Лаба 5: Redis Streams очередь + асинхронная обработка изображений (оценка 4)

Переиспользован стек lab4 (LB/2 реплики/Postgres/SeaweedFS). upload теперь
202 queued; отдельный worker-процесс генерирует thumbnail/crop/optimized.
Transactional outbox закрывает разрыв COMMIT/publish; durable retry
(job_retry_schedule) переживает падение worker во время backoff; lease
recovery через XAUTOCLAIM; DLQ-реестр (job_failures) с явным operator
replay (scripts/replay_failed.py). make test/test-persistence/test-failover/
validate зелёные; в процессе экспериментов найдены и исправлены два
реальных бага (неполный try/except вокруг S3-записи вариантов, ошибка типов
asyncpg в построении interval) — см. REPORT.md и evidence/.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
2026-10-30 01:21:42 +03:00
co-authored by Claude Sonnet 5
parent 276dbdad8a
commit bdba93f022
31 changed files with 2255 additions and 17 deletions
+18
View File
@@ -0,0 +1,18 @@
FROM python:3.12.7-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY app ./app
COPY migrations ./migrations
RUN useradd --create-home --shell /usr/sbin/nologin appuser
USER appuser
EXPOSE 8080
HEALTHCHECK --interval=5s --timeout=3s --start-period=10s --retries=5 \
CMD python -c "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8080/health/live', timeout=2)" || exit 1
CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8080", "--timeout-graceful-shutdown", "5"]
+77 -17
View File
@@ -1,37 +1,97 @@
# Отчёт — лабораторная 5
- Автор, группа: TODO
- Целевая оценка; выбранный трек 5, если нужен: TODO
- Commit/PR: TODO после публикации
- Использованные источники, библиотеки и помощь ИИ: TODO
- Автор, группа: Сарибекян, БПИ2303
- Целевая оценка: 4
- Commit/PR: см. ветку `solution/БПИ2303-Сарибекян`
- Использованные источники, библиотеки и помощь ИИ: база lab4 (FastAPI/asyncpg/redis-py/Pillow/boto3/SeaweedFS/Nginx) плюс Redis Streams как брокер (тот же Redis, что и под сессиями). Реализация подготовлена с помощью AI-ассистента (Claude Code); проверено `make check/test/test-persistence/test-failover/validate` и серией ручных экспериментов отказов (см. `evidence/`).
## Запуск из чистого clone
TODO: версии runtime/образов, prerequisites, команды создания локальных secrets/.env, build, миграций, запуска, ожидания готовности, тестов и остановки. Указать необходимые действия оператора. Не вставлять секреты. Отдельно указать безопасную остановку и команду намеренного удаления учебных данных.
Версии: Python 3.12.7, PostgreSQL 16.4, Redis 7.4.1, Nginx 1.27.2, SeaweedFS 3.73.
```sh
git clone <ваш форк>
cd mtusi2026lab5
mkdir -p secrets
openssl rand -hex 20 > secrets/postgres_password.txt
printf 'postgresql://messenger:%s@postgres:5432/messenger' "$(cat secrets/postgres_password.txt)" > secrets/database_url.txt
printf 'redis://redis:6379/0' > secrets/cache_url.txt
printf 'messenger-api' > secrets/s3_access_key_id.txt
openssl rand -hex 20 > secrets/s3_secret_access_key.txt
AK=$(cat secrets/s3_access_key_id.txt); SK=$(cat secrets/s3_secret_access_key.txt)
sed -e "s/__S3_ACCESS_KEY_ID__/$AK/" -e "s/__S3_SECRET_ACCESS_KEY__/$SK/" \
infra/seaweedfs/s3.json.template > secrets/s3.json
chmod 600 secrets/*.txt secrets/s3.json
printf 'SESSION_TTL_SECONDS=3600\nMAX_UPLOAD_BYTES=10485760\nMAX_IMAGE_PIXELS=40000000\nJOB_MAX_ATTEMPTS=5\nJOB_TIMEOUT_SECONDS=60\n' > .env
docker compose --env-file .env config --quiet
docker compose --env-file .env up -d --build
curl -s localhost:8080/health/ready
make test
chmod +x scripts/restart.sh scripts/stop-one.sh
make test-persistence ACTION_SCRIPT=scripts/restart.sh
make test-failover ACTION_SCRIPT=scripts/stop-one.sh
# после отказа одной реплики: docker compose start api1 (или api2), повторить для второй.
```
Остановка: `docker compose stop`. Намеренное удаление: `docker compose down -v`.
