual(psql("SELECT count(*) FROM attachments"), before)
|
||||
|
||||
def test_corrupted_and_spoofed_files_are_415_not_500(self):
|
||||
cases = {
|
||||
"truncated png": (PNG[: len(PNG) // 2], "image/png"),
|
||||
"garbage png": (b"\x89PNG\r\n\x1a\n" + b"\x00" * 200, "image/png"),
|
||||
"random bytes": (os.urandom(500), "image/png"),
|
||||
"truncated jpeg": (JPG[: len(JPG) // 2], "image/jpeg"),
|
||||
"garbage jpeg": (b"\xff\xd8\xff\xe0" + b"\x00" * 200, "image/jpeg"),
|
||||
"png as jpeg": (PNG, "image/jpeg"),
|
||||
"jpeg as png": (JPG, "image/png"),
|
||||
"text as png": (b"hello, i am not an image", "image/png"),
|
||||
"gif": (b"GIF89a" + b"\x00" * 50, "image/png"),
|
||||
}
|
||||
for name, (data, ctype) in cases.items():
|
||||
with self.subTest(name):
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], data, ctype), 415, "unsupported_media_type")
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], PNG, "text/plain"), 415)
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], PNG, "multipart/form-data; boundary=x"), 415)
|
||||
|
||||
def test_decompression_bomb_header_is_413(self):
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], bomb_png()), 413, "payload_too_large")
|
||||
# 6000x6600 = 39.6 млн пикселей < лимита: по размерам допустимо, но данных изображения нет -> файл повреждён (415)
|
||||
self.assertEqual(upload(self.a["token"], self.chat["id"], bomb_png(6000, 6600))[0], 415)
|
||||
|
||||
def test_dangerous_filenames_are_400(self):
|
||||
bad = ["a%2Fb.png", "..%5Cx.png", "x%00.png", "x%0A.png", "x%7F.png", "", "%20%20", "n" * 256]
|
||||
for name in bad:
|
||||
with self.subTest(name):
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], PNG, raw_filename="filename=" + name), 400, "invalid_request")
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], PNG, raw_filename=""), 400)
|
||||
self.assertEqual(upload(self.a["token"], self.chat["id"], PNG, raw_filename="filename=" + "n" * 255)[0], 202)
|
||||
|
||||
def test_empty_body_is_400(self):
|
||||
self.assertError(upload(self.a["token"], self.chat["id"], b""), 400)
|
||||
|
||||
def test_foreign_user_cannot_upload_read_or_download(self):
|
||||
self.assertError(upload(self.eve["token"], self.chat["id"], PNG), 404)
|
||||
att = upload(self.a["token"], self.chat["id"], PNG)[2]
|
||||
wait_ready(self.a["token"], att["id"])
|
||||
for suffix in ("", "/content"):
|
||||
self.assertEqual(download(self.eve["token"], att["id"], suffix)[0], 404)
|
||||
self.assertEqual(download(None, att["id"], suffix)[0], 401)
|
||||
self.assertEqual(download(self.a["token"], str(uuid.uuid4()))[0], 404)
|
||||
self.assertEqual(download(self.a["token"], "not-a-uuid")[0], 400)
|
||||
self.assertEqual(download(self.a["token"], att["id"], "/content?variant=thumbnail")[0], 200) # после ready вариант доступен
|
||||
self.assertEqual(download(self.a["token"], att["id"], "/content?variant=bogus")[0], 400)
|
||||
|
||||
def test_message_with_attachments_and_chat_binding(self):
|
||||
att = upload(self.a["token"], self.chat["id"], PNG)[2]
|
||||
foreign_chat_att = upload(self.a["token"], self.other["id"], JPG, "image/jpeg")[2]
|
||||
path = "/api/v1/chats/%s/messages" % self.chat["id"]
|
||||
sent = call("POST", path, {"text": "", "attachment_ids": [att["id"]]}, self.a["token"])
|
||||
self.assertEqual(sent[0], 201, sent[2])
|
||||
self.assertEqual(sent[2]["attachment_ids"], [att["id"]])
|
||||
self.assertEqual(call("POST", path, {"text": "ещё раз то же вложение", "attachment_ids": [att["id"]]}, self.b["token"])[0], 201)
|
||||
listed = call("GET", path + "?limit=10", token=self.b["token"])[2]["items"]
|
||||
self.assertEqual([m["attachment_ids"] for m in listed], [[att["id"]], [att["id"]]])
|
||||
# вложение другого чата / несуществующее -> 404; дубликаты, >10, пустое сообщение без вложений -> 400
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [foreign_chat_att["id"]]}, self.a["token"])[0], 404)
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [str(uuid.uuid4())]}, self.a["token"])[0], 404)
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [att["id"], att["id"]]}, self.a["token"])[0], 400)
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [str(uuid.uuid4()) for _ in range(11)]}, self.a["token"])[0], 400)
|
||||
self.assertEqual(call("POST", path, {"text": " ", "attachment_ids": []}, self.a["token"])[0], 400)
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": "nope"}, self.a["token"])[0], 400)
|
||||
# чужой пользователь не может сослаться на вложение
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [att["id"]]}, self.eve["token"])[0], 404)
|
||||
|
||||
def test_idempotent_replay_with_attachments(self):
|
||||
att = upload(self.a["token"], self.chat["id"], PNG)[2]
|
||||
path = "/api/v1/chats/%s/messages" % self.chat["id"]
|
||||
key = uuid.uuid4().hex
|
||||
h = {"Idempotency-Key": key}
|
||||
first = call("POST", path, {"text": "t", "attachment_ids": [att["id"]]}, self.a["token"], headers=h)
|
||||
again = call("POST", path, {"attachment_ids": [att["id"]], "text": "t"}, self.a["token"], headers=h) # другой порядок ключей
|
||||
self.assertEqual((first[0], again[0], first[2]), (201, 201, again[2]))
|
||||
self.assertEqual(call("POST", path, {"text": "t"}, self.a["token"], headers=h)[0], 409) # другое тело
|
||||
text_key = uuid.uuid4().hex
|
||||
