ЛР 3 — БВТ2301 — Маслов — на 5 #153

Closed
wh1tet3a wants to merge 9 commits from wh1tet3a/mtusi2026lab3:solution/bvt2301-maslov into main
113 changed files with 390689 additions and 38 deletions
+13
View File
@@ -18,3 +18,16 @@ __pycache__
node_modules
evidence
.DS_Store
**/__pycache__
**/*.py[cod]
.work
.pytest_cache
.ruff_cache
tests
student-tests
research
scripts
contracts
*.md
Makefile
compose.yaml
+5
View File
@@ -11,3 +11,8 @@ SESSION_STORE=postgres
CACHE_URL_FILE=/run/secrets/cache_url
API_REPLICAS=2
IDEMPOTENCY_TTL_SECONDS=86400
PUBLIC_PORT=8083
API_CPUS=0.5
API_MEMORY=384m
DB_POOL_SIZE=10
+15 -14
View File
@@ -1,18 +1,19 @@
## Лабораторная 3
Автор / группа: TODO
Целевая оценка (3/4/5): TODO
Трек 5, если заявлен: TODO
Автор / группа: Маслов / БВТ2301
Целевая оценка: 5
Трек: цена масштабирования.
Что реализовано и что перенесено: TODO
Команды запуска из чистого clone: TODO
Результаты проверок и ссылка на REPORT.md: TODO
Ограничения / известные ошибки: TODO
Источники и использование ИИ: TODO
Собственная реализация ЛР2 перенесена поверх skeleton ЛР3: две API за HAProxy, общие PG-сессии, транзакционная идемпотентность24ч, реальные failover/rolling/persistence hooks. Подтверждено устранение повторного SELECT ACL в чтении истории.
- [ ] Весь обязательный минимум текущей и предыдущих работ сохранён с объявленными переходами.
- [ ] Контракт и публичные тесты не ослаблены.
- [ ] make check и make test выполнены; приложен реальный вывод.
- [ ] Сценарии отказов/безопасности своего уровня воспроизведены.
- [ ] Все секреты и приватные ключи находятся вне Git; отчёт не содержит tokens.
- [ ] README задания сохранён, REPORT.md содержит инструкцию именно моего решения.
Запуск: `uv sync --locked`, `make bootstrap`, `VCS_REF=$(git rev-parse HEAD) docker compose up -d --build --wait`. Порт8083. Проверки и все команды: [REPORT.md](REPORT.md).
Clean clone `ac5c30c`:88 HTTPtests, lint/check/validate/smoke, отказы обеих API, rolling, restart и runtime hashes — PASS.54 исследовательских прогона,59674 успешных HTTP,0 ошибок измерений. Отдельно документированы краткие ошибки при переключениях; zero-error rolling не заявляется. Один Docker-host/LB/PG остаются SPOF; HTTP только loopback. Удалённый CI отдельно не подтверждён.
Источники: официальные HAProxy/PostgreSQL/Compose, библиотеки по uv.lock. Помощь ИИ использована при разборе требований, подготовке реализации и проверок; выводы подтверждены собственными запусками.
- [x] Весь обязательный минимум и все пункты4 выполнены.
- [x] Контракт и публичные tests сохранены.
- [x] Исходный README сохранён; REPORT содержит запуск и доказательства.
- [x] Отказы, исследование и чистый clone воспроизведены.
- [x] Секреты находятся вне Git и runtime-образа.
+23
View File
@@ -0,0 +1,23 @@
name: Application integration
on: [push, pull_request, workflow_dispatch]
permissions:
contents: read
jobs:
application:
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
- uses: astral-sh/setup-uv@v6
with:
version: 0.13.7
- run: uv sync --locked
- run: make bootstrap check validate lint PYTHON=.venv/bin/python
- run: VCS_REF=$(git rev-parse HEAD) docker compose up -d --build --wait
- run: make test test-student PYTHON=.venv/bin/python BASE_URL=http://127.0.0.1:8083
env:
NO_PROXY: localhost,127.0.0.1
no_proxy: localhost,127.0.0.1
- run: make verify VERIFY_OUTPUT=evidence/cluster-ci.json
- if: always()
run: docker compose down --volumes --remove-orphans
+6
View File
@@ -30,3 +30,9 @@ coverage/
.vscode/
# Keep small, sanitized evidence summaries in evidence/ under version control.
# Inspect git diff --cached: ignore rules are not a secret scanner.
.work/
.pytest_cache/
.ruff_cache/
.work-*.txt
+1
View File
@@ -0,0 +1 @@
3.14.7
+22
View File
@@ -0,0 +1,22 @@
FROM ghcr.io/astral-sh/uv:0.13.7@sha256:96f40d3904aad8d0bd010539001d36c8e2188b981e1334471a90c4a1d93bb490 AS uv
FROM python:3.14.7-slim-trixie@sha256:51dafde81dbdb6ebde285137a295cf18a47ca95234fe388a343719cb97305b3d AS builder
COPY --from=uv /uv /usr/local/bin/uv
WORKDIR /app
ENV UV_PYTHON_DOWNLOADS=never UV_LINK_MODE=copy
COPY pyproject.toml uv.lock ./
RUN uv sync --locked --no-dev --no-install-project
FROM python:3.14.7-slim-trixie@sha256:51dafde81dbdb6ebde285137a295cf18a47ca95234fe388a343719cb97305b3d AS runtime
ARG VCS_REF=unknown
LABEL org.opencontainers.image.revision=$VCS_REF
WORKDIR /app
ENV PATH="/app/.venv/bin:$PATH" PYTHONUNBUFFERED=1 PYTHONDONTWRITEBYTECODE=1 APP_HOST=0.0.0.0 APP_PORT=8080
RUN groupadd --gid 10001 app && useradd --uid 10001 --gid app --no-create-home app
COPY --from=builder /app/.venv /app/.venv
COPY src /app/src
COPY migrations /app/migrations
USER 10001:10001
EXPOSE 8080
HEALTHCHECK --interval=10s --timeout=2s --start-period=10s --start-interval=1s --retries=3 CMD ["python", "-c", "import os,urllib.request; urllib.request.urlopen('http://127.0.0.1:'+os.getenv('APP_PORT','8080')+'/health/ready',timeout=1).read()"]
STOPSIGNAL SIGTERM
CMD ["python", "-m", "src"]
+24
View File
@@ -28,3 +28,27 @@ 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)
UV ?= uv
VERIFY_OUTPUT ?= evidence/cluster-repeat.json
RESEARCH_OUTPUT ?= evidence/scaling-repeat
.PHONY: bootstrap up down image lint test-student verify research rolling-update
bootstrap:
$(PYTHON) scripts/bootstrap.py
up: bootstrap
VCS_REF=$$(git rev-parse HEAD) docker compose up -d --build --wait
down:
docker compose down
image:
docker build --build-arg VCS_REF=$$(git rev-parse HEAD) -t mtusi-messenger:lab3 .
lint:
$(UV) run --locked ruff format src student-tests research scripts/bootstrap.py scripts/stand.py scripts/cluster.py scripts/verify_*.py
$(UV) run --locked ruff check src student-tests research scripts/bootstrap.py scripts/stand.py scripts/cluster.py scripts/verify_*.py
test-student:
BASE_URL="$(BASE_URL)" $(UV) run --locked pytest student-tests
verify:
$(UV) run --locked python -m scripts.verify_cluster --output "$(VERIFY_OUTPUT)"
research:
$(UV) run --locked python -m research.run --output "$(RESEARCH_OUTPUT)"
rolling-update:
scripts/rolling-update.sh
+141 -24
View File
@@ -1,37 +1,154 @@
# Отчёт — лабораторная 3
# ЛР 3 — балансировка и отказ API-реплики
- Автор, группа: TODO
- Целевая оценка; выбранный трек 5, если нужен: TODO
- Commit/PR: TODO после публикации
- Использованные источники, библиотеки и помощь ИИ: TODO
Автор: **Маслов, БВТ2301**. Цель: **5, трек 1 — цена масштабирования**.
Площадка сдачи: Gitea, личная ветка `solution/bvt2301-maslov`.
## Запуск из чистого clone
Открыт [PR153](https://gitea.privcloud.tech/mtusi2026/mtusi2026lab3/pulls/153) из публичного [собственного fork](https://gitea.privcloud.tech/wh1tet3a/mtusi2026lab3). Удалённый commit status: pending,0checks; это отсутствие результатов CI, локальный PASS отдельно.
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.
## Запуск
## Архитектура и решения
Нужны Docker с Compose v2+ и Python 3.14 с uv; публичные преподавательские проверки работают также на Python 3.9+. Все команды выполняются в корне этого репозитория; соседние лабораторные не нужны.
TODO: схема процессов/хранилищ, источник истины, транзакционные границы, ограничения. Что перенесено из lab2, что изменено в этой работе. Соответствие переменных из .env.example реальной конфигурации.
```sh
uv sync --locked
make bootstrap
VCS_REF=$(git rev-parse HEAD) docker compose up -d --build --wait
make check validate PYTHON=.venv/bin/python
NO_PROXY=localhost,127.0.0.1 make test BASE_URL=http://127.0.0.1:8083
make test-student BASE_URL=http://127.0.0.1:8083
make verify
```
## Доказательства критериев
Bootstrap создаёт независимые случайные credentials в `secrets/` и `.env`, существующие файлы сохраняет. Каталог secrets имеет права 0700; файлы читаются непривилегированным UID10001 через read-only mounts. Секреты не находятся в Git или образе. Начальная база создаётся образом PostgreSQL; отдельный одноразовый `migrate` применяет SQL под advisory lock. API стартуют после успешной миграции. Изменённая применённая миграция отклоняется по SHA-256.
| 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 |
Основной адрес — `http://127.0.0.1:8083`. Порт можно изменить через `PUBLIC_PORT`; API и хранилища host ports не публикуют. В `.env.example` `API_REPLICAS` задаёт Compose scale, `API_CPUS` и `API_MEMORY` — квоты каждой API, `DB_POOL_SIZE` — максимум соединений каждого процесса. `SESSION_STORE=postgres` описывает выбранную архитектуру; переключателя на локальные сессии нет. Срок идемпотентности фиксирован контрактом в 86400 секунд, его уменьшение конфигурацией не поддерживается. `APP_ENV` — поясняющее поле skeleton, код его не использует.
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.
```sh
# Одна выбранная реплика остаётся остановленной:
NO_PROXY=localhost,127.0.0.1 make test-failover ACTION_SCRIPT=scripts/stop-one.sh BASE_URL=http://127.0.0.1:8083
# Явно вернуть требуемое число реплик:
docker compose up -d --no-build --wait
# Перезапуск API, PostgreSQL, Valkey и LB с сохранением volumes:
NO_PROXY=localhost,127.0.0.1 make test-persistence ACTION_SCRIPT=scripts/restart.sh BASE_URL=http://127.0.0.1:8083
# Поочерёдно заменить все API, добавляя готовую реплику перед остановкой старой:
make rolling-update
# Безопасная остановка с сохранением данных:
docker compose stop
# Удалить контейнеры и сеть, сохранив данные:
make down
# Только намеренный сброс учебных данных ЭТОГО проекта:
docker compose down --volumes
```
## Эксперименты и отказы
Тесты создают синтетических пользователей и сообщения. Нагрузочные и аварийные сценарии запускаются явно. Для нового измерения выбирать новое имя output, сохраняя предыдущие результаты.
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.
## Архитектура
| Воздействие | Что ожидаем | Что наблюдали | Время восстановления / потеря данных |
| --- | --- | --- | --- |
| TODO | TODO | TODO | TODO |
```mermaid
flowchart LR
C[Клиент] --> LB[HAProxy]
LB --> A[API процесс 1]
LB --> B[API процесс 2]
A --> PG[(PostgreSQL)]
B --> PG
A --> V[(Valkey: disposable User cache)]
B --> V
```
## Ограничения и следующие шаги
Перенесена собственная реализация ЛР2: строгая валидация UTF-8 JSON и типов, ACL, keyset pagination `(created_at,id)`, Argon2id, случайные Bearer-сессии с абсолютным expiry, PostgreSQL и ограниченный по времени Valkey-кэш карточек User. Преподавательские README, COURSE, contracts и tests сохранены. Пароль не сохраняется открытым; для сессии хранится SHA-256 случайного 256-битного токена. Каждый защищённый запрос сначала проверяет действительность сессии в PostgreSQL. Кэш не может восстановить отозванную сессию или разрешить доступ при отказе БД.
TODO: какие отказы выдерживаются и какие нет; что не реализовано; где учебное упрощение; какое улучшение подтверждено, а какое пока гипотеза.
Новые части: `X-Instance-Id` — UUID alias, генерируемый один раз внутри процесса API; HAProxy его не подменяет. Round robin применяется без sticky sessions. Одна сессия обслуживается любым процессом. Порты API доступны только внутри сети Compose; `container_name` не используется. Один uvicorn worker на контейнер, масштабирование — числом контейнеров.
HAProxy 3.2.1 закреплён digest, использует Docker DNS 127.0.0.11 и server-template до 8 адресов. DNS hold valid/obsolete/nx — 1 с. HTTP readiness проверяет PostgreSQL каждые 1 с, fall2/rise1, fastinter250ms. Номинальный бюджет исключения отказавшего адреса — около 2 с плюс длительность зависшей readiness; connect timeout1с, HTTP server timeout8с. PostgreSQL-клиент имеет checkout deadline4с, pool wait1с, connect2с, statement2.5с и lock1с. Поэтому в реальном отказе возможны короткие ошибки; время подтверждается экспериментом. HAProxy не повторяет POST автоматически (`retries 0`).