## Архитектура и решения
TODO: схема процессов/хранилищ, источник истины, транзакционные границы, ограничения. Что перенесено из lab4, что изменено в этой работе. Соответствие переменных из .env.example реальной конфигурации.
```
Клиент → lb → api1/api2 → durable staging (= финальный S3-prefix) + attachments(status=queued) + job_outbox (1 транзакция Postgres)
│
▼ best-effort publish + фоновый relay (устраняет разрыв COMMIT↔publish)
Redis Streams (consumer group "workers")
│
▼
worker: thumbnail/crop/optimized → S3 + attachment_variants → status=ready
```
- **Брокер — Redis Streams на том же Redis**, что уже используется под сессии (AOF-персистентность уже настроена в lab2). Выбор задокументирован: это разрешённая контрактом альтернатива RabbitMQ/NATS JetStream, и переиспользование инфраструктуры избавляет от отдельного сервиса.
- **Staging = финальный S3-prefix**: контракт явно разрешает эту честную границу, если staging уже в S3 (см. `contracts/attachment-states.md`). Оригинал не "переезжает" между каталогами — он валидируется и сохраняется синхронно при upload (как в lab4), меняется только то, что готовность (`ready`) теперь выставляет **worker**, не API.
- **Transactional outbox** (`app/outbox.py`): `job_outbox` пишется в той же транзакции, что `INSERT` в `attachments`. После COMMIT — best-effort немедленная публикация; фоновый `relay_loop` (в каждой API-реплике, тик каждые 2с) публикует всё, что осталось неопубликованным дольше 1с. Проверено: искусственно созданная неопубликованная строка была подхвачена и доведена до `ready` без участия API-запроса (`evidence/outbox_gap.txt`).
- **Durable delayed retry**: Redis Streams не умеет отложенную доставку, поэтому backoff реализован через `job_retry_schedule` (Postgres) + `retry_scheduler_loop` в worker, а не через `sleep()` в процессе — переживает падение worker в любой момент backoff-окна.
- **Lease recovery**: `XAUTOCLAIM` (`broker.reclaim_stale`, `JOB_TIMEOUT_SECONDS`=60с) забирает незаподтверждённые сообщения у возможно умершего воркера.
- **Идемпотентность эффекта**: `process_job` — no-op для уже `ready`/`failed`; `attachment_variants` — `ON CONFLICT DO UPDATE`; финальный статус — условный `UPDATE ... WHERE status != 'ready'`.
- **DLQ упрощённо**: таблица `job_failures` (не отдельная очередь брокера) — читается оператором напрямую; explicit replay — `scripts/replay_failed.py`.
- **Совместимость с lab4**: `attachment_ids` в сообщении теперь может ссылаться на `queued/processing` вложения (не только `ready`), запрещён только `failed` (`409`) — согласно обновлённому контракту lab5.
## Доказательства критериев
| Критерий README | Команда/сценарий | Наблюдаемый результат | Файл доказательства |
| --- | --- | --- | --- |
| Публичный smoke | make test | TODO | TODO |
| Остальные пункты уровня 3 | TODO | TODO | TODO |
| Все пункты уровня 4, если заявлен | TODO | TODO | TODO |
| Выбранный трек 5, если заявлен | TODO | TODO | TODO |
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.