m1 = call("POST", path, {"text": "без вложений"}, self.a["token"], headers={"Idempotency-Key": text_key})
|
||||
m2 = call("POST", path, {"text": "без вложений", "attachment_ids": []}, self.a["token"], headers={"Idempotency-Key": text_key})
|
||||
self.assertEqual(m1[2], m2[2]) # отсутствие == []
|
||||
self.assertEqual(m1[2]["attachment_ids"], [])
|
||||
|
||||
def test_aborted_upload_leaves_nothing(self):
|
||||
before = (psql("SELECT count(*) FROM attachments"))
|
||||
host, port = BASE.split("//")[1].split(":")
|
||||
with socket.create_connection((host, int(port))) as s:
|
||||
s.sendall(("POST /api/v1/chats/%s/attachments?filename=a.png HTTP/1.1\r\nHost: x\r\nAuthorization: Bearer %s\r\n"
|
||||
"Content-Type: image/png\r\nContent-Length: 200000\r\n\r\n" % (self.chat["id"], self.a["token"])).encode() + PNG[:5000])
|
||||
import time
|
||||
time.sleep(1)
|
||||
self.assertEqual(psql("SELECT count(*) FROM attachments"), before)
|
||||
|
||||
def test_bucket_is_private_outside_the_api(self):
|
||||
att = upload(self.a["token"], self.chat["id"], PNG)[2]
|
||||
wait_ready(self.a["token"], att["id"])
|
||||
key = psql("SELECT s3_key FROM attachments WHERE id = '%s'" % att["id"])
|
||||
import subprocess
|
||||
code = ("import urllib.request,urllib.error,sys\n"
|
||||
"try:\n urllib.request.urlopen('http://seaweedfs:8333/messenger-attachments/%s', timeout=5); print('OPEN')\n"
|
||||
"except urllib.error.HTTPError as e: print(e.code)\n" % key)
|
||||
out = subprocess.run(["docker", "compose", "--env-file", ".env", "exec", "-T", "api", "python", "-c", code], cwd=ROOT,
|
||||
capture_output=True, text=True).stdout.strip()
|
||||
self.assertEqual(out, "403", "анонимный GET объекта напрямую из S3 должен быть запрещён")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,121 @@
|
||||
"""Идемпотентность отправки сообщений (lab3) и распределение по репликам. Запуск:
|
||||
python student-tests/test_idempotency.py (BASE_URL по умолчанию http://127.0.0.1:8080, через LB)
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
import unittest
|
||||
import uuid
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
from urllib.parse import quote
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import call, psql, register # noqa: E402
|
||||
|
||||
|
||||
def chat_with(owner, *others):
|
||||
ids = [owner["user"]["id"]] + [o["user"]["id"] for o in others]
|
||||
return call("POST", "/api/v1/chats", {"title": "идемпотентность", "member_ids": ids}, owner["token"])[2]
|
||||
|
||||
|
||||
def history(token, chat_id):
|
||||
items, cursor = [], None
|
||||
while True:
|
||||
page = call("GET", "/api/v1/chats/%s/messages?limit=100%s" % (chat_id, "&cursor=" + quote(cursor, safe="") if cursor else ""), token=token)[2]
|
||||
items += page["items"]
|
||||
cursor = page["next_cursor"]
|
||||
if not cursor:
|
||||
return items
|
||||
|
||||
|
||||
class IdempotencyTest(unittest.TestCase):
|
||||
def setUp(self):
|
||||
self.a, self.b, self.eve = register(), register(), register()
|
||||
self.chat = chat_with(self.a, self.b)
|
||||
self.path = "/api/v1/chats/%s/messages" % self.chat["id"]
|
||||
|
||||
def post(self, actor, body, key, path=None):
|
||||
return call("POST", path or self.path, body, actor["token"], headers={"Idempotency-Key": key})
|
||||
|
||||
def test_key_validation(self):
|
||||
for headers in ({}, {"Idempotency-Key": ""}, {"Idempotency-Key": "a b"},
|
||||
{"Idempotency-Key": "x" * 129}, {"Idempotency-Key": "a/b"}):
|
||||
h = dict(headers)
|
||||
h.setdefault("Content-Type", "application/json")
|
||||
r = call("POST", self.path, raw=b'{"text":"x"}', token=self.a["token"], headers=h)
|
||||
self.assertEqual(r[0], 400, headers)
|
||||
self.assertEqual(r[2]["error"]["code"], "invalid_request")
|
||||
self.assertEqual(self.post(self.a, {"text": "ok"}, "A-z_0.9:-" + "k" * 100)[0], 201)
|
||||
|
||||
def test_replay_returns_same_message_and_creates_no_duplicate(self):
|
||||
key = uuid.uuid4().hex
|
||||
first = self.post(self.a, {"text": "один раз"}, key)
|
||||
second = self.post(self.a, {"text": "один раз"}, key)
|
||||
self.assertEqual((first[0], second[0]), (201, 201))
|
||||
self.assertEqual(first[2], second[2])
|
||||
self.assertEqual(len([m for m in history(self.b["token"], self.chat["id"]) if m["text"] == "один раз"]), 1)
|
||||
|
||||
def test_different_body_same_key_is_409(self):
|
||||
key = uuid.uuid4().hex
|
||||
self.post(self.a, {"text": "первое"}, key)
|
||||
r = self.post(self.a, {"text": "другое"}, key)
|
||||
self.assertEqual(r[0], 409)
|
||||
self.assertEqual(r[2]["error"]["code"], "conflict")
|
||||
|
||||
def test_scope_is_user_and_chat(self):
|
||||
key = uuid.uuid4().hex
|
||||
m1 = self.post(self.a, {"text": "общий ключ"}, key)[2]
|
||||
m2 = self.post(self.b, {"text": "общий ключ"}, key)[2] # другой пользователь
|
||||
other = chat_with(self.a)
|
||||
m3 = self.post(self.a, {"text": "общий ключ"}, key, "/api/v1/chats/%s/messages" % other["id"])[2] # другой чат
|
||||
self.assertEqual(len({m1["id"], m2["id"], m3["id"]}), 3)
|
||||
|
||||
def test_replay_rechecks_session_and_membership(self):
|
||||
key = uuid.uuid4().hex
|
||||
self.post(self.a, {"text": "защищено"}, key)
|
||||
self.assertEqual(self.post(self.eve, {"text": "защищено"}, key)[0], 404) # посторонний
|
||||
r = call("POST", self.path, {"text": "защищено"}, "Z" * 43, headers={"Idempotency-Key": key})