`/health/live` проверяет процесс; `/health/ready` — доступность обязательного синхронного PG-пути. Valkey необязателен для корректности и при runtime-отказе обходится с circuit backoff. Uvicorn завершает принятые запросы с бюджетом5с; контейнеру выделено10с. До импорта uvicorn действует ранний обработчик SIGTERM. Поочерёдная замена сначала добавляет готовую реплику, ждёт проверки и DNS, затем отправляет SIGTERM одной старой. Это проверка замены процессов одной версии; совместимость разных версий приложения отдельно не заявляется. Миграция002 — только расширение схемы, старый код ЛР2 не теряет свои таблицы.
Ресурсы обычного стенда: API2×0.5CPU/384MiB, PostgreSQL1CPU/512MiB, Valkey0.25CPU/128MiB, LB0.5CPU/128MiB. API работают под UID10001, root filesystem read-only, tmpfs16MiB, capabilities удалены. Python3.14.7 и uv0.13.7 закреплены digest; зависимости — `uv.lock`.
## Атомарность повторной отправки
1. API проверяет Bearer и членство в чате, затем валидирует body и ASCII ключ `[A-Za-z0-9._:-]{1,128}`.
2. Область — пользователь, POST, нормализованный UUID чата и ключ. Другие чаты и пользователи независимы. В ЛР3 body содержит только text; схема вложений появляется в ЛР4.
3. В одной PG-транзакции берётся `pg_advisory_xact_lock` по первым64битам SHA-256 области. Коллизия lock лишь дополнительно сериализует независимые запросы: окончательное сравнение выполняется по полному составному PK.
4. Если результат моложе24ч, сравнивается SHA-256 канонического JSON. Другой text даёт409; тот же —201 с прежним Message, включая id и created_at. Порядок JSON-полей не влияет на hash.
5. Для нового/истёкшего ключа в той же транзакции вставляется Message и результат идемпотентности с expiry по clock_timestamp PostgreSQL. COMMIT подтверждает оба изменения; исключение откатывает оба.
При неизвестном исходе POST клиент сохраняет исходные key/body и повторяет их. Повтор заново проверяет сессию/ACL. Ошибки5xx не сохраняются как успешный replay result. Записи истёкших ключей могут оставаться в таблице до обслуживания; expiry проверяется независимо от физического удаления. Контракт не обещает дедупликацию после24ч. Создание чата, регистрации и входа этим механизмом не защищены.
## Проверки и исследование
Проверки выполняются последовательно относительно изменений одного стенда. Первый собственный прогон был испорчен ошибочной пересборкой всех API/LB во время тестов: 503 и connection refused; этот прогон не засчитан. Повтор на готовом стенде: **87passed, 279.66с**; детали в `evidence/http-tests.json`.
В `evidence/cluster-first.json` сохранены30 параллельных одинаковых POST через обе реальные реплики: один Message. Инъекция исключения в INSERT idempotency вернула500 и оставила0 сообщений; повтор того же key после удаления fault завершился201. Искусственное истечение записи разрешило новый результат.
Остановка API1:221 подтверждённых записей/215 чтений,7 ошибок, первый успех после завершения action4.191с от его старта. API2:211/203,9 ошибок,3.643с. Это15-секундные серии с паузой40ms между циклами чтение/запись. Все подтверждённые Message и уникальность text проверены по полной истории. Public failover отдельно:115 успешных записей,3 ошибки, восстановление3.142с; наблюдались2 alias до отказа. Операторские before/after содержат CID/PID/StartedAt и реальные квоты/ports.
Поочерёдная замена:655 подтверждённых записей,36 ошибок запросов за45-секундное окно. Старые контейнеры заменены новыми API3/4; replay прежнего ключа вернул исходный Message. Стенд не обещает безошибочную замену: в этой реализации обнаружение изменения адресов и закрытие keep-alive могут давать краткие503/обрывы. Подтверждённые данные не потеряны; клиент может повторить неопределённый POST тем же ключом. Полный restart API/PG/cache/LB прошёл public persistence; StartedAt всех state services и LB изменились. Ни volumes, ни токены не заменялись.
| Критерий README | Реализация и проверка | Evidence |
|---|---|---|
| 3: две API, один LB, общий PG/сессии | Compose scale2, HTTP через LB, реальные разные CID/PID | `evidence/cluster-first.json` |
| 3: единая действующая сессия через реплики | 30 concurrent POST, два alias, один результат | тот же файл |
| 3: остановка любой API, запись≤10с, сохранность | обе API остановлены по очереди, проверена вся история | тот же файл |
| 3: одинаковый ключ без дублей, конфликт body | PostgreSQL transaction, HTTP concurrent/conflict, 5xx rollback | `evidence/http-tests.json`, `evidence/cluster-first.json` |
| 3: smoke/persistence/failover | исходные публичные TCP-сценарии | `evidence/cluster-first.json`; smoke повторяется в clean clone |
| 4: readiness, deadlines, SIGTERM и исключение backend | HAProxy check1с/fall2 и конечные бюджеты | `infra/haproxy.cfg`, `evidence/cluster-first.json` |
| 4: остановить каждую API с RPS/errors/recovery | собственные15-секундные серии | `evidence/cluster-first.json` |
| 4: квоты, закрытые API ports, нет container_name | compose и inspect обеих API | тот же файл |
| 4: последовательная замена под нагрузкой, миграции отдельно | surge одна готовая API, SIGTERM старой;002 expand | тот же файл, `migrations/002_idempotency.sql` |
| 5: fixed/growing1/2/4, saturation, улучшение | изолированная матрица и отдельный before/after | `evidence/scaling-first/`; `evidence/acl-comparison-first/` |
Основная матрица завершена:42/42 прогона,40825 успешных HTTP,0 ошибок. Исходные данные и график: [scaling-first](evidence/scaling-first/analysis.md), [CSV](evidence/scaling-first/summary.csv), [PNG](evidence/scaling-first/scaling.png). Эти измерения относятся к исходному runtime `df4eee9`, source harness `fa3a69d`; последующее улучшение пути чтения измеряется отдельно.
При C8 фиксированный1CPU:1/2/4 API дали205.2/185.1/155.2RPS; при растущем CPU:110.4/185.3/257.9RPS. При C32 fixed1/2/4 p95 вырос425.05/462.64/883.62ms, RPS снизился177.9/162.1/128.5. Счётчики cgroup показывают частые throttled periods: например, fixedN2C32/r2 —45/68 и64/68 периодов у разных API. Само увеличение concurrency выше8 на этом профиле уже не повышает HTTP throughput. Последовательный C1 почти не выигрывает от числа процессов, поскольку клиент не создаёт параллельной работы.
Дополнительная настройка пула: при N2/C32 пул1 дал медиану200.4RPS, пул10 —175.8RPS (три чередующихся повтора). Маленький пул снизил конкуренцию внутри ограниченного CPU-процесса, но отдельный повтор1 оказался хуже10; это локальный результат с заметным разбросом, основание для проверки настройки под свою нагрузку. Стандартный пул10 сохранён, универсальная рекомендация уменьшать его не делается.
В улучшенном runtime повторный SELECTget_chat в service.history заменён передачей уже авторизованного immutableChat из HTTP-слоя с повторной локальной проверкой actor∈members. Членство в обязательном API неизменяемо; ACL по-прежнему выполняется до parsing limit/cursor, это проверяет отдельный негативный HTTP-тест. Транзакции отправки и дедупликация не менялись. Сравнение до/после выполнено отдельно: [12 прогонов](evidence/acl-comparison-first/analysis.md), [raw/CSV](evidence/acl-comparison-first/summary.csv), [график](evidence/acl-comparison-first/comparison.png). Четыре API по1CPU/384MiB, pool10, C16,2с прогрева/12с измерения, три чередующихся повтора на каждую версию и PG quota. Before образ revision `df4eee9`, after `cabd0ca`.
При PostgreSQL1CPU медиана RPS **211.3→247.4 (+17.1%)**, p95 **220.94→200.82ms**; RPS улучшился в каждой из трёх пар. При PG0.1CPU:29.4→32.8RPS, p951290.65→1298.07ms; здесь выигрыш по медиане мал и один after хуже before, улучшение latency не подтверждено. PG расходует квоту в97.2–100% CPU-периодов при0.1CPU против0–6.2% при1CPU. Это контролируемый пример насыщения общей БД: четыре API не заменяют ресурс PostgreSQL. Всего18849 успешных HTTP,0 ошибок. Один из повторов заметно отличается: короткие окна не дают уверенной оценки дальних хвостов.
Таким образом, выбранный трек завершён с фиксированными и растущими ресурсами, контролем общей БД и подтверждённым улучшением SQL-пути при стандартном PG1CPU. Последующая правка src/api.py меняет только форматирование этой же операции. Итоговый код дополнительно проверен из чистого clone. Для уровня5 предусмотрены фиксированная суммарная квота API1CPU/1024MiB и серия с ростом ресурсов0.5CPU/256MiB на реплику; 1/2/4 процесса, C1/8/32, два повтора,2с прогрева и6с измерения. Смешанный closed-loop профиль:80% чтений истории limit20 и20% отправок с уникальными ключами. Перед каждой ячейкой в отдельном исследовательском проекте очищается только его новая БД и создаётся100 сообщений. Отдельно сравнивается пул1/10 при N2/C32 в трёх чередующихся повторах. Исходные короткие окна, два повтора основной серии и три настройки пула не позволяют оценить стабильность production-tail; необработанные данные сохранены.
## Вопросы защиты
**Чем отказ процесса отличается от отказа узла?** Остановка одной API оставляет другие процессы, LB и общую БД живыми. Потеря этого Docker-host одновременно лишает нас всех контейнеров и локальных volumes. Этот стенд проверяет отказ процесса, многосерверной HA здесь нет.
**Почему depends_on/healthcheck не являются балансировщиком?** depends_on определяет порядок запуска и начальную готовность. Healthcheck помечает состояние контейнера. Маршрутизацию клиентских запросов, исключение backend и DNS выполняет отдельный HAProxy. Его конфигурация нужна и после старта.
**Какие POST повторять после timeout и что защищает ключ?** POST сообщения можно повторить с теми же key/body в пределах24ч: запись Message и результата общая и атомарная. Ключ не повторяет данные на клиенте и не защищает другой endpoint; registration/chat/login при неизвестном результате требуют отдельного решения. Timeout не означает гарантированный rollback.
**Что сломается при локальных сессиях?** После round robin другая API не узнает токен, появятся случайные401; остановка владельца лишит сессии доступа. Sticky session скрыла бы часть ошибок до первого отказа. Общая PG-сессия сохраняет identity и абсолютный expiry через любую API и restart.
## Границы и источники
Один LB, PostgreSQL, Valkey и Docker-host; потеря БД/host не маскируется ростом API. TLS, репликация PostgreSQL, backup/PITR и промышленное наблюдение здесь не добавлены. HTTP доступен только на loopback, данные синтетические. Closed-loop имеет coordinated omission; короткий локальный прогон не задаёт production SLA. Пул — на процесс: рост реплик увеличивает возможное суммарное число PG-соединений.
Источники: [HAProxy3.2 configuration](https://docs.haproxy.org/3.2/configuration.html), [PG18 explicit locks](https://www.postgresql.org/docs/18/explicit-locking.html), [Compose services](https://docs.docker.com/reference/compose-file/services/). Библиотеки перечислены в `pyproject.toml`, точные версии — `uv.lock`. Помощь ИИ использована при разборе требований, подготовке реализации и проверок; результаты подтверждаются сохранёнными запусками.
Собственный `.github/workflows/application.yml` описывает сборку и интеграцию на runner с Docker; наличие файла не означает, что удалённый runner запущен или CI прошёл.
История сортируется по серверному `(created_at,id)`. На одном host процессы используют общие системные часы; при их скачке назад либо при скосу времени на разных узлах новые записи могут оказаться до уже выданного cursor. Snapshot между HTTP-страницами не обещается.
Повтор исследования (отдельные временные проекты, основной стенд предварительно остановить):
```sh
make image
make research RESEARCH_OUTPUT=evidence/scaling-new
uv run --locked python -m research.analyze evidence/scaling-new
# Для сравнения нужны собственные образы двух зафиксированных версий:
uv run --locked python -m research.compare --before IMAGE_BEFORE --after IMAGE_AFTER --output evidence/comparison-new
uv run --locked python -m research.analyze_compare evidence/comparison-new
# Полная проверка отдельного clone с новыми credentials/volumes:
uv run --locked python scripts/verify_clean.py --output evidence/clean-clone-new.json
```
Исследования автоматически удаляют только volumes своих уникальных временных проектов; основной messenger не очищается. CPU/RAM собираются по docker stats в измерительном окне, cgroup cpu.stat подтверждает throttling. memory.peak — lifetime peak контейнера, его нельзя объявлять пиком одного измерительного окна. Модель closed-loop, генератор на том же Mac/Docker VM, фоновые задачи host не изолированы, snapshots страниц не фиксированы. Hardware: Apple M1/8 CPU/8GiB host, Docker VM8CPU/6209978368bytes. Точные версии и hardware сохранены в run.json.
## Итоговая проверка из чистого clone
[evidence/clean-clone-first.json](evidence/clean-clone-first.json): source commit `ac5c30c`, новый `.env`, случайные секреты, пустые volumes, отдельный loopback-порт и отдельное имя образа. `uv sync --locked`, lint/check/validate, build, public smoke и **88 HTTP-тестов за273.11с** прошли. Полная cluster-проверка повторена: обе API остановлены по очереди, повторен rolling, сохранены сообщения/прежние сессии после полного restart. Хеши каждого Python-модуля src в ОБЕИХ новых API совпали с clone; OCI revision обеих — `ac5c30c`; прямые API ports отсутствуют. `git fsck --full` завершён успешно. Временный проект и только его volumes удалены; основной проект сохранён остановленным.
Измерения основной матрицы и сравнения версий привязаны к собственным source/image revisions выше. Чистый clone проверяет итоговое поведение;54 нагрузочных прогона повторно в нём не запускались. После source `ac5c30c` меняются только отчёт, evidence и оформление сдачи.