| upload 201→202 | `make test` | `PASS: lab5 public API scenario...` (включает ожидание ready, проверку thumbnail 256×160) | вывод команды |
| Персистентность | `make test-persistence ACTION_SCRIPT=scripts/restart.sh` | `PASS` | вывод команды |
| Failover (дважды) | `make test-failover ACTION_SCRIPT=scripts/stop-one.sh` | `successful_writes:101/103, failed_attempts:0` | вывод команды |
| Worker стоп→upload→API restart→worker старт→ready | ручной сценарий | `queued`→`409` на content→после старта worker: `ready` с 3 вариантами | `evidence/worker_resilience.txt` |
| Ограниченный retry с backoff+jitter | симулированный transient failure (`_handle_failure`, attempt=1) | `job_retry_schedule` с `next_attempt_at`+1.9с; автопланировщик republish; `attempt=2` → `ready` | `evidence/retry_backoff.txt` |
| DLQ/failed + явный replay | симулировано attempt=5 → `failed` | `job_failures` запись, `409` на content, `scripts/replay_failed.py` → `queued`→`ready` | `evidence/dlq_and_replay.txt` |
| Разрыв COMMIT/publish (outbox) | вставлена неопубликованная `job_outbox`-строка | фоновый relay опубликовал её сам, вложение дошло до `ready` | `evidence/outbox_gap.txt` |
| Повторная доставка ready-задачи — no-op | ручной повторный `process_job` на готовом вложении | лог `"already ready; no-op"`, данные не изменены | `evidence/outbox_gap.txt` |
| crop 128×128, optimized WebP (4) | тот же smoke/ручной upload | 3 варианта с верными размерами в ответе | `evidence/worker_resilience.txt`, `evidence/retry_backoff.txt` |
| `make validate` | `make validate PYTHON=.venv/bin/python` | `PASS` | вывод команды |
## Эксперименты и отказы
TODO: гипотеза, hardware/лимиты, версии, объём данных, профиль нагрузки, concurrency/arrival rate, длительность/прогрев, повторы, успешные RPS, ошибки, p50/p95/p99, CPU/RAM. Сохранить raw summary без tokens и объяснить ограничения измерения.
| Воздействие | Что ожидаем | Что наблюдали | Время восстановления / потеря данных |
| Воздействие | Что ожидаем | Что наблюдали | Восстановление / потеря данных |
| --- | --- | --- | --- |
| TODO | TODO | TODO | TODO |
| `docker compose stop worker`, upload, `docker compose restart api1 api2` (worker всё ещё выключен) | Вложение остаётся `queued`, API-рестарт не теряет job | подтверждено, `status: queued` после рестарта API | нет потерь; задача ждёт worker |
| `GET .../content` при `status=queued` | `409`, не `200`/`404` | `409 conflict: attachment is not ready yet` | — |
| Старт worker после простоя | Принятая ранее задача доходит до `ready` | `ready` с 3 вариантами через ~5с после старта | без повторной загрузки файла |
| Симулированный transient failure, attempt=1 | Durable retry schedule, не "тихая" потеря | `job_retry_schedule` создана, автопубликация через ~2с, `attempt=2` → `ready` | без потерь, с управляемой задержкой |
| Симулированное исчерпание попыток (attempt=5) | `failed` + DLQ-запись, `409` на content | подтверждено; `scripts/replay_failed.py` восстановил до `ready` | без потерь, восстановление — операторское решение |
| Неопубликованная `job_outbox`-строка (разрыв COMMIT/publish) | Фоновый relay публикует без вмешательства | задача дошла до `ready` автоматически | без потерь, без ручного создания нового задания |
| Повторный вызов `process_job` на `ready`-вложении | No-op, без дублей/изменений | лог `already ready; no-op` | — |
**Ограничение:** трек исследования уровня 5 (очередь под нагрузкой / доказанная идемпотентность с fault injection) не выполнялся — цель лабы — оценка 4. Нагрузочный тест `tests/load.py` на нескольких одновременных upload с измерением throughput конвейера не проводился отдельно от общего `make test`.
## Ограничения и следующие шаги
TODO: какие отказы выдерживаются и какие нет; что не реализовано; где учебное упрощение; какое улучшение подтверждено, а какое пока гипотеза.
- Найдено и исправлено **две реальные ошибки** в процессе ручных экспериментов (не поймались бы одним happy-path smoke):
1. Блок `try/except` в `process_job` изначально не покрывал загрузку сгенерированных вариантов в S3 — ошибка на этом шаге "сбегала" из структурированного retry-механизма. Исправлено расширением границ `try`.
2. SQL-ошибка типов `asyncpg` (`($1 || ' seconds')::interval` с параметром-`float`) в `app/worker.py` и `app/outbox.py` — заменено на `($1 * interval '1 second')`.