|
||||
self.assertEqual(r[0], 401) # недействительный токен
|
||||
logout = register()
|
||||
chat = chat_with(logout)
|
||||
path = "/api/v1/chats/%s/messages" % chat["id"]
|
||||
self.post(logout, {"text": "x"}, key, path)
|
||||
call("DELETE", "/api/v1/auth/sessions/current", token=logout["token"])
|
||||
self.assertEqual(call("POST", path, {"text": "x"}, logout["token"], headers={"Idempotency-Key": key})[0], 401)
|
||||
|
||||
def test_semantically_equal_body_is_replayed(self):
|
||||
key = uuid.uuid4().hex
|
||||
first = self.post(self.a, {"text": "эквивалент"}, key)
|
||||
raw = call("POST", self.path, raw=b'{ "text" : "\u044d\u043a\u0432\u0438\u0432\u0430\u043b\u0435\u043d\u0442" }',
|
||||
token=self.a["token"], headers={"Idempotency-Key": key, "Content-Type": "application/json"})
|
||||
self.assertEqual((raw[0], raw[2]), (201, first[2]))
|
||||
|
||||
def test_concurrent_same_key_creates_exactly_one_message_across_replicas(self):
|
||||
key = uuid.uuid4().hex
|
||||
with ThreadPoolExecutor(24) as pool:
|
||||
results = list(pool.map(lambda _: self.post(self.a, {"text": "гонка ключа"}, key), range(24)))
|
||||
self.assertEqual({r[0] for r in results}, {201})
|
||||
self.assertEqual(len({r[2]["id"] for r in results}), 1)
|
||||
instances = {r[1]["X-Instance-Id"] for r in results}
|
||||
self.assertGreaterEqual(len(instances), 2, "повторы должны обслуживаться разными репликами: %s" % instances)
|
||||
found = [m for m in history(self.b["token"], self.chat["id"]) if m["text"] == "гонка ключа"]
|
||||
self.assertEqual(len(found), 1)
|
||||
|
||||
def test_expired_key_is_reusable_after_ttl(self):
|
||||
key = uuid.uuid4().hex
|
||||
first = self.post(self.a, {"text": "ttl"}, key)[2]
|
||||
psql("UPDATE idempotency_keys SET created_at = now() - interval '25 hours' WHERE idem_key = '%s'" % key)
|
||||
again = self.post(self.a, {"text": "ttl"}, key)
|
||||
self.assertEqual(again[0], 201)
|
||||
self.assertNotEqual(again[2]["id"], first["id"], "после 24 ч ключ считается новым")
|
||||
|
||||
def test_session_is_shared_between_replicas_without_relogin(self):
|
||||
seen = set()
|
||||
for _ in range(30):
|
||||
r = call("GET", "/api/v1/users/me", token=self.a["token"])
|
||||
self.assertEqual(r[0], 200)
|
||||
seen.add(r[1]["X-Instance-Id"])
|
||||
self.assertGreaterEqual(len(seen), 2, seen)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,142 @@
|
||||
"""Проверка наблюдаемости lab6: метрики, логи, трассы, связь API -> брокер -> worker, отсутствие секретов в телеметрии.
|
||||
python -m unittest discover -s student-tests -p test_observability.py -v (стенд запущен; нужен ~1 минута)
|
||||
Запросы к Prometheus/Loki/Tempo выполняются изнутри сети стенда (их порты наружу не опубликованы).
|
||||
"""
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
import urllib.parse
|
||||
import uuid
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import BASE, ROOT, call, register # noqa: E402
|
||||
from test_attachments import PNG, upload # noqa: E402
|
||||
from test_pipeline import wait_status # noqa: E402
|
||||
|
||||
|
||||
def net_get(url, accept=None):
|
||||
cmd = ["docker", "compose", "--env-file", ".env", "exec", "-T", "prometheus", "wget", "-q", "-O", "-"]
|
||||
if accept:
|
||||
cmd += ["--header", "Accept: " + accept]
|
||||
out = subprocess.run(cmd + [url], cwd=ROOT, capture_output=True, text=True, encoding="utf-8")
|
||||
return out.stdout
|
||||
|
||||
|
||||
def prom(query):
|
||||
data = json.loads(net_get("http://127.0.0.1:9090/api/v1/query?query=" + urllib.parse.quote(query)))
|
||||
return data["data"]["result"]
|
||||
|
||||
|
||||
def loki(query, limit=200):
|
||||
now = int(time.time() * 1e9)
|
||||
url = "http://loki:3100/loki/api/v1/query_range?limit=%d&start=%d&end=%d&query=%s" % (limit, now - 600 * 10 ** 9, now, urllib.parse.quote(query))
|
||||
data = json.loads(net_get(url))
|
||||
return [(s["stream"], v) for s in data["data"]["result"] for v in s["values"]]
|
||||
|
||||
|
||||
def tempo_trace(trace_id):
|
||||
raw = net_get("http://tempo:3200/api/traces/" + trace_id, "application/json")
|
||||
return json.loads(raw) if raw.strip() else None
|
||||
|
||||
|
||||
def spans_of(trace):
|
||||
out = []
|
||||
for batch in trace.get("batches", trace.get("resourceSpans", [])):
|
||||
service = next((a["value"].get("stringValue") for a in batch["resource"]["attributes"] if a["key"] == "service.name"), None)
|
||||
for scope in batch.get("scopeSpans", batch.get("instrumentationLibrarySpans", [])):
|
||||
for sp in scope["spans"]:
|
||||
attrs = {a["key"]: next(iter(a["value"].values())) for a in sp.get("attributes", [])}
|
||||
out.append({"service": service, "name": sp["name"], "id": sp["spanId"], "parent": sp.get("parentSpanId", ""), "kind": sp.get("kind"), "attrs": attrs})
|
||||
return out
|
||||
|
||||
|
||||
class ObservabilityTest(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.a = register("obs")
|
||||
cls.chat = call("POST", "/api/v1/chats", {"title": "наблюдаемость", "member_ids": [cls.a["user"]["id"]]}, cls.a["token"])[2]
|
||||
cls.trace_id = uuid.uuid4().hex
|
||||
cls.secret_text = "СЕКРЕТНЫЙ-ТЕКСТ-" + uuid.uuid4().hex
|
||||
# загрузка с входящим traceparent: так мы знаем trace_id заранее
|
||||
import urllib.request
|
||||