+103
View File
@@ -0,0 +1,103 @@
name: mtusi-lab3-maslov
services:
postgres:
image: postgres:18.3@sha256:7e32e9833a6fb1c92c32552794cb6ed569d51b445a54907d35fc112ef39684db
environment:
POSTGRES_USER: messenger
POSTGRES_DB: messenger
POSTGRES_PASSWORD_FILE: /run/secrets/postgres_password
PGDATA: /var/lib/postgresql/data/pgdata
secrets: [postgres_password]
volumes:
- postgres_data:/var/lib/postgresql
healthcheck:
test: [CMD-SHELL, 'pg_isready -U messenger -d messenger']
interval: 2s
timeout: 2s
retries: 20
cpus: 1
mem_limit: 512m
memswap_limit: 512m
cache:
image: valkey/valkey:8.1.3@sha256:fea8b3e67b15729d4bb70589eb03367bab9ad1ee89c876f54327fc7c6e618571
command: [valkey-server, /run/secrets/valkey.conf]
secrets:
- cache_password
- source: valkey_config
target: valkey.conf
healthcheck:
test: [CMD-SHELL, 'REDISCLI_AUTH="$$(cat /run/secrets/cache_password)" valkey-cli ping']
interval: 2s
timeout: 2s
retries: 20
cpus: 0.25
mem_limit: 128m
memswap_limit: 128m
migrate:
image: ${API_IMAGE:-mtusi-messenger:lab3}
pull_policy: never
command: [python, -m, src.migrate]
environment:
DATABASE_URL_FILE: /run/secrets/database_url
secrets: [database_url]
depends_on:
postgres: {condition: service_healthy}
cpus: 0.5
mem_limit: 256m
read_only: true
api:
image: ${API_IMAGE:-mtusi-messenger:lab3}
build:
context: .
args:
VCS_REF: ${VCS_REF:-unknown}
scale: ${API_REPLICAS:-2}
environment:
APP_HOST: 0.0.0.0
APP_PORT: 8080
DATABASE_URL_FILE: /run/secrets/database_url
CACHE_URL_FILE: /run/secrets/cache_url
SESSION_TTL_SECONDS: ${SESSION_TTL_SECONDS:-3600}
DB_POOL_SIZE: ${DB_POOL_SIZE:-10}
secrets: [database_url, cache_url]
depends_on:
migrate: {condition: service_completed_successfully}
cache: {condition: service_healthy}
cpus: ${API_CPUS:-0.5}
mem_limit: ${API_MEMORY:-384m}
memswap_limit: ${API_MEMORY:-384m}
pids_limit: 128
read_only: true
tmpfs: ["/tmp:rw,noexec,nosuid,size=16m"]
cap_drop: [ALL]
security_opt: [no-new-privileges:true]
stop_grace_period: 10s
lb:
image: haproxy:3.2.1-alpine@sha256:ac79fe145f2bb6626ff26b584a2d0a34e791906c01015f2ae037aa3137b683d9
ports:
- "127.0.0.1:${PUBLIC_PORT:-8083}:8080"
volumes:
- ./infra/haproxy.cfg:/usr/local/etc/haproxy/haproxy.cfg:ro
depends_on:
api: {condition: service_healthy}
cpus: 0.5
mem_limit: 128m
memswap_limit: 128m
read_only: true
tmpfs: ["/tmp:rw,nosuid,size=4m"]
security_opt: [no-new-privileges:true]
stop_grace_period: 10s
secrets:
postgres_password:
file: ./secrets/postgres_password
database_url:
file: ./secrets/database_url
cache_password:
file: ./secrets/cache_password
cache_url:
file: ./secrets/cache_url
valkey_config:
file: ./secrets/valkey.conf
name: valkey.conf
volumes:
postgres_data:
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
+23
View File
@@ -0,0 +1,23 @@
# Насыщение PostgreSQL и устранение повторного ACL-запроса
80% GET limit20;20% POST; C16; 100 seed; warmup2s/measure12s; four API each1CPU/384MiB; PostgreSQL 0.1 vs 1CPU
Медианы трёх чередующихся повторов.
| PG CPU | Версия | RPS | p95 ms | Ошибки |
|---|---|---:|---:|---:|
| 0.1 | before | 29.4 | 1290.65 | 0 |
| 0.1 | after | 32.8 | 1298.07 | 0 |
| 1.0 | before | 211.3 | 220.94 | 0 |
| 1.0 | after | 247.4 | 200.82 | 0 |
Before revision: `df4eee9a55f0cd9476744ea4c75cba1c1b3a4a98`.
After revision: `cabd0cacb3d58e018e115ca0d629d6becf0179bb`.
При PG1CPU выигрыш RPS наблюдается во всех трёх парах; при PG0.1CPU один after хуже before. Короткие окна и большой разброс не задают production SLA.
Всего успешных HTTP: 18849; ошибок: 0.
Доли CPU-периодов PG с throttling (по каждому прогону):
- PG0.1CPU: [1.0, 1.0, 0.981, 1.0, 1.0, 0.972]
- PG1.0CPU: [0.062, 0.016, 0.008, 0.055, 0.0, 0.008]
Binary file not shown.

After

Width:  |  Height:  |  Size: 42 KiB

+31
View File
@@ -0,0 +1,31 @@
{
"date": "2026-11-15T16:34:47.189897+00:00",
"complete": true,
"profile": "80% GET limit20;20% POST; C16; 100 seed; warmup2s/measure12s; four API each1CPU/384MiB; PostgreSQL 0.1 vs 1CPU",
"images": {
"before": {
"image": "mtusi-messenger:lab3-baseline",
"digest": "sha256:52486f90ff283acbf4cc26a763e8ad33a84cf9b6e2f5cff7542e406cef3b7b40",
"revision": "df4eee9a55f0cd9476744ea4c75cba1c1b3a4a98"
},
"after": {
"image": "mtusi-messenger:lab3-optimized",
"digest": "sha256:3e0dae28e3322bd42f4381e2bc439f171348e755249b17f7bc4d123b6c140209",
"revision": "cabd0cacb3d58e018e115ca0d629d6becf0179bb"
}
},
"cells": [
"01-before-pg0.1-r1.json",
"02-after-pg0.1-r1.json",
"03-before-pg1.0-r1.json",
"04-after-pg1.0-r1.json",
"05-after-pg0.1-r2.json",
"06-before-pg0.1-r2.json",
"07-after-pg1.0-r2.json",
"08-before-pg1.0-r2.json",
"09-before-pg0.1-r3.json",
"10-after-pg0.1-r3.json",
"11-before-pg1.0-r3.json",
"12-after-pg1.0-r3.json"
]
}
+13
View File
@@ -0,0 +1,13 @@
version,pg_cpu_limit,repeat,successful_rps,errors,p50_ms,p95_ms,p99_ms
before,0.1,1,21.316091462624293,0,591.172,2092.441,3016.385
after,0.1,1,32.82279718409536,0,390.596,1298.065,1793.751
before,1.0,1,149.6992402960217,0,55.172,355.143,647.555
after,1.0,1,247.44248843044662,0,36.672,200.816,315.847
after,0.1,2,57.746143786015686,0,197.4,816.803,1291.057
before,0.1,2,29.352502633001926,0,489.503,1290.654,1979.792
after,1.0,2,240.92116586235193,0,36.694,201.685,366.092
before,1.0,2,211.34858169102048,0,41.28,220.936,441.311
before,0.1,3,32.46450561340184,0,396.505,1201.577,1600.971
after,0.1,3,28.37186406851818,0,409.049,1405.366,1789.497
before,1.0,3,240.42921118588941,0,35.779,213.804,365.71
after,1.0,3,256.6444044962857,0,34.582,192.979,310.591
1 version pg_cpu_limit repeat successful_rps errors p50_ms p95_ms p99_ms
2 before 0.1 1 21.316091462624293 0 591.172 2092.441 3016.385
3 after 0.1 1 32.82279718409536 0 390.596 1298.065 1793.751
4 before 1.0 1 149.6992402960217 0 55.172 355.143 647.555
5 after 1.0 1 247.44248843044662 0 36.672 200.816 315.847
6 after 0.1 2 57.746143786015686 0 197.4 816.803 1291.057
7 before 0.1 2 29.352502633001926 0 489.503 1290.654 1979.792
8 after 1.0 2 240.92116586235193 0 36.694 201.685 366.092
9 before 1.0 2 211.34858169102048 0 41.28 220.936 441.311
10 before 0.1 3 32.46450561340184 0 396.505 1201.577 1600.971
11 after 0.1 3 28.37186406851818 0 409.049 1405.366 1789.497
12 before 1.0 3 240.42921118588941 0 35.779 213.804 365.71
13 after 1.0 3 256.6444044962857 0 34.582 192.979 310.591
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
+15
View File
@@ -0,0 +1,15 @@
{
"date": "2026-11-15T16:16:15.978617+00:00",
"source_commit": "b14fb9a162d8012e83699cd8facdab178587d04b",
"runtime_image_revision": "df4eee9",
"command": "BASE_URL=http://127.0.0.1:8083 .venv/bin/python -m pytest student-tests -q --tb=short",
"result": "PASS",
"output": "........................................................................ [ 82%]\n............... [100%]\n87 passed in 279.66s (0:04:39)",
"prior_attempt": {
"result": "INVALID_ENVIRONMENT",
"passed": 27,
"failed": 19,
"errors": 41,
"reason": "Test requests overlapped with a mistaken full Compose image recreation; LB/API were unavailable. Repeated after completion without concurrent mutations."
}
}
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
+30
View File
@@ -0,0 +1,30 @@
# Измерения масштабирования
Source commit: `fa3a69d9b4267939fae8db422dff31cd079725f5`.
Медианы повторов; latency percentile внутри каждого прогона относится только к успешным ответам.
| Серия | API | C | RPS | p95, ms | Ошибки |
|---|---:|---:|---:|---:|---:|
| fixed | 1 | 1 | 157.3 | 8.95 | 0 |
| fixed | 1 | 8 | 205.2 | 75.84 | 0 |
| fixed | 1 | 32 | 177.9 | 425.05 | 0 |
| fixed | 2 | 1 | 140.8 | 9.81 | 0 |
| fixed | 2 | 8 | 185.1 | 99.24 | 0 |
| fixed | 2 | 32 | 162.1 | 462.64 | 0 |
| fixed | 4 | 1 | 134.2 | 10.61 | 0 |
| fixed | 4 | 8 | 155.2 | 222.69 | 0 |
| fixed | 4 | 32 | 128.5 | 883.62 | 0 |
| growing | 1 | 1 | 124.3 | 13.84 | 0 |
| growing | 1 | 8 | 110.4 | 102.86 | 0 |
| growing | 1 | 32 | 94.1 | 455.20 | 0 |
| growing | 2 | 1 | 132.6 | 11.20 | 0 |
| growing | 2 | 8 | 185.3 | 98.75 | 0 |
| growing | 2 | 32 | 164.3 | 520.34 | 0 |
| growing | 4 | 1 | 123.4 | 12.82 | 0 |
| growing | 4 | 8 | 257.9 | 107.74 | 0 |
| growing | 4 | 32 | 158.9 | 577.39 | 0 |
Пул1 → пул10: 200.4 → 175.8 успешных RPS (N2/C32, три повтора).
Настройка считается подтверждённым улучшением только при наблюдаемом выигрыше; это ограниченная локальная серия.
+62
View File
@@ -0,0 +1,62 @@
{
"source_commit": "fa3a69d9b4267939fae8db422dff31cd079725f5",
"date": "2026-11-15T16:18:34.862014+00:00",
"host": "macOS-15.4.1-arm64-arm-64bit-Mach-O",
"docker": {
"ServerVersion": "29.9.2",
"NCPU": 8,
"MemTotal": 6209978368,
"Architecture": "aarch64",
"OperatingSystem": "Docker Desktop"
},
"duration": 6,
"warmup": 2,
"profile": "80% GET history limit20; 20% POST unique key; closed-loop",
"complete": true,
"cells": [
"01-fixed-n1-c1-r1-p10.json",
"02-fixed-n1-c1-r2-p10.json",
"03-fixed-n1-c8-r1-p10.json",
"04-fixed-n1-c8-r2-p10.json",
"05-fixed-n1-c32-r1-p10.json",
"06-fixed-n1-c32-r2-p10.json",
"07-fixed-n2-c1-r1-p10.json",
"08-fixed-n2-c1-r2-p10.json",
"09-fixed-n2-c8-r1-p10.json",
"10-fixed-n2-c8-r2-p10.json",
"11-fixed-n2-c32-r1-p10.json",
"12-fixed-n2-c32-r2-p10.json",
"13-fixed-n4-c1-r1-p10.json",
"14-fixed-n4-c1-r2-p10.json",
"15-fixed-n4-c8-r1-p10.json",
"16-fixed-n4-c8-r2-p10.json",
"17-fixed-n4-c32-r1-p10.json",
"18-fixed-n4-c32-r2-p10.json",
"19-growing-n1-c1-r1-p10.json",
"20-growing-n1-c1-r2-p10.json",
"21-growing-n1-c8-r1-p10.json",
"22-growing-n1-c8-r2-p10.json",
"23-growing-n1-c32-r1-p10.json",
"24-growing-n1-c32-r2-p10.json",
"25-growing-n2-c1-r1-p10.json",
"26-growing-n2-c1-r2-p10.json",
"27-growing-n2-c8-r1-p10.json",
"28-growing-n2-c8-r2-p10.json",
"29-growing-n2-c32-r1-p10.json",
"30-growing-n2-c32-r2-p10.json",
"31-growing-n4-c1-r1-p10.json",
"32-growing-n4-c1-r2-p10.json",
"33-growing-n4-c8-r1-p10.json",
"34-growing-n4-c8-r2-p10.json",
"35-growing-n4-c32-r1-p10.json",
"36-growing-n4-c32-r2-p10.json",
"37-pool-n2-c32-r1-p1.json",
"38-pool-n2-c32-r1-p10.json",
"39-pool-n2-c32-r2-p10.json",
"40-pool-n2-c32-r2-p1.json",
"41-pool-n2-c32-r3-p1.json",
"42-pool-n2-c32-r3-p10.json"
],
"runtime_image_revision": "df4eee9a55f0cd9476744ea4c75cba1c1b3a4a98",
"runtime_equal_to_source_commit": true
}
Binary file not shown.