req = urllib.request.Request("%s/api/v1/chats/%s/attachments?filename=obs.png" % (BASE, cls.chat["id"]), data=PNG, method="POST",
|
||||
headers={"Authorization": "Bearer " + cls.a["token"], "Content-Type": "image/png",
|
||||
"traceparent": "00-%s-%s-01" % (cls.trace_id, uuid.uuid4().hex[:16])})
|
||||
with urllib.request.urlopen(req, timeout=30) as r:
|
||||
cls.att = json.loads(r.read())
|
||||
wait_status(cls.a["token"], cls.att["id"], ("ready",), 60)
|
||||
call("POST", "/api/v1/chats/%s/messages" % cls.chat["id"], {"text": cls.secret_text, "attachment_ids": [cls.att["id"]]}, cls.a["token"])
|
||||
deadline = time.time() + 40 # экспорт SDK (2 с) + пакетная обработка Collector + индексация
|
||||
cls.trace = None
|
||||
while time.time() < deadline:
|
||||
cls.trace = tempo_trace(cls.trace_id)
|
||||
if cls.trace and {s["service"] for s in spans_of(cls.trace)} >= {"messenger-api", "messenger-worker"}:
|
||||
break
|
||||
time.sleep(2)
|
||||
|
||||
def test_one_trace_spans_api_outbox_broker_and_worker(self):
|
||||
self.assertIsNotNone(self.trace, "трасса не найдена в Tempo")
|
||||
spans = spans_of(self.trace)
|
||||
by_name = {s["name"]: s for s in spans}
|
||||
self.assertEqual({s["service"] for s in spans} & {"messenger-api", "messenger-worker"}, {"messenger-api", "messenger-worker"})
|
||||
server = by_name["HTTP POST /api/v1/chats/{chat_id}/attachments"]
|
||||
producer, consumer = by_name["attachment.process publish"], by_name["attachment.process process"]
|
||||
self.assertEqual((server["service"], producer["service"], consumer["service"]), ("messenger-api", "messenger-api", "messenger-worker"))
|
||||
# цепочка parent-child: server (API) -> producer (publish из outbox) -> consumer (worker)
|
||||
self.assertEqual(producer["parent"], server["id"])
|
||||
self.assertEqual(consumer["parent"], producer["id"])
|
||||
names = {s["name"] for s in spans}
|
||||
self.assertIn("s3.put_object", names) # запись staging (API) и результатов (worker)
|
||||
self.assertIn("s3.get_object", names) # чтение staging worker-ом
|
||||
self.assertIn("db.connection", names)
|
||||
self.assertEqual(server["attrs"].get("http.route"), "/api/v1/chats/{chat_id}/attachments")
|
||||
self.assertEqual(int(server["attrs"].get("http.response.status_code")), 202)
|
||||
|
||||
def test_logs_are_structured_and_linked_by_trace_id(self):
|
||||
rows = loki('{service_name=~"messenger-.*"} | trace_id="%s"' % self.trace_id)
|
||||
services = {stream["service_name"] for stream, _ in rows}
|
||||
messages = {v[1] for _, v in rows}
|
||||
self.assertIn("messenger-api", services)
|
||||
self.assertIn("messenger-worker", services)
|
||||
self.assertIn("job_ready", messages)
|
||||
self.assertTrue(all(stream.get("span_id") for stream, _ in rows))
|
||||
self.assertTrue(all(stream.get("service_version") is not None and stream.get("service_instance_id") for stream, _ in rows))
|
||||
|
||||
def test_metrics_for_http_jobs_queue_pool_worker(self):
|
||||
for query in ('http_server_request_duration_seconds_count{http_route="/api/v1/chats/{chat_id}/attachments"}',
|
||||
'messenger_jobs_processed_total{outcome="ready"}',
|
||||
'messenger_job_duration_seconds_bucket', 'messenger_job_time_to_ready_seconds_bucket',
|
||||
'messenger_queue_depth', 'messenger_queue_oldest_job_age_seconds', 'messenger_worker_capacity', 'messenger_worker_busy',
|
||||
'db_client_connections_usage{state="used"}', 'db_client_connections_waiting', 'process_memory_rss_bytes',
|
||||
'process_cpu_time_seconds_total'):
|
||||
with self.subTest(query):
|
||||
deadline = time.time() + 40 # экспорт SDK (5 с) + scrape Prometheus (5 с)
|
||||
while not prom(query) and time.time() < deadline:
|
||||
time.sleep(2)
|
||||
self.assertTrue(prom(query), "нет данных: " + query)
|
||||
|
||||
def test_no_high_cardinality_or_secret_labels(self):
|
||||
forbidden = ("user_id", "chat_id", "message_id", "attachment_id", "request_id", "filename")
|
||||
series = json.loads(net_get("http://127.0.0.1:9090/api/v1/series?match[]=%s" % urllib.parse.quote('{__name__=~"http_server.*|messenger_.*|db_client.*|process_.*"}')))["data"]
|
||||
labels = {k for s in series for k in s}
|
||||
self.assertFalse(labels & set(forbidden), labels & set(forbidden))
|
||||
text = json.dumps(series)
|
||||
self.assertNotIn(self.a["token"], text)
|
||||
route_values = {s.get("http_route") for s in series if "http_route" in s}
|
||||
import re
|
||||
self.assertFalse([r for r in route_values if re.search(r"[0-9a-f]{8}-[0-9a-f]{4}-", r)], route_values) # шаблоны, не реальные UUID
|
||||
|
||||
def test_secrets_and_message_bodies_are_not_exported(self):
|
||||
self.assertEqual(loki('{service_name=~"messenger-.*"} |= "%s"' % self.secret_text), [])
|
||||
self.assertEqual(loki('{service_name=~"messenger-.*"} |= "%s"' % self.a["token"]), [])
|
||||
self.assertEqual(loki('{service_name=~"messenger-.*"} |= "%s"' % self.a["password"]), [])
|
||||
raw = json.dumps(self.trace)
|
||||
for secret in (self.secret_text, self.a["token"], self.a["password"]):
|
||||
self.assertNotIn(secret, raw)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,58 @@
|
||||
"""Цепочка трассировки с повторами (уровень 4): все попытки одного задания находятся в ОДНОЙ трассе.
|
||||
python student-tests/test_observability_retry.py
|
||||
Worker пересоздаётся с внедрением сбоя (fail:after_claim, 3 попытки), затем возвращается к штатной конфигурации.