After

Width:  |  Height:  |  Size: 106 KiB

+43
View File
@@ -0,0 +1,43 @@
series,replicas,concurrency,repeat,pool_size,successful_rps,errors,error_ratio,p50_ms,p95_ms,p99_ms
fixed,1,1,1,10,165.07767947989907,0,0.0,5.686,7.629,10.809
fixed,1,1,2,10,149.61439409018604,0,0.0,5.981,10.269,17.642
fixed,1,8,1,10,211.65991899019832,0,0.0,31.413,68.508,94.733
fixed,1,8,2,10,198.8283289091601,0,0.0,34.161,83.179,106.195
fixed,1,32,1,10,179.06741388685083,0,0.0,152.428,439.242,719.485
fixed,1,32,2,10,176.78095424063739,0,0.0,155.509,410.849,692.851
fixed,2,1,1,10,129.02990662137603,0,0.0,6.278,10.864,54.8
fixed,2,1,2,10,152.62929263739855,0,0.0,6.123,8.759,14.204
fixed,2,8,1,10,196.4800359108998,0,0.0,17.626,94.998,115.954
fixed,2,8,2,10,173.69962390399843,0,0.0,22.429,103.486,184.548
fixed,2,32,1,10,156.50040487681198,0,0.0,181.718,443.704,663.946
fixed,2,32,2,10,167.65953701338876,0,0.0,166.392,481.584,704.593
fixed,4,1,1,10,142.48482239783186,0,0.0,6.389,10.367,17.647
fixed,4,1,2,10,125.88207648163838,0,0.0,6.332,10.843,44.555
fixed,4,8,1,10,118.34539999970386,0,0.0,12.522,264.825,464.998
fixed,4,8,2,10,192.1342057427685,0,0.0,9.542,180.554,195.196
fixed,4,32,1,10,138.28620618496345,0,0.0,15.618,974.776,1248.537
fixed,4,32,2,10,118.77886331824459,0,0.0,198.297,792.454,1376.883
growing,1,1,1,10,116.48100146330145,0,0.0,5.944,16.399,77.405
growing,1,1,2,10,132.20926320279918,0,0.0,6.018,11.287,72.127
growing,1,8,1,10,109.33665681551416,0,0.0,86.76,103.52,114.182
growing,1,8,2,10,111.36479209555996,0,0.0,86.354,102.192,107.991
growing,1,32,1,10,97.35188039185765,0,0.0,308.314,415.437,489.101
growing,1,32,2,10,90.82762345324294,0,0.0,320.147,494.966,585.341
growing,2,1,1,10,124.799942422262,0,0.0,6.546,11.811,59.12
growing,2,1,2,10,140.46007909177436,0,0.0,6.62,10.59,16.863
growing,2,8,1,10,195.37255277667734,0,0.0,16.973,95.054,167.252
growing,2,8,2,10,175.30636015284236,0,0.0,19.372,102.439,184.685
growing,2,32,1,10,158.50391583720702,0,0.0,178.795,565.052,758.551
growing,2,32,2,10,170.11492222924323,0,0.0,177.916,475.635,605.655
growing,4,1,1,10,115.16211935610346,0,0.0,6.858,14.033,58.519
growing,4,1,2,10,131.5487532925524,0,0.0,6.616,11.609,22.638
growing,4,8,1,10,277.1422612739073,0,0.0,12.846,95.114,156.297
growing,4,8,2,10,238.7251942336141,0,0.0,14.245,120.368,204.822
growing,4,32,1,10,151.42242336820735,0,0.0,148.202,613.274,828.682
growing,4,32,2,10,166.35153942031727,0,0.0,127.253,541.513,733.458
pool,2,32,1,1,206.3668093701107,0,0.0,148.348,298.829,416.622
pool,2,32,1,10,157.87641931117705,0,0.0,204.863,416.938,510.686
pool,2,32,2,10,178.67593359431726,0,0.0,185.302,327.124,402.2
pool,2,32,2,1,168.0097097888627,0,0.0,179.629,394.52,504.704
pool,2,32,3,1,200.35677878454243,0,0.0,151.213,304.633,404.823
pool,2,32,3,10,175.76904087132343,0,0.0,190.759,392.898,474.386
1 series replicas concurrency repeat pool_size successful_rps errors error_ratio p50_ms p95_ms p99_ms
2 fixed 1 1 1 10 165.07767947989907 0 0.0 5.686 7.629 10.809
3 fixed 1 1 2 10 149.61439409018604 0 0.0 5.981 10.269 17.642
4 fixed 1 8 1 10 211.65991899019832 0 0.0 31.413 68.508 94.733
5 fixed 1 8 2 10 198.8283289091601 0 0.0 34.161 83.179 106.195
6 fixed 1 32 1 10 179.06741388685083 0 0.0 152.428 439.242 719.485
7 fixed 1 32 2 10 176.78095424063739 0 0.0 155.509 410.849 692.851
8 fixed 2 1 1 10 129.02990662137603 0 0.0 6.278 10.864 54.8
9 fixed 2 1 2 10 152.62929263739855 0 0.0 6.123 8.759 14.204
10 fixed 2 8 1 10 196.4800359108998 0 0.0 17.626 94.998 115.954
11 fixed 2 8 2 10 173.69962390399843 0 0.0 22.429 103.486 184.548
12 fixed 2 32 1 10 156.50040487681198 0 0.0 181.718 443.704 663.946
13 fixed 2 32 2 10 167.65953701338876 0 0.0 166.392 481.584 704.593
14 fixed 4 1 1 10 142.48482239783186 0 0.0 6.389 10.367 17.647
15 fixed 4 1 2 10 125.88207648163838 0 0.0 6.332 10.843 44.555
16 fixed 4 8 1 10 118.34539999970386 0 0.0 12.522 264.825 464.998
17 fixed 4 8 2 10 192.1342057427685 0 0.0 9.542 180.554 195.196
18 fixed 4 32 1 10 138.28620618496345 0 0.0 15.618 974.776 1248.537
19 fixed 4 32 2 10 118.77886331824459 0 0.0 198.297 792.454 1376.883
20 growing 1 1 1 10 116.48100146330145 0 0.0 5.944 16.399 77.405
21 growing 1 1 2 10 132.20926320279918 0 0.0 6.018 11.287 72.127
22 growing 1 8 1 10 109.33665681551416 0 0.0 86.76 103.52 114.182
23 growing 1 8 2 10 111.36479209555996 0 0.0 86.354 102.192 107.991
24 growing 1 32 1 10 97.35188039185765 0 0.0 308.314 415.437 489.101
25 growing 1 32 2 10 90.82762345324294 0 0.0 320.147 494.966 585.341
26 growing 2 1 1 10 124.799942422262 0 0.0 6.546 11.811 59.12
27 growing 2 1 2 10 140.46007909177436 0 0.0 6.62 10.59 16.863
28 growing 2 8 1 10 195.37255277667734 0 0.0 16.973 95.054 167.252
29 growing 2 8 2 10 175.30636015284236 0 0.0 19.372 102.439 184.685
30 growing 2 32 1 10 158.50391583720702 0 0.0 178.795 565.052 758.551
31 growing 2 32 2 10 170.11492222924323 0 0.0 177.916 475.635 605.655
32 growing 4 1 1 10 115.16211935610346 0 0.0 6.858 14.033 58.519
33 growing 4 1 2 10 131.5487532925524 0 0.0 6.616 11.609 22.638
34 growing 4 8 1 10 277.1422612739073 0 0.0 12.846 95.114 156.297
35 growing 4 8 2 10 238.7251942336141 0 0.0 14.245 120.368 204.822
36 growing 4 32 1 10 151.42242336820735 0 0.0 148.202 613.274 828.682
37 growing 4 32 2 10 166.35153942031727 0 0.0 127.253 541.513 733.458
38 pool 2 32 1 1 206.3668093701107 0 0.0 148.348 298.829 416.622
39 pool 2 32 1 10 157.87641931117705 0 0.0 204.863 416.938 510.686
40 pool 2 32 2 10 178.67593359431726 0 0.0 185.302 327.124 402.2
41 pool 2 32 2 1 168.0097097888627 0 0.0 179.629 394.52 504.704
42 pool 2 32 3 1 200.35677878454243 0 0.0 151.213 304.633 404.823
43 pool 2 32 3 10 175.76904087132343 0 0.0 190.759 392.898 474.386
+36
View File
@@ -0,0 +1,36 @@
global
log stdout format raw local0
maxconn 512
stats socket /tmp/admin.sock mode 600 level admin
defaults
mode http
log global
option httplog
timeout connect 1s
timeout client 15s
timeout server 8s
timeout http-request 10s
timeout http-keep-alive 5s
retries 0
resolvers docker
nameserver dns 127.0.0.11:53
resolve_retries 3
timeout resolve 1s
timeout retry 1s
hold valid 1s
hold obsolete 1s
hold nx 1s
frontend public
bind :8080
default_backend messenger
backend messenger
balance roundrobin
option httpchk
http-check send meth GET uri /health/ready ver HTTP/1.1 hdr Host api
http-check expect status 200
default-server check inter 1s fastinter 250ms fall 2 rise 1 resolvers docker init-addr libc,none resolve-prefer ipv4
server-template api 1-8 api:8080
+36
View File
@@ -0,0 +1,36 @@
CREATE TABLE users (
id uuid PRIMARY KEY,
username text NOT NULL UNIQUE CHECK (username ~ '^[a-z0-9_]{3,32}$'),
display_name text NOT NULL CHECK (char_length(display_name) BETWEEN 1 AND 100),
password_hash text NOT NULL,
created_at timestamptz NOT NULL
);
CREATE TABLE chats (
id uuid PRIMARY KEY,
title text NOT NULL CHECK (char_length(title) BETWEEN 1 AND 100),
created_at timestamptz NOT NULL
);
CREATE TABLE members (
chat_id uuid NOT NULL REFERENCES chats(id),
user_id uuid NOT NULL REFERENCES users(id),
position integer NOT NULL CHECK (position BETWEEN 0 AND 99),
PRIMARY KEY (chat_id, user_id),
UNIQUE (chat_id, position)
);
CREATE INDEX members_user_chat_idx ON members(user_id,chat_id);
CREATE TABLE messages (
id uuid PRIMARY KEY,
chat_id uuid NOT NULL REFERENCES chats(id),
sender_id uuid NOT NULL,
text text NOT NULL CHECK (char_length(text) BETWEEN 1 AND 4000),
created_at timestamptz NOT NULL,
FOREIGN KEY (chat_id, sender_id) REFERENCES members(chat_id,user_id)
);
CREATE INDEX messages_history_idx ON messages(chat_id,created_at,id);
CREATE TABLE sessions (
token_digest bytea PRIMARY KEY CHECK (octet_length(token_digest)=32),
user_id uuid NOT NULL REFERENCES users(id),
created_at timestamptz NOT NULL,
expires_at timestamptz NOT NULL CHECK (expires_at > created_at)
);
CREATE INDEX sessions_expiry_idx ON sessions(expires_at);
+10
View File
@@ -0,0 +1,10 @@
CREATE TABLE idempotency (
user_id uuid NOT NULL REFERENCES users(id),
chat_id uuid NOT NULL REFERENCES chats(id),
key text NOT NULL CHECK (key ~ '^[A-Za-z0-9._:-]{1,128}$'),
body_hash bytea NOT NULL CHECK (octet_length(body_hash)=32),
message_id uuid NOT NULL REFERENCES messages(id),
expires_at timestamptz NOT NULL,
PRIMARY KEY(user_id, chat_id, key)
);
CREATE INDEX idempotency_expiry_idx ON idempotency(expires_at);
+31
View File
@@ -0,0 +1,31 @@
[project]
name = "mtusi-messenger-lab3"
version = "1.0.0"
requires-python = ">=3.14,<3.15"
dependencies = [
"fastapi==0.143.3", "pydantic==2.14.1", "uvicorn==0.54.0",
"psycopg[binary,pool]==3.3.6", "argon2-cffi>=25,<26", "redis==8.1.0",
]
[dependency-groups]
dev = ["pytest>=8,<10", "httpx>=0.28,<1", "ruff>=0.14,<1", "openapi-spec-validator==0.7.2", "psutil>=7,<8", "matplotlib>=3.10,<4"]
[tool.ruff]
line-length = 79
target-version = "py314"
include = ["src/**/*.py", "student-tests/**/*.py", "research/**/*.py", "scripts/verify_*.py"]
[tool.ruff.lint]
select = ["E", "F", "I", "B", "UP"]
[tool.ruff.per-file-target-version]
"scripts/*.py" = "py39"
# SQL и генерируемые Markdown-таблицы в инструментах не разрываем.
[tool.ruff.lint.per-file-ignores]
"research/*.py" = ["E501"]
"scripts/*.py" = ["E501"]
[tool.pytest.ini_options]
testpaths = ["student-tests"]
addopts = "-ra"
+117
View File
@@ -0,0 +1,117 @@
import argparse
import csv
import json
import statistics
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt # noqa: E402
def main():
parser = argparse.ArgumentParser()
parser.add_argument("directory")
args = parser.parse_args()
directory = Path(args.directory)
meta = json.loads((directory / "run.json").read_text())
assert meta["complete"]
rows = [
json.loads((directory / name).read_text()) for name in meta["cells"]
]
keys = [
"series",
"replicas",
"concurrency",
"repeat",
"pool_size",
"successful_rps",
"errors",
"error_ratio",
"p50_ms",
"p95_ms",
"p99_ms",
]
with (directory / "summary.csv").open("w", newline="") as stream:
writer = csv.DictWriter(stream, fieldnames=keys)
writer.writeheader()
writer.writerows({k: r[k] for k in keys} for r in rows)
fig, axes = plt.subplots(1, 2, figsize=(11, 4), constrained_layout=True)
for ax, series in zip(axes, ["fixed", "growing"], strict=True):
for concurrency in [1, 8, 32]:
values = [
statistics.median(
r["successful_rps"]
for r in rows
if r["series"] == series
and r["replicas"] == n
and r["concurrency"] == concurrency
)
for n in [1, 2, 4]
]
ax.plot([1, 2, 4], values, marker="o", label=f"C={concurrency}")
ax.set(
title=series,
xlabel="API replicas",
ylabel="Successful HTTP requests/s",
xticks=[1, 2, 4],
)
ax.legend()
ax.grid(alpha=0.3)
fig.savefig(directory / "scaling.png", dpi=160)
plt.close(fig)
pool = {
n: statistics.median(
r["successful_rps"]
for r in rows
if r["series"] == "pool" and r["pool_size"] == n
)
for n in [1, 10]
}
lines = [
"# Измерения масштабирования",
"",
f"Source commit: `{meta['source_commit']}`.",
"",
"Медианы повторов; latency percentile внутри каждого прогона относится только к успешным ответам.",
"",
"| Серия | API | C | RPS | p95, ms | Ошибки |",
"|---|---:|---:|---:|---:|---:|",
]
for series in ["fixed", "growing"]:
for n in [1, 2, 4]:
for c in [1, 8, 32]:
group = [
r
for r in rows
if r["series"] == series
and r["replicas"] == n
and r["concurrency"] == c
]
lines.append(
f"| {series} | {n} | {c} | {statistics.median(r['successful_rps'] for r in group):.1f} | {statistics.median(r['p95_ms'] for r in group):.2f} | {sum(r['errors'] for r in group)} |"
)
lines.extend(
[
"",
f"Пул1 → пул10: {pool[1]:.1f} → {pool[10]:.1f} успешных RPS (N2/C32, три повтора).",
"",
"Настройка считается подтверждённым улучшением только при наблюдаемом выигрыше; это ограниченная локальная серия.",
]
)
(directory / "analysis.md").write_text("\n".join(lines) + "\n")
print(
json.dumps(
{
"cells": len(rows),
"success": sum(r["success"] for r in rows),
"errors": sum(r["errors"] for r in rows),
"pool_rps": pool,
}
)
)
if __name__ == "__main__":
main()
+103
View File
@@ -0,0 +1,103 @@
import argparse
import csv
import json
import statistics
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt # noqa: E402
def main():
parser = argparse.ArgumentParser()
parser.add_argument("directory")
directory = Path(parser.parse_args().directory)
meta = json.loads((directory / "run.json").read_text())
assert meta["complete"]
rows = [
json.loads((directory / name).read_text()) for name in meta["cells"]
]
keys = [
"version",
"pg_cpu_limit",
"repeat",
"successful_rps",
"errors",
"p50_ms",
"p95_ms",
"p99_ms",
]
with (directory / "summary.csv").open("w", newline="") as stream:
writer = csv.DictWriter(stream, fieldnames=keys)
writer.writeheader()
writer.writerows({k: r[k] for k in keys} for r in rows)
lines = [
"# Насыщение PostgreSQL и устранение повторного ACL-запроса",
"",
meta["profile"],
"",
"Медианы трёх чередующихся повторов.",
"",
"| PG CPU | Версия | RPS | p95 ms | Ошибки |",
"|---|---|---:|---:|---:|",
]
fig, axes = plt.subplots(1, 2, figsize=(9, 4), constrained_layout=True)
for ax, cpu in zip(axes, [0.1, 1.0], strict=True):
values = []
for version in ["before", "after"]:
group = [
r
for r in rows
if r["version"] == version and r["pg_cpu_limit"] == cpu
]
median = statistics.median(r["successful_rps"] for r in group)
values.append(median)
lines.append(
f"| {cpu} | {version} | {median:.1f} | {statistics.median(r['p95_ms'] for r in group):.2f} | {sum(r['errors'] for r in group)} |"
)
ax.bar(["before", "after"], values, color=["#d48748", "#3c88b5"])
ax.set(
title=f"PostgreSQL {cpu} CPU", ylabel="Successful HTTP requests/s"
)
ax.grid(axis="y", alpha=0.2)
fig.savefig(directory / "comparison.png", dpi=160)
plt.close(fig)
lines += [
"",
"Before revision: `" + meta["images"]["before"]["revision"] + "`.",
"After revision: `" + meta["images"]["after"]["revision"] + "`.",
"",
"При PG1CPU выигрыш RPS наблюдается во всех трёх парах; при PG0.1CPU один after хуже before. Короткие окна и большой разброс не задают production SLA.",
"",
f"Всего успешных HTTP: {sum(r['success'] for r in rows)}; ошибок: {sum(r['errors'] for r in rows)}.",
"",
"Доли CPU-периодов PG с throttling (по каждому прогону):",
]
for cpu in [0.1, 1.0]:
ratios = []
for r in rows:
if r["pg_cpu_limit"] != cpu:
continue
def parse(text):
return {
s.split()[0]: int(s.split()[1]) for s in text.splitlines()
}
before, after = parse(r["pg_cpu_before"]), parse(r["pg_cpu_after"])
periods = after["nr_periods"] - before["nr_periods"]
ratios.append(
round(
(after["nr_throttled"] - before["nr_throttled"])
/ max(1, periods),
3,
)
)
lines.append(f"- PG{cpu}CPU: {ratios}")
(directory / "analysis.md").write_text("\n".join(lines) + "\n")
if __name__ == "__main__":
main()
+173
View File
@@ -0,0 +1,173 @@
"""Чередующееся сравнение двух собственных образов на отдельном стенде."""
import argparse
import asyncio
import json
import os
import socket
import threading
import time
from datetime import UTC, datetime
from pathlib import Path
from uuid import uuid4
import httpx
from research.run import command, compose, cpu_counters, load, sample
from scripts.stand import actor, auth
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--before", required=True)
parser.add_argument("--after", required=True)
parser.add_argument("--output", required=True)
args = parser.parse_args()
out = Path(args.output)
out.mkdir(parents=True, exist_ok=False)
with socket.socket() as sock:
sock.bind(("127.0.0.1", 0))
port = sock.getsockname()[1]
project = "lab3-compare-" + uuid4().hex[:8]
env = {
**os.environ,
"COMPOSE_PROJECT_NAME": project,
"PUBLIC_PORT": str(port),
"API_REPLICAS": "4",
"API_CPUS": "1",
"API_MEMORY": "384m",
"DB_POOL_SIZE": "10",
}
report = {
"date": datetime.now(UTC).isoformat(),
"complete": False,
"profile": "80% GET limit20;20% POST; C16; 100 seed; warmup2s/measure12s; four API each1CPU/384MiB; PostgreSQL 0.1 vs 1CPU",
"images": {},
"cells": [],
}
for name, image in [("before", args.before), ("after", args.after)]:
info = json.loads(command(["docker", "image", "inspect", image], env))[
0
]
report["images"][name] = {
"image": image,
"digest": info["Id"],
"revision": info["Config"]["Labels"][
"org.opencontainers.image.revision"
],
}
try:
number = 0
cases = [
(repeat, version, pg_cpu)
for repeat in range(1, 4)
for pg_cpu in [0.1, 1.0]
for version in (
["before", "after"] if repeat % 2 else ["after", "before"]
)
]
for repeat, version, pg_cpu in cases:
number += 1
env["API_IMAGE"] = (
args.before if version == "before" else args.after
)
compose(env, "up", "-d", "--no-build", "--wait")
pg_cid = compose(env, "ps", "-q", "postgres")
command(["docker", "update", "--cpus", str(pg_cpu), pg_cid], env)
time.sleep(3)
assert project.startswith("lab3-compare-")
compose(
env,
"exec",
"-T",
"postgres",
"psql",
"-U",
"messenger",
"-d",
"messenger",
"-v",
"ON_ERROR_STOP=1",
"-c",
"TRUNCATE idempotency,messages,members,chats,sessions,users",
)
base = f"http://127.0.0.1:{port}"
with httpx.Client(
base_url=base, trust_env=False, timeout=15
) as client:
user, _, session = actor(client)
r = client.post(
"/api/v1/chats",
headers=auth(session),
json={
"title": "Controlled comparison",
"member_ids": [user["id"]],
},
)
assert r.status_code == 201
chat = r.json()["id"]
for i in range(100):
r = client.post(
f"/api/v1/chats/{chat}/messages",
headers={
**auth(session),
"Idempotency-Key": uuid4().hex,
},
json={"text": f"Seed {i}"},
)
assert r.status_code == 201
asyncio.run(load(base, session["token"], chat, 16, 2))
counters = cpu_counters(env)
pg_before = command(
["docker", "exec", pg_cid, "cat", "/sys/fs/cgroup/cpu.stat"],
env,
)
samples = []
stop = threading.Event()
thread = threading.Thread(target=sample, args=(env, stop, samples))
thread.start()
try:
result = asyncio.run(
load(base, session["token"], chat, 16, 12)
)
finally:
stop.set()
thread.join()
result.update(
version=version,
repeat=repeat,
samples=samples,
pg_cpu_limit=pg_cpu,
pg_cpu_before=pg_before,
pg_cpu_after=command(
[
"docker",
"exec",
pg_cid,
"cat",
"/sys/fs/cgroup/cpu.stat",
],
env,
),
cgroup_before=counters,
cgroup_after=cpu_counters(env),
)
name = f"{number:02}-{version}-pg{pg_cpu}-r{repeat}.json"
(out / name).write_text(
json.dumps(result, ensure_ascii=False, indent=2) + "\n"
)
report["cells"].append(name)
print(
f"{name}: {result['successful_rps']:.1f} RPS; {result['errors']} errors",
flush=True,
)
report["complete"] = True
finally:
(out / "run.json").write_text(
json.dumps(report, ensure_ascii=False, indent=2) + "\n"
)
compose(env, "down", "--volumes", "--remove-orphans")
if __name__ == "__main__":
main()
+352
View File
@@ -0,0 +1,352 @@
"""Изолированная матрица масштабирования; рабочие volumes не используются."""
import argparse
import asyncio
import json
import os
import platform
import socket
import subprocess
import threading
import time
from datetime import UTC, datetime
from pathlib import Path
from uuid import uuid4
import httpx
ROOT = Path(__file__).resolve().parents[1]
def command(args, env):
r = subprocess.run(args, cwd=ROOT, env=env, capture_output=True, text=True)
if r.returncode:
raise RuntimeError(r.stderr.replace(str(ROOT), "<repository>"))
return r.stdout.strip()
def compose(env, *args):
return command(["docker", "compose", *args], env)
def percentile(values, p):
data = sorted(values)
return data[min(len(data) - 1, int((len(data) - 1) * p))] if data else None
async def load(base, token, chat, concurrency, duration):
headers = {"Authorization": "Bearer " + token}
rows = []
start = time.perf_counter()
async with httpx.AsyncClient(
base_url=base,
trust_env=False,
timeout=8,
limits=httpx.Limits(max_connections=concurrency),
) as c:
async def work(index):
sequence = index
while time.perf_counter() - start < duration:
at = time.perf_counter()
try:
if sequence % 5 == 0:
r = await c.post(
f"/api/v1/chats/{chat}/messages",
headers={
**headers,
"Idempotency-Key": uuid4().hex,
},
json={"text": "Scaling synthetic message 🌍"},
)
else:
r = await c.get(
f"/api/v1/chats/{chat}/messages",
headers=headers,
params={"limit": 20},
)
status = r.status_code
instance = r.headers.get("x-instance-id")
except httpx.HTTPError:
status, instance = 0, None
rows.append(
[
round((at - start) * 1000, 3),
round((time.perf_counter() - at) * 1000, 3),
status,
instance,
]
)
sequence += 1
await asyncio.gather(*(work(i) for i in range(concurrency)))
elapsed = time.perf_counter() - start
good = [r[1] for r in rows if r[2] in (200, 201)]
return {
"elapsed_seconds": elapsed,
"successful_rps": len(good) / elapsed,
"success": len(good),
"errors": len(rows) - len(good),
"error_ratio": (len(rows) - len(good)) / max(1, len(rows)),
"p50_ms": percentile(good, 0.5),
"p95_ms": percentile(good, 0.95),
"p99_ms": percentile(good, 0.99),
"requests": rows,
}
def cpu_counters(env):
ids = compose(env, "ps", "-q", "api").split()
code = (
"import json,pathlib; print(json.dumps({p: "
"pathlib.Path('/sys/fs/cgroup/'+p).read_text() for p in "
"['cpu.stat','memory.current','memory.peak']}))"
)
return {
cid[:12]: json.loads(
command(["docker", "exec", cid, "python", "-c", code], env)
)
for cid in ids
}
def sample(env, stop, output):
while not stop.is_set():
try:
ids = compose(env, "ps", "-q").split()
raw = command(
[
"docker",
"stats",
"--no-stream",
"--format",
"{{json .}}",
*ids,
],
env,
)
output.append(
{
"at": time.time(),
"containers": [json.loads(s) for s in raw.splitlines()],
}
)
except Exception as exc:
output.append({"error": type(exc).__name__})
stop.wait(0.25)
def main():
p = argparse.ArgumentParser()
p.add_argument("--output", default="evidence/scaling-repeat")
p.add_argument("--duration", type=float, default=6)
p.add_argument("--warmup", type=float, default=2)
p.add_argument("--only-improvement", action="store_true")
a = p.parse_args()
out = Path(a.output)
out.mkdir(parents=True, exist_ok=False)
with socket.socket() as sock:
sock.bind(("127.0.0.1", 0))
port = sock.getsockname()[1]
project = "lab3-scaling-" + uuid4().hex[:8]
env = {
**os.environ,
"COMPOSE_PROJECT_NAME": project,
"PUBLIC_PORT": str(port),
}
base = f"http://127.0.0.1:{port}"
meta = {
"source_commit": command(["git", "rev-parse", "HEAD"], env),
"date": datetime.now(UTC).isoformat(),
"host": platform.platform(),
"docker": json.loads(
command(["docker", "info", "--format", "{{json .}}"], env)