|
||||
"""
|
||||
import json
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
import urllib.request
|
||||
import uuid
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import BASE, call, register # noqa: E402
|
||||
from test_attachments import PNG # noqa: E402
|
||||
from test_observability import loki, spans_of, tempo_trace # noqa: E402
|
||||
from test_pipeline import wait_status # noqa: E402
|
||||
from test_resilience5 import compose, healthy, recreate_worker, wait # noqa: E402
|
||||
|
||||
|
||||
class RetryTraceTest(unittest.TestCase):
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
recreate_worker()
|
||||
|
||||
def test_all_attempts_in_one_trace_and_logs(self):
|
||||
recreate_worker({"WORKER_FAULT_INJECTION": "fail:after_claim", "MAX_ATTEMPTS": "3", "RETRY_BASE_DELAY": "0.3"})
|
||||
a = register("obsr")
|
||||
chat = call("POST", "/api/v1/chats", {"title": "retry-trace", "member_ids": [a["user"]["id"]]}, a["token"])[2]
|
||||
trace_id = uuid.uuid4().hex
|
||||
req = urllib.request.Request("%s/api/v1/chats/%s/attachments?filename=r.png" % (BASE, chat["id"]), data=PNG, method="POST",
|
||||
headers={"Authorization": "Bearer " + a["token"], "Content-Type": "image/png",
|
||||
"traceparent": "00-%s-%s-01" % (trace_id, uuid.uuid4().hex[:16])})
|
||||
with urllib.request.urlopen(req, timeout=30) as r:
|
||||
att = json.loads(r.read())
|
||||
failed = wait_status(a["token"], att["id"], ("failed",), 60)
|
||||
self.assertEqual(failed["error_code"], "retry_exhausted")
|
||||
spans, deadline = [], time.time() + 40
|
||||
while time.time() < deadline:
|
||||
trace = tempo_trace(trace_id)
|
||||
spans = spans_of(trace) if trace else []
|
||||
consumers = [s for s in spans if s["name"] == "attachment.process process"]
|
||||
if len(consumers) >= 3:
|
||||
break
|
||||
time.sleep(2)
|
||||
consumers = sorted([s for s in spans if s["name"] == "attachment.process process"], key=lambda s: int(s["attrs"]["job.attempt"]))
|
||||
publishers = {s["id"]: s for s in spans if s["name"] == "attachment.process publish"}
|
||||
self.assertEqual([int(c["attrs"]["job.attempt"]) for c in consumers], [1, 2, 3])
|
||||
for c in consumers:
|
||||
self.assertIn(c["parent"], publishers, "consumer-спан каждой попытки — потомок publish-спана той же трассы")
|
||||
logs = loki('{service_name="messenger-worker"} | trace_id="%s"' % trace_id)
|
||||
messages = [v[1] for _, v in logs]
|
||||
self.assertEqual(messages.count("job_rescheduled"), 2)
|
||||
self.assertIn("job_retry_exhausted", messages)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,182 @@
|
||||
"""Тесты асинхронного конвейера вложений lab6 (контракт вариантов, состояния, права). Запуск:
|
||||
python -m unittest discover -s student-tests -p "test_pipeline.py" -v
|
||||
Пиксельные проверки выполняются, если установлен Pillow (иначе пропускаются); размеры проверяются по заголовкам файлов.
|
||||
"""
|
||||
import hashlib
|
||||
import io
|
||||
import os
|
||||
import struct
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
import uuid
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import call, register # noqa: E402
|
||||
from test_attachments import FIX, JPG, PNG, download, upload # noqa: E402
|
||||
|
||||
try:
|
||||
from PIL import Image
|
||||
except ImportError: # pragma: no cover
|
||||
Image = None
|
||||
|
||||
|
||||
def fixture(name):
|
||||
return open(os.path.join(FIX, name), "rb").read()
|
||||
|
||||
|
||||
def jpeg_size(data):
|
||||
i = 2
|
||||
while i < len(data):
|
||||
while data[i] == 0xFF:
|
||||
i += 1
|
||||
marker = data[i]
|
||||
i += 1
|
||||
if marker in (0xD8, 0x01) or 0xD0 <= marker <= 0xD7:
|
||||
continue
|
||||
length = int.from_bytes(data[i:i + 2], "big")
|
||||
if marker in (0xC0, 0xC1, 0xC2):
|
||||
return int.from_bytes(data[i + 5:i + 7], "big"), int.from_bytes(data[i + 3:i + 5], "big")
|
||||
i += length
|
||||
raise AssertionError("JPEG dimensions not found")
|
||||
|
||||
|
||||
def webp_size(data):
|
||||
assert data[:4] == b"RIFF" and data[8:12] == b"WEBP", "not WebP"
|
||||
kind = data[12:16]
|
||||
if kind == b"VP8X":
|
||||
return 1 + int.from_bytes(data[24:27], "little"), 1 + int.from_bytes(data[27:30], "little")
|
||||
if kind == b"VP8L":
|
||||
bits = int.from_bytes(data[21:25], "little")
|
||||
return (bits & 0x3FFF) + 1, ((bits >> 14) & 0x3FFF) + 1
|
||||
if kind == b"VP8 ":
|
||||
return struct.unpack("<HH", data[26:30])[0] & 0x3FFF, struct.unpack("<HH", data[26:30])[1] & 0x3FFF
|
||||
raise AssertionError("unknown WebP chunk")
|
||||
|
||||
|
||||
def wait_status(token, att_id, wanted=("ready",), seconds=60):
|
||||
deadline = time.time() + seconds
|
||||
while time.time() < deadline:
|
||||
code, _, body = download(token, att_id, "")
|
||||
att = __import__("json").loads(body)
|
||||
if att["status"] in wanted:
|
||||
return att
|
||||
time.sleep(0.3)
|
||||
raise AssertionError("вложение не достигло %s за %d с: %s" % (wanted, seconds, att["status"]))
|
||||
|
||||
|
||||
class PipelineTest(unittest.TestCase):
|
||||
def setUp(self):
|
||||
self.a, self.b, self.eve = register("pipe"), register("pipe"), register("pipe")
|
||||
self.chat = call("POST", "/api/v1/chats", {"title": "конвейер", "member_ids": [self.a["user"]["id"], self.b["user"]["id"]]}, self.a["token"])[2]
|
||||
|
||||
def ready(self, data, ctype="image/png", name="x.png"):
|
||||
status, _, att = upload(self.a["token"], self.chat["id"], data, ctype, name)
|
||||
self.assertEqual(status, 202, att)
|
||||
self.assertEqual((att["status"], att["variants"], att["error_code"]), ("queued", [], None))
|
||||
return wait_status(self.b["token"], att["id"])
|
||||
|
||||
def variants(self, att):
|
||||
return {v["name"]: v for v in att["variants"]}
|
||||
|
||||
def fetch(self, att, variant):
|
||||
code, hdr, body = download(self.b["token"], att["id"], "/content?variant=" + variant)
|
||||
self.assertEqual(code, 200)
|
||||
return hdr.get_content_type(), body
|
||||
|
||||
def test_sample_png_produces_all_variants_per_contract(self):
|
||||
att = self.ready(PNG)
|
||||
v = self.variants(att)
|
||||
self.assertEqual(set(v), {"thumbnail", "crop", "optimized"})
|
||||
self.assertEqual((att["sha256"], att["size_bytes"]), (hashlib.sha256(PNG).hexdigest(), len(PNG))) # описывают оригинал
|
||||
self.assertEqual((v["thumbnail"]["width"], v["thumbnail"]["height"], v["thumbnail"]["content_type"]), (256, 160, "image/jpeg"))
|
||||
self.assertEqual((v["crop"]["width"], v["crop"]["height"], v["crop"]["content_type"]), (128, 128, "image/jpeg"))
|
||||
self.assertEqual(v["optimized"]["content_type"], "image/webp")
|
||||
self.assertEqual((v["optimized"]["width"], v["optimized"]["height"]), (320, 200)) # без увеличения
|
||||
ctype, thumb = self.fetch(att, "thumbnail")
|
||||
self.assertEqual((ctype, jpeg_size(thumb)), ("image/jpeg", (256, 160)))
|
||||
self.assertEqual((v["thumbnail"]["size_bytes"], v["thumbnail"]["sha256"]), (len(thumb), hashlib.sha256(thumb).hexdigest()))
|
||||
_, crop = self.fetch(att, "crop")
|
||||
self.assertEqual(jpeg_size(crop), (128, 128))
|
||||
ctype, opt = self.fetch(att, "optimized")
|
||||
self.assertEqual((ctype, webp_size(opt)), ("image/webp", (320, 200)))
|
||||
code, hdr, original = download(self.b["token"], att["id"])
|
||||
self.assertEqual((code, original, hdr.get_content_type()), (200, PNG, "image/png")) # оригинал неизменен
|
||||
|
||||
def test_exif_orientation_is_applied(self):
|
||||
att = self.ready(fixture("exif_rot6.jpg"), "image/jpeg", "rot.jpg")
|
||||
v = self.variants(att)
|
||||
# 300x100 с Orientation=6 -> 100x300 -> вписывается в 256: 85x256 (округление вниз)
|
||||
self.assertEqual((v["thumbnail"]["width"], v["thumbnail"]["height"]), (85, 256))
|
||||
self.assertEqual(jpeg_size(self.fetch(att, "thumbnail")[1]), (85, 256))
|
||||
|
||||
def test_small_image_is_not_upscaled_but_crop_is_128(self):
|
||||
att = self.ready(fixture("small.png"), "image/png", "small.png")
|
||||
v = self.variants(att)
|
||||
self.assertEqual((v["thumbnail"]["width"], v["thumbnail"]["height"]), (50, 30))
|
||||
self.assertEqual((v["crop"]["width"], v["crop"]["height"]), (128, 128))
|
||||
self.assertEqual((v["optimized"]["width"], v["optimized"]["height"]), (50, 30))
|
||||
|
||||
@unittest.skipIf(Image is None, "нужен Pillow")
|
||||
def test_alpha_is_composited_on_white(self):
|
||||
att = self.ready(fixture("alpha.png"), "image/png", "alpha.png")
|
||||
thumb = Image.open(io.BytesIO(self.fetch(att, "thumbnail")[1])).convert("RGB")
|
||||
left, right = thumb.getpixel((10, thumb.height // 2)), thumb.getpixel((thumb.width - 10, thumb.height // 2))
|
||||
self.assertTrue(all(c > 235 for c in left), "прозрачная область должна стать белой: %s" % (left,))
|
||||
self.assertTrue(right[0] > 200 and right[1] < 60, "непрозрачная красная область сохранена: %s" % (right,))
|
||||
|
||||
def test_message_can_reference_attachment_before_ready_and_failed_cannot(self):
|
||||
status, _, att = upload(self.a["token"], self.chat["id"], PNG)
|
||||
path = "/api/v1/chats/%s/messages" % self.chat["id"]
|
||||
sent = call("POST", path, {"text": "", "attachment_ids": [att["id"]]}, self.a["token"])
|
||||
self.assertEqual((status, sent[0], sent[2]["attachment_ids"]), (202, 201, [att["id"]]))
|
||||
poison = upload(self.a["token"], self.chat["id"], fixture("poison.png"))
|
||||
self.assertEqual(poison[0], 202, poison[2])
|
||||
failed = wait_status(self.b["token"], poison[2]["id"], ("failed",))
|
||||
self.assertEqual((failed["status"], failed["error_code"], failed["variants"]), ("failed", "invalid_image", []))
|
||||
self.assertEqual(call("POST", path, {"text": "x", "attachment_ids": [failed["id"]]}, self.a["token"])[0], 409)
|
||||
self.assertEqual(download(self.b["token"], failed["id"], "/content")[0], 409)
|
||||
self.assertEqual(download(self.b["token"], failed["id"], "/content?variant=thumbnail")[0], 409)
|
||||
|
||||
def test_content_is_409_while_not_ready_and_original_is_untouched(self):
|
||||
status, _, att = upload(self.a["token"], self.chat["id"], PNG)
|
||||
code = download(self.b["token"], att["id"], "/content")[0]
|
||||
self.assertIn(code, (200, 409)) # быстрый worker мог успеть: 200 допустим, но не 404/500
|
||||
done = wait_status(self.b["token"], att["id"])
|
||||
self.assertEqual(download(self.b["token"], done["id"], "/content")[2], PNG)
|
||||
|
||||
def test_other_variants_visible_only_to_chat_members(self):
|
||||
att = self.ready(PNG)
|
||||
for variant in ("thumbnail", "crop", "optimized", "original"):
|
||||
self.assertEqual(download(self.eve["token"], att["id"], "/content?variant=" + variant)[0], 404)
|
||||
self.assertEqual(download(self.b["token"], att["id"], "/content?variant=bogus")[0], 400)
|
||||
|
||||
def test_ready_job_redelivery_does_not_duplicate_variants(self):
|
||||
from test_api import psql
|
||||
att = self.ready(PNG)
|
||||
before = psql("SELECT count(*) FROM attachment_variants WHERE attachment_id = '%s'" % att["id"])
|
||||
self.assertEqual(before, "3")
|
||||
psql("INSERT INTO outbox (job_id, attachment_id, pipeline_version, attempt) VALUES (gen_random_uuid(), '%s', 1, 2)" % att["id"])
|
||||
time.sleep(4)
|
||||
self.assertEqual(psql("SELECT count(*) FROM attachment_variants WHERE attachment_id = '%s'" % att["id"]), "3")
|
||||
self.assertEqual(wait_status(self.b["token"], att["id"])["status"], "ready")
|
||||
|
||||
def test_staging_is_cleaned_after_ready_and_kept_for_failed(self):
|
||||
import subprocess
|
||||
from test_api import ROOT, psql
|
||||
att = self.ready(PNG)
|
||||
poison = upload(self.a["token"], self.chat["id"], fixture("poison.png"))[2]
|
||||
wait_status(self.b["token"], poison["id"], ("failed",))
|
||||
|
||||
def exists(att_id):
|
||||
code = "\n".join(["import asyncio", "from app.s3 import S3", "async def m():",
|
||||
" s = S3(); await s.start(); print((await s.head('staging/%s')) is not None); await s.stop()" % att_id,
|
||||
"asyncio.run(m())"])
|
||||
return subprocess.run(["docker", "compose", "--env-file", ".env", "exec", "-T", "api", "python", "-c", code], cwd=ROOT,
|
||||
capture_output=True, text=True).stdout.strip()
|
||||
self.assertEqual(exists(att["id"]), "False", "staging готового вложения должен быть удалён после COMMIT результата")
|
||||
self.assertEqual(exists(poison["id"]), "True", "staging неудавшегося вложения хранится для разбора/replay (retention)")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,154 @@
|
||||
"""Отказы конвейера lab6 (окна отказа из contracts/attachment-states.md). Тесты управляют контейнерами (volumes сохраняются):
|
||||
python student-tests/test_resilience5.py # ~5-7 минут
|
||||
|
||||
Окна: (1) worker остановлен -> upload -> рестарт API -> worker поднят; (2) рестарт брокера с volume; (3) брокер недоступен
|
||||
при приёме (outbox); (4) crash API между COMMIT и publish; (5) crash worker после записи в S3 до COMMIT результата;
|
||||
(6) crash после COMMIT до ack; (7) исчерпание попыток и операторский replay.