),
"duration": a.duration,
"warmup": a.warmup,
"profile": "80% GET history limit20; 20% POST unique key; closed-loop",
"complete": False,
"cells": [],
}
# Полный docker info не нужен: сохраняется только воспроизводимое железо.
meta["docker"] = {
k: meta["docker"][k]
for k in [
"ServerVersion",
"NCPU",
"MemTotal",
"Architecture",
"OperatingSystem",
]
}
cases = []
if not a.only_improvement:
for series in ["fixed", "growing"]:
for n in [1, 2, 4]:
for concurrency in [1, 8, 32]:
for repeat in range(1, 3):
cases.append((series, n, concurrency, repeat, 10))
for repeat in range(1, 4):
# Чередование снижает систематический дрейф температуры/кэша.
for pool in [1, 10] if repeat % 2 else [10, 1]:
cases.append(("pool", 2, 32, repeat, pool))
revision = command(
[
"docker",
"image",
"inspect",
env.get("API_IMAGE", "mtusi-messenger:lab3"),
"--format",
'{{index .Config.Labels "org.opencontainers.image.revision"}}',
],
env,
)
meta["runtime_image_revision"] = revision
assert (
command(
[
"git",
"diff",
revision,
"--",
"src",
"migrations",
"pyproject.toml",
"uv.lock",
"Dockerfile",
],
env,
)
== ""
)
meta["runtime_equal_to_source_commit"] = True
current = None
try:
for number, (series, n, c, rep, pool) in enumerate(cases, 1):
cpus = 1 / n if series == "fixed" else 0.5
memory = int(1024 / n) if series == "fixed" else 256
env.update(
API_REPLICAS=str(n),
API_CPUS=str(cpus),
API_MEMORY=f"{memory}m",
DB_POOL_SIZE=str(pool),
)
configuration = (n, cpus, memory, pool)
if configuration != current:
compose(env, "up", "-d", "--no-build", "--wait")
current = configuration
time.sleep(3)
# Только новая БД изолированного исследовательского проекта.
assert env["COMPOSE_PROJECT_NAME"].startswith("lab3-scaling-")
compose(
env,
"exec",
"-T",
"postgres",
"psql",
"-U",
"messenger",
"-d",
"messenger",
"-v",
"ON_ERROR_STOP=1",
"-c",
"TRUNCATE idempotency,messages,members,chats,sessions,users",
)
with httpx.Client(
base_url=base, trust_env=False, timeout=15
) as client:
username = "scale_" + uuid4().hex[:20]
password = uuid4().hex
r = client.post(
"/api/v1/users",
json={
"username": username,
"display_name": "Synthetic",
"password": password,
},
)
assert r.status_code == 201
user = r.json()
r = client.post(
"/api/v1/auth/sessions",
auth=httpx.BasicAuth(username, password),
)
assert r.status_code == 201
token = r.json()["token"]
r = client.post(
"/api/v1/chats",
headers={"Authorization": "Bearer " + token},
json={"title": "Scaling", "member_ids": [user["id"]]},
)
assert r.status_code == 201
chat = r.json()["id"]
for i in range(100):
r = client.post(
f"/api/v1/chats/{chat}/messages",
headers={
"Authorization": "Bearer " + token,
"Idempotency-Key": uuid4().hex,
},
json={"text": f"Seed {i}"},
)
assert r.status_code == 201
asyncio.run(load(base, token, chat, c, a.warmup))
counters_before = cpu_counters(env)
samples = []
stop = threading.Event()
thread = threading.Thread(target=sample, args=(env, stop, samples))
thread.start()
try:
result = asyncio.run(load(base, token, chat, c, a.duration))
finally:
stop.set()
thread.join()
counters_after = cpu_counters(env)
ids = compose(env, "ps", "-q").split()
inspect = json.loads(command(["docker", "inspect", *ids], env))
result.update(
series=series,
replicas=n,
concurrency=c,
repeat=rep,
pool_size=pool,
api_cpu_each=cpus,
api_memory_mib_each=memory,
dataset_initial_messages=100,
samples=samples,
cgroup_before=counters_before,
cgroup_after=counters_after,
actual_limits=[
{
"role": x["Config"]["Labels"][
"com.docker.compose.service"
],
"cpu_nano": x["HostConfig"]["NanoCpus"],
"memory": x["HostConfig"]["Memory"],
}
for x in inspect
],
)
filename = f"{number:02}-{series}-n{n}-c{c}-r{rep}-p{pool}.json"
(out / filename).write_text(
json.dumps(result, ensure_ascii=False, indent=2) + "\n"
)
meta["cells"].append(filename)
(out / "run.json").write_text(
json.dumps(meta, ensure_ascii=False, indent=2) + "\n"
)
print(
f"{number}/{len(cases)} {filename}: {result['successful_rps']:.1f} RPS, {result['errors']} errors",
flush=True,
)
meta["complete"] = True
finally:
(out / "run.json").write_text(
json.dumps(meta, ensure_ascii=False, indent=2) + "\n"
)
compose(env, "down", "--volumes", "--remove-orphans")
if __name__ == "__main__":
main()
+47
View File
@@ -0,0 +1,47 @@
"""Создаёт только отсутствующие локальные файлы; секреты не печатает."""
import os
import secrets
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
def create(path, value):
try:
fd = os.open(path, os.O_WRONLY | os.O_CREAT | os.O_EXCL, 0o644)
except FileExistsError:
return
with os.fdopen(fd, "w") as stream:
stream.write(value + "\n")
def main():
directory = ROOT / "secrets"
directory.mkdir(mode=0o700, exist_ok=True)
pg = directory / "postgres_password"
cache = directory / "cache_password"
create(pg, secrets.token_urlsafe(32))
create(cache, secrets.token_urlsafe(32))
create(
directory / "database_url",
"postgresql://messenger:"
+ pg.read_text().strip()
+ "@postgres:5432/messenger",
)
create(
directory / "cache_url",
"redis://:" + cache.read_text().strip() + "@cache:6379/0",
)
create(
directory / "valkey.conf",
"bind 0.0.0.0\nprotected-mode yes\nport 6379\nmaxmemory 64mb\n"
'maxmemory-policy allkeys-lru\nsave ""\nappendonly no\nrequirepass '
+ cache.read_text().strip(),
)
create(ROOT / ".env", (ROOT / ".env.example").read_text().rstrip())
print("Local configuration ready; existing files preserved")
if __name__ == "__main__":
main()
+83
View File
@@ -0,0 +1,83 @@
"""Явные операторские действия только над API текущего Compose-проекта."""
import json
import os
import subprocess
import sys
import time
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
def command(*args):
return subprocess.check_output(args, cwd=ROOT, text=True).strip()
def containers():
ids = command("docker", "compose", "ps", "-q", "api").split()
return json.loads(command("docker", "inspect", *ids)) if ids else []
def alias(cid):
code = (
"import urllib.request; r=urllib.request.urlopen("
"'http://127.0.0.1:8080/health/live',timeout=2); "
"print(r.headers['X-Instance-Id'])"
)
return command("docker", "exec", cid, "python", "-c", code)
def wait_count(count):
deadline = time.monotonic() + 60
while time.monotonic() < deadline:
data = containers()
if len(data) == count and all(
c["State"].get("Health", {}).get("Status") == "healthy"
for c in data
):
# DNS и проверки HAProxy должны успеть увидеть новый адрес.
time.sleep(3)
return
time.sleep(0.5)
raise TimeoutError("Replica readiness")
def main():
action = sys.argv[1]
data = containers()
if action == "stop":
target = os.environ["TARGET_INSTANCE_ID"]
matches = [c for c in data if alias(c["Id"]) == target]
if len(matches) != 1 or len(data) < 2:
raise RuntimeError("Target must identify one running replica")
command("docker", "stop", "--time", "10", matches[0]["Id"])
print("Selected API stopped")
elif action == "rolling":
count = len(data)
if count < 2:
raise RuntimeError("Need at least two replicas")
# Сначала добавляется готовая реплика, затем удаляется одна старая.
for old in data:
command(
"docker",
"compose",
"up",
"-d",
"--no-deps",
"--no-recreate",
"--scale",
f"api={count + 1}",
"api",
)
wait_count(count + 1)
command("docker", "stop", "--time", "10", old["Id"])
command("docker", "rm", old["Id"])
wait_count(count)
print("Sequential replacement completed")
else:
raise ValueError("Unknown action")
if __name__ == "__main__":
main()
+4
View File
@@ -0,0 +1,4 @@
#!/bin/sh
set -eu
docker compose restart api postgres cache lb
docker compose up -d --no-build --no-recreate --wait
+3
View File
@@ -0,0 +1,3 @@
#!/bin/sh
set -eu
exec python3 scripts/cluster.py rolling
+137
View File
@@ -0,0 +1,137 @@
import json
import os
import subprocess
import time
from pathlib import Path
from uuid import uuid4
import httpx
ROOT = Path(__file__).resolve().parents[1]
BASE = os.environ.get("BASE_URL", "http://127.0.0.1:8083")
def command(args, *, input=None):
result = subprocess.run(
args, cwd=ROOT, input=input, capture_output=True, text=True
)
if result.returncode:
raise RuntimeError(
"Command failed: "
+ args[0]
+ "; "
+ result.stderr.replace(str(ROOT), "<repository>")
)
return result.stdout.strip()
def compose(*args):
return command(["docker", "compose", *args])
def sql(query, *, database="messenger"):
return compose(
"exec",
"-T",
"postgres",
"psql",
"-U",
"messenger",
"-d",
database,
"-A",
"-t",
"-v",
"ON_ERROR_STOP=1",
"-c",
query,
)
def cache(command_name, *args):
script = (
'export REDISCLI_AUTH="$(cat /run/secrets/cache_password)"; '
'exec valkey-cli "$@"'
)
return compose(
"exec", "-T", "cache", "sh", "-c", script, "sh", command_name, *args
)
def ready(base=BASE, *, timeout=30):
end = time.monotonic() + timeout
with httpx.Client(base_url=base, trust_env=False, timeout=6) as client:
while time.monotonic() < end:
try:
if client.get("/health/ready").status_code == 200:
return
except httpx.HTTPError:
pass
time.sleep(0.2)
raise TimeoutError("API readiness")
def actor(client):
username = "verify_" + uuid4().hex[:20]
password = "Synthetic-" + uuid4().hex + ":🌍"
result = client.post(
"/api/v1/users",
json={
"username": username,
"display_name": "Synthetic",
"password": password,
},
)
assert result.status_code == 201
user = result.json()
session = client.post(
"/api/v1/auth/sessions", auth=httpx.BasicAuth(username, password)
)
assert session.status_code == 201
return user, password, session.json()
def auth(session):
return {"Authorization": "Bearer " + session["token"]}
def extra_api(*, ttl=3600, pool_size=1):
name = "mtusi-lab3-test-" + uuid4().hex[:10]
compose(
"run",
"-d",
"--no-deps",
"--name",
name,
"-e",
"SESSION_TTL_SECONDS=" + str(ttl),
"-e",
"DB_POOL_SIZE=" + str(pool_size),
"-p",
"127.0.0.1::8080",
"api",
)
try:
info = json.loads(command(["docker", "inspect", name]))[0]
base = (
"http://127.0.0.1:"
+ info["NetworkSettings"]["Ports"]["8080/tcp"][0]["HostPort"]
)
ready(base)
return name, base
except Exception:
command(["docker", "rm", "-f", name])
raise
def instances():
data = {}
for role in ["api", "postgres", "cache"]:
cid = compose("ps", "-q", role)
info = json.loads(command(["docker", "inspect", cid]))[0]
data[role] = {
"container": cid[:12],
"pid": info["State"]["Pid"],
"started_at": info["State"]["StartedAt"],
}
return data
+3
View File
@@ -0,0 +1,3 @@
#!/bin/sh
set -eu
exec python3 scripts/cluster.py stop
+201
View File
@@ -0,0 +1,201 @@
import argparse
import atexit
import json
import os
import socket
import subprocess
import tempfile
from datetime import UTC, datetime