|
||||
"""
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
|
||||
sys.path.insert(0, os.path.dirname(__file__))
|
||||
from test_api import ROOT, call, psql, register # noqa: E402
|
||||
from test_attachments import PNG, download, upload # noqa: E402
|
||||
from test_pipeline import wait_status # noqa: E402
|
||||
|
||||
|
||||
def compose(*args, env=None, check=True):
|
||||
return subprocess.run(["docker", "compose", "--env-file", ".env", *args], cwd=ROOT, capture_output=True, text=True,
|
||||
env=dict(os.environ, **(env or {})), check=check).stdout
|
||||
|
||||
|
||||
def wait(predicate, seconds=90, message="условие"):
|
||||
deadline = time.time() + seconds
|
||||
while time.time() < deadline:
|
||||
try:
|
||||
if predicate():
|
||||
return
|
||||
except Exception:
|
||||
pass
|
||||
time.sleep(1)
|
||||
raise AssertionError("%s не выполнено за %d с" % (message, seconds))
|
||||
|
||||
|
||||
def healthy(service):
|
||||
ids = compose("ps", "-q", service).split()
|
||||
return bool(ids) and all(subprocess.run(["docker", "inspect", "-f", "{{.State.Health.Status}}", i], capture_output=True, text=True).stdout.strip() == "healthy" for i in ids)
|
||||
|
||||
|
||||
def api_ready():
|
||||
return call("GET", "/health/ready")[0] == 200
|
||||
|
||||
|
||||
def recreate_worker(env=None):
|
||||
compose("up", "-d", "--no-deps", "--force-recreate", "worker", env=env)
|
||||
wait(lambda: healthy("worker"), 90, "worker healthy")
|
||||
|
||||
|
||||
class ResilienceTest(unittest.TestCase):
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
recreate_worker()
|
||||
compose("up", "-d", "--no-deps", "rabbitmq", "api")
|
||||
|
||||
def setUp(self):
|
||||
compose("start", "rabbitmq", "postgres", "minio", "valkey", check=False)
|
||||
wait(lambda: healthy("rabbitmq") and healthy("postgres"), 90, "rabbitmq/postgres healthy")
|
||||
wait(api_ready, 60, "api ready")
|
||||
self.a = register("res")
|
||||
self.chat = call("POST", "/api/v1/chats", {"title": "отказы", "member_ids": [self.a["user"]["id"]]}, self.a["token"])[2]
|
||||
|
||||
def accept(self, data=PNG):
|
||||
status, _, att = upload(self.a["token"], self.chat["id"], data)
|
||||
self.assertEqual(status, 202, att)
|
||||
return att
|
||||
|
||||
def done(self, att, seconds=90):
|
||||
ready = wait_status(self.a["token"], att["id"], ("ready",), seconds)
|
||||
self.assertEqual(psql("SELECT count(*) FROM attachment_variants WHERE attachment_id = '%s'" % att["id"]), "3") # без дублей
|
||||
code, _, body = download(self.a["token"], att["id"])
|
||||
self.assertEqual((code, body), (200, PNG))
|
||||
return ready
|
||||
|
||||
def test_1_worker_down_upload_api_restart_worker_up(self):
|
||||
compose("stop", "worker")
|
||||
att = self.accept()
|
||||
sent = call("POST", "/api/v1/chats/%s/messages" % self.chat["id"], {"text": "", "attachment_ids": [att["id"]]}, self.a["token"])
|
||||
self.assertEqual(sent[0], 201)
|
||||
time.sleep(3)
|
||||
self.assertEqual(wait_status(self.a["token"], att["id"], ("queued",), 5)["status"], "queued") # не обрабатывается без worker
|
||||
compose("restart", "api")
|
||||
wait(api_ready, 60, "api ready")
|
||||
compose("start", "worker")
|
||||
self.done(att)
|
||||
|
||||
def test_2_broker_restart_keeps_accepted_jobs(self):
|
||||
compose("stop", "worker")
|
||||
att = self.accept()
|
||||
time.sleep(2)
|
||||
compose("restart", "rabbitmq")
|
||||
wait(lambda: healthy("rabbitmq"), 90, "rabbitmq healthy")
|
||||
compose("start", "worker")
|
||||
self.done(att)
|
||||
|
||||
def test_3_broker_down_during_acceptance_outbox_delivers_later(self):
|
||||
compose("stop", "rabbitmq")
|
||||
att = self.accept() # 202 даже без брокера: намерение зафиксировано в outbox
|
||||
self.assertEqual(psql("SELECT count(*) FROM outbox WHERE attachment_id = '%s' AND published_at IS NULL" % att["id"]), "1")
|
||||
time.sleep(3)
|
||||
self.assertEqual(wait_status(self.a["token"], att["id"], ("queued",), 5)["status"], "queued")
|
||||
compose("start", "rabbitmq")
|
||||
self.done(att)
|
||||
self.assertEqual(psql("SELECT count(*) FROM outbox WHERE attachment_id = '%s' AND published_at IS NULL" % att["id"]), "0")
|
||||
|
||||
def test_4_api_crash_between_commit_and_publish(self):
|
||||
compose("stop", "rabbitmq")
|
||||
att = self.accept() # COMMIT выполнен, publish невозможен
|
||||
for cid in compose("ps", "-q", "api").split():
|
||||
subprocess.run(["docker", "kill", cid], capture_output=True) # внезапная смерть всех реплик API
|
||||
compose("start", "rabbitmq")
|
||||
wait(lambda: healthy("rabbitmq"), 90, "rabbitmq healthy")
|
||||
compose("up", "-d", "--no-deps", "api")
|
||||
wait(api_ready, 90, "api ready")
|
||||
self.done(att) # relay (в worker/новом API) довёл задание без ручного вмешательства
|
||||
|
||||
def test_5_worker_crash_after_s3_before_commit_lease_recovery(self):
|
||||
recreate_worker({"WORKER_FAULT_INJECTION": "crash_once:after_s3_before_commit", "LEASE_SECONDS": "8"})
|
||||
att = self.accept()
|
||||
ready = self.done(att, 120)
|
||||
events = psql("SELECT string_agg(event, ',' ORDER BY id) FROM job_events WHERE attachment_id = '%s'" % att["id"])
|
||||
self.assertIn("lease_expired", events)
|
||||
self.assertIn("ready", events)
|
||||
self.assertEqual(ready["status"], "ready")
|
||||
|
||||
def test_6_worker_crash_after_commit_before_ack_redelivery_is_noop(self):
|
||||
recreate_worker({"WORKER_FAULT_INJECTION": "crash_once:after_commit_before_ack"})
|
||||
att = self.accept()
|
||||
self.done(att, 120)
|
||||
time.sleep(5) # повторная доставка: no-op с ack, вариантов по-прежнему 3
|
||||
self.assertEqual(psql("SELECT count(*) FROM attachment_variants WHERE attachment_id = '%s'" % att["id"]), "3")
|
||||
self.assertEqual(psql("SELECT status FROM attachments WHERE id = '%s'" % att["id"]), "ready")
|
||||
|
||||
def test_7_retry_exhaustion_registry_and_operator_replay(self):
|
||||
recreate_worker({"WORKER_FAULT_INJECTION": "fail:after_claim", "MAX_ATTEMPTS": "3", "RETRY_BASE_DELAY": "0.3"})
|
||||
att = self.accept()
|
||||
failed = wait_status(self.a["token"], att["id"], ("failed",), 60)
|
||||
self.assertEqual((failed["error_code"], failed["variants"]), ("retry_exhausted", []))
|
||||
events = psql("SELECT string_agg(event || ':' || coalesce(attempt::text,''), ',' ORDER BY id) FROM job_events WHERE attachment_id = '%s'" % att["id"])
|
||||
self.assertEqual(events.count("retry_scheduled"), 2, events)
|
||||
registry = json.loads(compose("run", "--rm", "-T", "worker", "python", "-m", "app.replay", "list"))
|
||||
self.assertIn(att["id"], [r["attachment_id"] for r in registry])
|
||||
recreate_worker() # причина устранена
|
||||
result = json.loads(compose("run", "--rm", "-T", "worker", "python", "-m", "app.replay", "replay", att["id"]))
|
||||
self.assertEqual(result["result"], "requeued")
|
||||
self.done(att)
|
||||
again = json.loads(compose("run", "--rm", "-T", "worker", "python", "-m", "app.replay", "replay", att["id"]))
|
||||
self.assertEqual(again["result"], "skipped") # replay не исполняет дубль уже успешной работы
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,59 @@
|
||||
"""Проверка топологии стенда (уровень 4): число реплик, отсутствие публикации портов API, лимиты ресурсов, healthcheck,
|
||||
отсутствие container_name, единая публичная точка входа. Запуск: python student-tests/test_topology.py"""
|
||||
import json
|
||||
import subprocess
|
||||
import unittest
|
||||
|
||||
PROJECT = "mtusi-lab6"
|
||||
|
||||
|
||||
def sh(*args):
|
||||
return subprocess.run(args, capture_output=True, text=True, encoding="utf-8").stdout.strip()
|
||||
|
||||
|
||||
def containers(service):
|
||||
ids = sh("docker", "ps", "-q", "--filter", "label=com.docker.compose.project=" + PROJECT,
|
||||
"--filter", "label=com.docker.compose.service=" + service).split()
|
||||
return [json.loads(sh("docker", "inspect", i))[0] for i in ids]
|
||||
|
||||
|
||||
class TopologyTest(unittest.TestCase):
|
||||
def test_at_least_two_api_replicas_without_published_ports(self):
|
||||
api = containers("api")
|
||||
self.assertGreaterEqual(len(api), 2)
|
||||
for c in api:
|
||||
self.assertFalse([p for p in c["HostConfig"]["PortBindings"] or {}], "реплика API публикует порт в обход LB")
|
||||
self.assertEqual(c["State"]["Health"]["Status"], "healthy")
|
||||
|
||||
def test_only_lb_is_published_and_only_on_localhost(self):
|
||||
published = {}
|
||||
for service in ("api", "postgres", "valkey", "lb", "seaweedfs", "rabbitmq", "otel-collector", "prometheus", "loki", "tempo", "grafana"):
|
||||
for c in containers(service):
|
||||
for port, binds in (c["HostConfig"]["PortBindings"] or {}).items():
|
||||
published.setdefault(service, []).extend((port, b["HostIp"], b["HostPort"]) for b in binds)
|
||||
self.assertEqual(sorted(published), ["grafana", "lb"], published) # lab6: LB и Grafana (UI), OTLP/метрики/админки наружу не открыты
|
||||
self.assertEqual(published["lb"], [("8080/tcp", "127.0.0.1", "8080")])
|
||||
self.assertEqual(published["grafana"], [("3000/tcp", "127.0.0.1", "3000")])
|
||||
|
||||
def test_resource_limits_are_set(self):
|
||||
for service in ("api", "postgres", "valkey", "lb", "seaweedfs", "rabbitmq", "otel-collector", "prometheus", "loki", "tempo", "grafana"):
|
||||
for c in containers(service):
|
||||
self.assertGreater(c["HostConfig"]["Memory"], 0, service)
|
||||
self.assertGreater(c["HostConfig"]["NanoCpus"], 0, service)
|
||||
|
||||
def test_no_fixed_container_names(self):
|
||||
for c in containers("api"):
|
||||
labels = c["Config"]["Labels"]
|
||||
self.assertEqual(c["Name"].lstrip("/"), "%s-api-%s" % (PROJECT, labels["com.docker.compose.container-number"]))
|
||||
|
||||
def test_api_runs_as_non_root(self):
|
||||
for c in containers("api"):
|
||||
self.assertEqual(c["Config"]["User"], "10001:10001")
|
||||
|
||||
def test_stop_grace_period_covers_drain(self):
|
||||
for c in containers("api"):
|
||||
self.assertGreaterEqual(c["Config"].get("StopTimeout", 10), 15)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||