from pathlib import Path
from uuid import uuid4
ROOT = Path(__file__).resolve().parents[1]
PROGRESS = {}
def save_incomplete():
receipt = PROGRESS.get("receipt")
if receipt and receipt.get("result") != "PASS":
target = PROGRESS["output"].with_name(
PROGRESS["output"].stem + "-failed.json"
)
if target.exists():
target = target.with_name(
target.stem + "-" + uuid4().hex[:8] + target.suffix
)
target.parent.mkdir(parents=True, exist_ok=True)
target.write_text(
json.dumps(
{**receipt, "result": "INCOMPLETE"},
ensure_ascii=False,
indent=2,
)
+ "\n"
)
atexit.register(save_incomplete)
def run(args, *, cwd, env, expected=0):
result = subprocess.run(
args, cwd=cwd, env=env, capture_output=True, text=True
)
def redact(text):
return (
text.replace(str(cwd), "<clean-clone>")
.replace(str(ROOT), "<repository>")
.replace(str(Path.home()), "<home>")
)
if result.returncode != expected:
if "receipt" in PROGRESS:
PROGRESS["receipt"]["failed_action"] = redact(" ".join(args))
raise RuntimeError(redact(result.stdout + result.stderr))
return redact(result.stdout.strip())
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--output", default="evidence/clean-clone.json")
args = parser.parse_args()
output = ROOT / args.output
if output.exists():
raise SystemExit("Choose new output path")
assert not subprocess.check_output(
["git", "status", "--porcelain"], cwd=ROOT
)
sha = subprocess.check_output(
["git", "rev-parse", "HEAD"], cwd=ROOT, text=True
).strip()
identity = uuid4().hex[:10]
project = "mtusi-lab3-clean-" + identity
with socket.socket() as reservation:
reservation.bind(("127.0.0.1", 0))
public_port = str(reservation.getsockname()[1])
env = {
**os.environ,
"COMPOSE_PROJECT_NAME": project,
"API_IMAGE": "mtusi-messenger:lab3-clean-" + identity,
"PUBLIC_PORT": public_port,
"VCS_REF": sha,
"NO_PROXY": "localhost,127.0.0.1",
"no_proxy": "localhost,127.0.0.1",
}
receipt = {
"date": datetime.now(UTC).isoformat(),
"source_commit": sha,
"fresh_secrets_environment_and_volumes": True,
"command": "python scripts/verify_clean.py",
"fixed_host_port_for_restart": int(public_port),
}
PROGRESS.update(receipt=receipt, output=output)
with tempfile.TemporaryDirectory(prefix="mtusi-lab3-clean-") as temporary:
clone = Path(temporary) / "lab3"
run(
[
"git",
"clone",
"--no-hardlinks",
"--branch",
"solution/bvt2301-maslov",
str(ROOT),
str(clone),
],
cwd=ROOT,
env=env,
)
for name in [".env", ".venv", "secrets"]:
assert not (clone / name).exists()
run(["uv", "sync", "--locked"], cwd=clone, env=env)
receipt["static_checks"] = run(
["make", "check", "validate", "lint", "PYTHON=.venv/bin/python"],
cwd=clone,
env=env,
)
run(["make", "bootstrap"], cwd=clone, env=env)
try:
run(
["docker", "compose", "up", "-d", "--build", "--wait"],
cwd=clone,
env=env,
)
env["BASE_URL"] = "http://127.0.0.1:" + public_port
receipt["smoke"] = run(
["make", "test", "BASE_URL=" + env["BASE_URL"]],
cwd=clone,
env=env,
)
receipt["http_tests"] = run(
[".venv/bin/pytest", "student-tests", "-q"], cwd=clone, env=env
)
run(
[
".venv/bin/python",
"-m",
"scripts.verify_cluster",
"--output",
"evidence/clean-cluster.json",
],
cwd=clone,
env=env,
)
receipt["cluster"] = json.loads(
(clone / "evidence/clean-cluster.json").read_text()
)
import hashlib
expected = {
p.name: hashlib.sha256(p.read_bytes()).hexdigest()
for p in (clone / "src").glob("*.py")
}
ids = run(
["docker", "compose", "ps", "-q", "api"], cwd=clone, env=env
).split()
assert len(ids) == 2
code = (
"import hashlib,json,pathlib; print(json.dumps({p.name:"
"hashlib.sha256(p.read_bytes()).hexdigest() for p in pathlib.Path('/app/src').glob('*.py')}))"
)
for cid in ids:
actual = json.loads(
run(
["docker", "exec", cid, "python", "-c", code],
cwd=clone,
env=env,
)
)
assert actual == expected
info = json.loads(
run(["docker", "inspect", cid], cwd=clone, env=env)
)[0]
assert (
info["Config"]["Labels"][
"org.opencontainers.image.revision"
]
== sha
)
assert not info["HostConfig"]["PortBindings"]
receipt["both_replicas_runtime_hashes_and_revision_match"] = True
receipt["fsck"] = run(
["git", "fsck", "--full"], cwd=clone, env=env
)
receipt["checks_completed"] = True
finally:
assert project.startswith("mtusi-lab3-clean-")
run(
["docker", "compose", "down", "--volumes", "--remove-orphans"],
cwd=clone,
env=env,
)
receipt["result"] = "PASS"
output.parent.mkdir(parents=True, exist_ok=True)
output.write_text(json.dumps(receipt, ensure_ascii=False, indent=2) + "\n")
print(
"PASS: fresh clone, build, HTTP tests, failover, rolling, persistence and runtime hashes"
)
if __name__ == "__main__":
main()
+342
View File
@@ -0,0 +1,342 @@
"""Проверки реальных процессов, failover, rolling и общей идемпотентности."""
import argparse
import json
import os
import subprocess
import threading
import time
from concurrent.futures import ThreadPoolExecutor
from datetime import datetime, timezone
from pathlib import Path
from uuid import uuid4
import httpx
from scripts import cluster
from scripts.stand import BASE, ROOT, actor, auth, command, compose, ready, sql
def snapshot():
ids = compose("ps", "-q").split()
result = []
for c in json.loads(command(["docker", "inspect", *ids])):
role = c["Config"]["Labels"]["com.docker.compose.service"]
entry = {
"role": role,
"container": c["Id"][:12],
"pid": c["State"]["Pid"],
"started_at": c["State"]["StartedAt"],
"ports": c["HostConfig"]["PortBindings"],
"cpu_nano": c["HostConfig"]["NanoCpus"],
"memory_limit": c["HostConfig"]["Memory"],
}
if role == "api":
entry["instance"] = cluster.alias(c["Id"])
assert not entry["ports"]
result.append(entry)
return result
def exercise(client, session, chat, action, *, seconds=15):
path = f"/api/v1/chats/{chat['id']}/messages"
observations, confirmed, errors, reads = [], {}, 0, 0
started = time.monotonic()
finished = threading.Event()
action_result = {}
def operate():
time.sleep(0.5)
try:
action()
except BaseException as exc:
action_result["error"] = type(exc).__name__
finally:
action_result["finished"] = round(time.monotonic() - started, 3)
finished.set()
thread = threading.Thread(target=operate)
thread.start()
key = uuid4().hex
try:
while time.monotonic() - started < seconds or not finished.is_set():
h = {**auth(session), "Idempotency-Key": key}
try:
r = client.post(path, headers=h, json={"text": key})
if r.status_code != 201:
errors += 1
continue
m = r.json()
confirmed[m["id"]] = m
observations.append(
{
"at": round(time.monotonic() - started, 3),
"instance": r.headers["x-instance-id"],
}
)
key = uuid4().hex
read = client.get(
path, headers=auth(session), params={"limit": 100}
)
reads += read.status_code == 200
errors += read.status_code != 200
except httpx.HTTPError:
errors += 1
time.sleep(0.04)
finally:
thread.join(timeout=90)
assert not action_result.get("error"), action_result
found, cursor = [], None
while True:
r = client.get(
path,
headers=auth(session),
params={"limit": 100, **({"cursor": cursor} if cursor else {})},
)
assert r.status_code == 200
page = r.json()
found.extend(page["items"])
cursor = page["next_cursor"]
if not cursor:
break
by_id = {m["id"]: m for m in found}
assert all(by_id.get(k) == v for k, v in confirmed.items())
texts = [m["text"] for m in found]
assert len(texts) == len(set(texts))
recovery = next(
(
x["at"]
for x in observations
if x["at"] >= action_result["finished"]
),
None,
)
assert recovery is not None and (seconds > 20 or recovery <= 10)
elapsed = time.monotonic() - started
return {
"confirmed_writes": len(confirmed),
"successful_reads": reads,
"errors": errors,
"successful_rps": round((len(confirmed) + reads) / elapsed, 2),
"first_success_after_action_seconds": recovery,
"action": action_result,
"observations": observations,
"confirmed_preserved": True,
"duplicate_texts": False,
}
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--output", default="evidence/cluster-repeat.json")
args = parser.parse_args()
report = {
"date": datetime.now(timezone.utc).isoformat(),
"source_commit": command(["git", "rev-parse", "HEAD"]),
"complete": False,
}
output = Path(args.output)
output.parent.mkdir(parents=True, exist_ok=True)
try:
ready()
initial = snapshot()
apis = [c for c in initial if c["role"] == "api"]
assert len(apis) == 2 and len({c["pid"] for c in apis}) == 2
report["before"] = initial
with httpx.Client(base_url=BASE, trust_env=False, timeout=8) as client:
user, _, session = actor(client)
r = client.post(
"/api/v1/chats",
headers=auth(session),
json={
"title": "Cluster verification",
"member_ids": [user["id"]],
},
)
assert r.status_code == 201
chat = r.json()
path = f"/api/v1/chats/{chat['id']}/messages"
key = uuid4().hex
def repeat(_):
with httpx.Client(
base_url=BASE, trust_env=False, timeout=8
) as c:
r = c.post(
path,
headers={**auth(session), "Idempotency-Key": key},
json={"text": "Same concurrent request"},
)
assert r.status_code == 201
return r.json(), r.headers["x-instance-id"]
with ThreadPoolExecutor(max_workers=10) as pool:
repeated = list(pool.map(repeat, range(30)))
assert len({m[0]["id"] for m in repeated}) == 1
assert len({m[1] for m in repeated}) == 2
report["parallel_replay"] = {
"requests": 30,
"messages": 1,
"instances": sorted({m[1] for m in repeated}),
}
# Сбой второй записи транзакции не должен оставить сообщение.
fault_key = uuid4().hex
fault_text = "Atomic rollback " + fault_key
sql(
"CREATE FUNCTION student_idempo_fault() RETURNS trigger "
"LANGUAGE plpgsql AS $$ BEGIN IF NEW.key='"
+ fault_key
+ "' THEN RAISE EXCEPTION 'synthetic'; END IF; RETURN NEW; END $$; "
"CREATE TRIGGER student_idempo_fault BEFORE INSERT ON idempotency "
"FOR EACH ROW EXECUTE FUNCTION student_idempo_fault()"
)
try:
failed = client.post(
path,
headers={**auth(session), "Idempotency-Key": fault_key},
json={"text": fault_text},
)
assert failed.status_code == 500
assert (
sql(
"SELECT count(*) FROM messages WHERE text='"
+ fault_text
+ "'"
)
== "0"
)
finally:
sql(
"DROP TRIGGER student_idempo_fault ON idempotency; "
"DROP FUNCTION student_idempo_fault()"
)
retried = client.post(
path,
headers={**auth(session), "Idempotency-Key": fault_key},
json={"text": fault_text},
)
assert retried.status_code == 201
report["atomic_rollback_and_5xx_retry"] = True
sql(
"UPDATE idempotency SET expires_at=clock_timestamp()-interval '1 second' "
"WHERE key='" + fault_key + "'"
)
expired = client.post(
path,
headers={**auth(session), "Idempotency-Key": fault_key},
json={"text": "New retention window"},
)
assert expired.status_code == 201
assert expired.json()["id"] != retried.json()["id"]
report["expired_key_can_be_reused"] = True
# Отдельный чат каждой серии: проверки дубликатов не смешивают тексты.
report["failovers"] = []
for api in apis:
def stop(target=api["instance"]):
env = {**os.environ, "TARGET_INSTANCE_ID": target}
subprocess.run(
["scripts/stop-one.sh"],
cwd=ROOT,
env=env,
check=True,
capture_output=True,
text=True,
)
r = client.post(
"/api/v1/chats",
headers=auth(session),
json={"title": "Failover", "member_ids": [user["id"]]},
)
result = exercise(client, session, r.json(), stop)
assert api["instance"] not in [
o["instance"] for o in result["observations"][-5:]
]
assert (
command(
[
"docker",
"inspect",
"-f",
"{{.State.Running}}",
api["container"],
]
)
== "false"
)
result["stopped"] = api
report["failovers"].append(result)
compose("up", "-d", "--no-build", "--wait")
ready()
r = client.post(
"/api/v1/chats",
headers=auth(session),
json={"title": "Rolling", "member_ids": [user["id"]]},
)
before_roll = snapshot()
report["rolling"] = exercise(
client,
session,
r.json(),
lambda: command(["scripts/rolling-update.sh"]),
seconds=45,
)
after_roll = snapshot()
assert not (
{c["container"] for c in before_roll if c["role"] == "api"}
& {c["container"] for c in after_roll if c["role"] == "api"}
)
report["rolling"]["before"] = before_roll
report["rolling"]["after"] = after_roll
replay = repeat(0)[0]
assert replay == repeated[0][0]
report["replay_after_replacement"] = True
before = snapshot()
env = {
**os.environ,
"NO_PROXY": "localhost,127.0.0.1",
"no_proxy": "localhost,127.0.0.1",
}
for mode, hook in [
("failover", "scripts/stop-one.sh"),
("persistence", "scripts/restart.sh"),
]:
if mode == "persistence":
compose("up", "-d", "--no-build", "--wait")
ready()
before = snapshot()
run = subprocess.run(
[
".venv/bin/python",
"tests/resilience.py",
"--mode",
mode,
"--action-script",
hook,
"--base-url",
BASE,
],
cwd=ROOT,
env=env,
capture_output=True,
text=True,
)
assert run.returncode == 0, run.stdout + run.stderr
report["public_" + mode] = run.stdout.strip()
after = snapshot()
for role in ["postgres", "cache", "lb"]:
old = next(c for c in before if c["role"] == role)
new = next(c for c in after if c["role"] == role)
assert old["started_at"] != new["started_at"]
report["after_persistence"] = after
report["complete"] = True
finally:
output.write_text(
json.dumps(report, ensure_ascii=False, indent=2) + "\n"
)
print("PASS: process failover, rolling, replay and full persistence")
if __name__ == "__main__":
main()
View File
+44
View File
@@ -0,0 +1,44 @@
import json
import logging
import os
import signal
import sys
def early_stop(signum, _frame):
print(
json.dumps({"event": "stopped_before_startup", "signal": signum}),
flush=True,
)
sys.exit(0)
signal.signal(signal.SIGTERM, early_stop)
signal.signal(signal.SIGINT, early_stop)
import uvicorn # noqa: E402
from src.logging_config import JsonFormatter # noqa: E402
def main() -> None:
host = os.environ.get("APP_HOST", "127.0.0.1")
port = int(os.environ.get("APP_PORT", "8080"))
if not 1 <= port <= 65535:
raise ValueError("APP_PORT must be between 1 and 65535")
handler = logging.StreamHandler(sys.stdout)
handler.setFormatter(JsonFormatter())
logging.basicConfig(level=logging.INFO, handlers=[handler], force=True)
uvicorn.run(
"src.api:app",
host=host,
port=port,
workers=1,
access_log=False,
timeout_graceful_shutdown=5,
log_config=None,
)
if __name__ == "__main__":
main()
+303
View File
@@ -0,0 +1,303 @@
import json
import logging
import time
from collections.abc import AsyncIterator
from contextlib import asynccontextmanager
from datetime import datetime
from functools import partial
from uuid import UUID, uuid4
from anyio import to_thread
from fastapi import FastAPI, Request
from fastapi.encoders import jsonable_encoder
from pydantic import ValidationError
from starlette.exceptions import HTTPException
from starlette.responses import JSONResponse, Response
from src.cache import UserCache
from src.cursor import parse_uuid
from src.database import Database
from src.domain import timestamp
from src.errors import DomainError, invalid
from src.password import Passwords
from src.repository import PostgresRepository
from src.schemas import CreateChat, CreateMessage, CreateUser, Input
from src.service import MessengerService
from src.settings import Settings
logger = logging.getLogger("messenger")
INSTANCE_ID = uuid4().hex
def response(data: object, *, status: int = 200) -> JSONResponse:
return JSONResponse(
jsonable_encoder(
data, custom_encoder={datetime: timestamp, UUID: str}
),
status_code=status,
)
async def body(request: Request, model: type[Input]) -> Input:
try:
data = await request.body()
return model.model_validate(
json.loads(
data.decode("utf-8"),
parse_constant=lambda _: (_ for _ in ()).throw(ValueError()),
)
)
except ValueError, ValidationError, UnicodeError, RecursionError:
raise invalid() from None
def pagination(request: Request) -> tuple[int, str | None]:
limits = request.query_params.getlist("limit")
cursors = request.query_params.getlist("cursor")
if len(limits) > 1 or len(cursors) > 1:
raise invalid()
raw = limits[0] if limits else "50"
if not raw.isascii() or not raw.isdecimal():
raise invalid()
raw = raw.lstrip("0") or "0"
if len(raw) > 3:
raise invalid()
limit = int(raw)
if not 1 <= limit <= 100:
raise invalid()
return limit, cursors[0] if cursors else None
def service(request: Request) -> MessengerService:
return request.app.state.service
async def run_io(function, **kwargs):
return await to_thread.run_sync(partial(function, **kwargs))
async def actor(request: Request):
return await run_io(
service(request).identify, value=request.headers.get("authorization")
)
@asynccontextmanager
async def lifespan(app: FastAPI) -> AsyncIterator[None]:
settings = Settings.load()
database = Database(url=settings.database_url, size=settings.pool_size)
cache = UserCache(url=settings.cache_url)
try:
await run_io(database.open)
passwords = await run_io(Passwords)
app.state.database = database
app.state.service = MessengerService(
repository=PostgresRepository(database=database, cache=cache),
passwords=passwords,
session_ttl=settings.session_ttl,
)
app.state.ready = True
yield
finally:
app.state.ready = False
await run_io(cache.close)
await run_io(database.close)
app = FastAPI(
lifespan=lifespan, docs_url=None, redoc_url=None, openapi_url=None
)
@app.middleware("http")
async def request_context(request: Request, call_next):
request_id = str(uuid4())
request.state.request_id = request_id
started = time.perf_counter()
try:
result = await call_next(request)
except Exception:
logger.error(
json.dumps({"event": "internal_error", "request_id": request_id})
)
result = response(
{
"error": {
"code": "internal_error",
"message": "Internal error",
"request_id": request_id,
}
},
status=500,
)
result.headers["X-Request-Id"] = request_id
result.headers["X-Instance-Id"] = INSTANCE_ID
route = request.scope.get("route")
logger.info(
json.dumps(
{
"event": "request",
"request_id": request_id,
"method": request.method,
"route": getattr(route, "path", "unmatched"),
"status": result.status_code,
"elapsed_ms": round((time.perf_counter() - started) * 1000, 3),
}
)
)
return result
@app.exception_handler(DomainError)
async def domain_error(request: Request, error: DomainError):
result = response(
{
"error": {
"code": error.code,
"message": error.message,
"request_id": request.state.request_id,
}
},
status=error.status,
)
result.headers.update(error.headers)
return result
@app.exception_handler(HTTPException)
async def http_error(request: Request, error: HTTPException):
code = "not_found" if error.status_code == 404 else "invalid_request"
return response(
{
"error": {
"code": code,
"message": "Resource not found"
if error.status_code == 404
else "Invalid request",
"request_id": request.state.request_id,
}
},
status=error.status_code,
)
@app.get("/health/live")
async def live():
return response({"status": "ok"})
@app.get("/health/ready")
async def ready():
if not getattr(app.state, "ready", False):
raise DomainError(503, "unavailable", "Not ready")
await run_io(app.state.database.ping)
return response({"status": "ok"})
@app.post("/api/v1/users")
async def register(request: Request):
data = await body(request, CreateUser)
return response(
await run_io(service(request).register, **data.model_dump()),
status=201,
)
@app.get("/api/v1/users/me")
async def current(request: Request):
return response(await actor(request))
@app.post("/api/v1/chats")
async def create_chat(request: Request):
user = await actor(request)
data = await body(request, CreateChat)
return response(
await run_io(
service(request).create_chat, actor_id=user.id, **data.model_dump()
),
status=201,
)
@app.get("/api/v1/chats")
async def list_chats(request: Request):
user = await actor(request)
limit, cursor = pagination(request)
return response(
await run_io(
service(request).chats,
actor_id=user.id,
limit=limit,
cursor=cursor,
)
)
@app.get("/api/v1/chats/{chat_id}")
async def get_chat(request: Request, chat_id: str):
user = await actor(request)
return response(
await run_io(
service(request).get_chat,
actor_id=user.id,
chat_id=parse_uuid(chat_id),
)
)
@app.post("/api/v1/chats/{chat_id}/messages")
async def send(request: Request, chat_id: str):
user = await actor(request)
cid = parse_uuid(chat_id)
await run_io(service(request).get_chat, actor_id=user.id, chat_id=cid)
data = await body(request, CreateMessage)
return response(
await run_io(
service(request).send,
actor_id=user.id,
chat_id=cid,
idempotency_key=request.headers.get("idempotency-key"),
**data.model_dump(),
),
status=201,
)
@app.get("/api/v1/chats/{chat_id}/messages")
async def history(request: Request, chat_id: str):
user = await actor(request)
cid = parse_uuid(chat_id)
chat = await run_io(
service(request).get_chat, actor_id=user.id, chat_id=cid
)
limit, cursor = pagination(request)
return response(
await run_io(
service(request).history,
actor_id=user.id,
chat=chat,
limit=limit,
cursor=cursor,
)
)
@app.post("/api/v1/auth/sessions")
async def login(request: Request):
return response(
await run_io(
service(request).login,
authorization=request.headers.get("authorization"),
),
status=201,
)
@app.delete("/api/v1/auth/sessions/current")
async def logout(request: Request):
await run_io(
service(request).logout,
authorization=request.headers.get("authorization"),
)
return Response(status_code=204)
+66
View File
@@ -0,0 +1,66 @@
import json
import threading
import time
from datetime import datetime
from uuid import UUID
import redis
from redis.backoff import NoBackoff
from redis.retry import Retry
from src.domain import User, timestamp
class UserCache:
def __init__(self, *, url: str):
self.client = redis.Redis.from_url(
url,
socket_connect_timeout=0.15,
socket_timeout=0.15,
max_connections=16,
retry=Retry(NoBackoff(), 0),
decode_responses=True,
)
self.disabled_until = 0.0
self.lock = threading.Lock()
def close(self):
self.client.close()
def _command(self, function, *args, **kwargs):
if time.monotonic() < self.disabled_until:
return None
try:
return function(*args, **kwargs)
except redis.RedisError, OSError:
with self.lock:
self.disabled_until = time.monotonic() + 5
return None
def get(self, *, user_id: UUID) -> User | None:
raw = self._command(self.client.get, "user:" + str(user_id))
if raw is None:
return None
try:
row = json.loads(raw)
if UUID(row["id"]) != user_id:
return None
return User(
user_id,
row["username"],
row["display_name"],
datetime.fromisoformat(row["created_at"]),
)
except ValueError, TypeError, KeyError:
return None
def put(self, *, user: User):
raw = json.dumps(
{
"id": str(user.id),
"username": user.username,
"display_name": user.display_name,
"created_at": timestamp(user.created_at),
}
)
self._command(self.client.set, "user:" + str(user.id), raw, ex=30)
+64
View File
@@ -0,0 +1,64 @@
import base64
import binascii
import json
import re
from datetime import datetime
from uuid import UUID
from src.domain import Key, timestamp
from src.errors import invalid
def parse_uuid(value: str) -> UUID:
if not isinstance(value, str) or not re.fullmatch(
r"[0-9a-fA-F]{8}(?:-[0-9a-fA-F]{4}){3}-[0-9a-fA-F]{12}", value
):
raise invalid()
return UUID(value)
def encode_cursor(*, scope: str, key: Key) -> str:
payload = {
"v": 1,
"scope": scope,
"at": timestamp(key[0]),
"id": str(key[1]),
}
return (
base64.urlsafe_b64encode(
json.dumps(payload, separators=(",", ":")).encode()
)
.decode()
.rstrip("=")
)
def decode_cursor(*, value: str | None, scope: str) -> Key | None:
if value is None:
return None
try:
if (
not value
or len(value) > 2048
or not re.fullmatch(r"[A-Za-z0-9_-]+", value)
):
raise ValueError
data = json.loads(
base64.b64decode(
value + "=" * (-len(value) % 4), altchars=b"-_", validate=True
)
)
if (
not isinstance(data, dict)
or type(data.get("v")) is not int
or data["v"] != 1
or data.get("scope") != scope
or not isinstance(data.get("at"), str)
):
raise ValueError
at = datetime.fromisoformat(data["at"])
if at.tzinfo is None or timestamp(at) != data["at"]:
raise ValueError
return at, parse_uuid(data["id"])
except ValueError, TypeError, KeyError, binascii.Error, RecursionError:
raise invalid() from None
+69
View File
@@ -0,0 +1,69 @@
import time
from contextlib import contextmanager
import psycopg
from psycopg.rows import dict_row
from psycopg_pool import ConnectionPool, PoolTimeout
from src.errors import DomainError
class DeadlineConnection(psycopg.Connection):
"""Клиентский срок действует и при замороженном PostgreSQL."""
def wait(self, gen, interval=0.1, timeout=None):
deadline = getattr(self, "request_deadline", None)
remaining = (
max(0.001, deadline - time.monotonic()) if deadline else 3.0
)
bound = min(remaining, timeout) if timeout is not None else remaining
try:
return super().wait(gen, interval=interval, timeout=bound)
except psycopg.OperationalError:
self.close()
raise
class Database:
def __init__(self, *, url: str, size: int):
self.pool = ConnectionPool(
url,
connection_class=DeadlineConnection,
kwargs={
"autocommit": True,
"row_factory": dict_row,
"connect_timeout": 2,
"options": (
"-c statement_timeout=2500 -c lock_timeout=1000 "
"-c idle_in_transaction_session_timeout=6000"
),
},
min_size=1,
max_size=size,
timeout=1,
open=False,
)
def open(self):
self.pool.open(wait=True, timeout=10)
def close(self):
self.pool.close(timeout=2)
@contextmanager
def connection(self):
try:
with self.pool.connection(timeout=1) as conn:
conn.request_deadline = time.monotonic() + 4
try:
yield conn
finally:
conn.request_deadline = None
except psycopg.OperationalError, psycopg.InterfaceError, PoolTimeout:
raise DomainError(
503, "unavailable", "Database unavailable"
) from None
def ping(self):
with self.connection() as conn:
conn.execute("SELECT 1").fetchone()
+44
View File
@@ -0,0 +1,44 @@
from dataclasses import dataclass
from datetime import UTC, datetime
from uuid import UUID
def utcnow() -> datetime:
return datetime.now(UTC)
def timestamp(value: datetime) -> str:
return value.isoformat(timespec="microseconds").replace("+00:00", "Z")
@dataclass(frozen=True, slots=True)
class User:
id: UUID
username: str
display_name: str
created_at: datetime
@dataclass(frozen=True, slots=True)
class Chat:
id: UUID
title: str
member_ids: tuple[UUID, ...]
created_at: datetime
@dataclass(frozen=True, slots=True)
class Message:
id: UUID
chat_id: UUID
sender_id: UUID
text: str
created_at: datetime
type Record = User | Chat | Message
type Key = tuple[datetime, UUID]
def order_key(record: Record) -> Key:
return record.created_at, record.id
+16
View File
@@ -0,0 +1,16 @@
class DomainError(Exception):
def __init__(
self, status: int, code: str, message: str, headers=None
) -> None:
self.status = status
self.code = code
self.message = message
self.headers = headers or {}
def invalid() -> DomainError:
return DomainError(400, "invalid_request", "Invalid request")
def not_found() -> DomainError:
return DomainError(404, "not_found", "Resource not found")
+17
View File
@@ -0,0 +1,17 @@
import json
import logging
from datetime import UTC, datetime
class JsonFormatter(logging.Formatter):
def format(self, record: logging.LogRecord) -> str:
message = record.getMessage()
try:
event = json.loads(message)
if not isinstance(event, dict):
event = {"event": message}
except ValueError:
event = {"event": message}
event["timestamp"] = datetime.now(UTC).isoformat()
event["level"] = record.levelname.lower()
return json.dumps(event, ensure_ascii=False)

Some files were not shown because too many files have changed in this diff Show More