- `WORKER_CONCURRENCY` из `.env.example` не используется впрямую как числовой множитель параллельных задач — в текущей реализации один worker-процесс обрабатывает задания последовательно (`consume_loop` с `count=1`), конкурентность между задачами не реализована отдельно от конкурентности между контейнерами worker (которых здесь один). Для реальной параллельности нужно либо несколько worker-реплик, либо `asyncio.gather` нескольких consume-корутин в одном процессе — не сделано ввиду бюджета времени этой лабы.
- Очистка staging (`STAGING_RETENTION_SECONDS`) не реализована отдельным cron/job — поскольку staging совпадает с финальным хранилищем, "неудачные/непривязанные" объекты — это ровно та же категория orphan-объектов, что в lab4, и покрывается тем же `scripts/reconcile_attachments.py` (перенесён без изменений; здесь не переиспользован явно, но логика применима).
- Трек исследования уровня 5 не реализован.
View File
+68
View File
@@ -0,0 +1,68 @@
"""Durable job queue on Redis Streams (persisted via the same AOF-enabled
Redis used for sessions). A consumer group gives at-least-once delivery with
per-message acknowledgement and lease-based recovery of a crashed worker's
unacked entries (XAUTOCLAIM) -- Redis Streams' equivalent of RabbitMQ's
redelivery-on-no-ack.
"""
import json
import redis.asyncio as redis
STREAM = "attachment_jobs"
GROUP = "workers"
_client: redis.Redis | None = None
def init(url: str) -> None:
global _client
_client = redis.from_url(url, decode_responses=True, socket_timeout=5, socket_connect_timeout=5)
def client() -> redis.Redis:
if _client is None:
raise RuntimeError("broker is not initialized")
return _client
async def ensure_group() -> None:
try:
await client().xgroup_create(STREAM, GROUP, id="0", mkstream=True)
except redis.ResponseError as exc:
if "BUSYGROUP" not in str(exc):
raise
async def ping() -> bool:
try:
return bool(await client().ping())
except Exception: # noqa: BLE001 - readiness probe
return False
async def publish(job: dict) -> str:
return await client().xadd(STREAM, {"payload": json.dumps(job, ensure_ascii=False)})
async def read(consumer: str, count: int = 1, block_ms: int = 2000):
"""Return [(message_id, job_dict), ...] of newly delivered entries."""
resp = await client().xreadgroup(GROUP, consumer, {STREAM: ">"}, count=count, block=block_ms)
out = []
for _stream, entries in resp or []:
for message_id, fields in entries:
out.append((message_id, json.loads(fields["payload"])))
return out
async def ack(message_id: str) -> None:
await client().xack(STREAM, GROUP, message_id)
async def reclaim_stale(consumer: str, min_idle_ms: int, count: int = 10):
"""Hand this consumer any entries some other (likely dead) consumer has
held unacked for longer than min_idle_ms -- recovery for a worker that
died mid-processing, not just redelivery of a fresh message."""
_cursor, entries, _deleted = await client().xautoclaim(
STREAM, GROUP, consumer, min_idle_time=min_idle_ms, start_id="0-0", count=count
)
return [(message_id, json.loads(fields["payload"])) for message_id, fields in entries]
+82
View File
@@ -0,0 +1,82 @@
"""PostgreSQL connection pool and versioned migration runner."""
import logging
from pathlib import Path
import asyncpg
logger = logging.getLogger("app.db")
MIGRATIONS_DIR = Path(__file__).resolve().parent.parent / "migrations"
_pool: asyncpg.Pool | None = None
async def init_pool(dsn: str) -> asyncpg.Pool:
global _pool
_pool = await asyncpg.create_pool(
dsn,
min_size=2,
max_size=10,
command_timeout=5,
timeout=5,
)
await _run_migrations(_pool)
return _pool
async def close_pool() -> None:
global _pool
if _pool is not None:
await _pool.close()
_pool = None
def pool() -> asyncpg.Pool:
if _pool is None:
raise RuntimeError("database pool is not initialized")
return _pool
MIGRATION_LOCK_KEY = 783451 # arbitrary advisory-lock id, serializes concurrent replicas
async def _run_migrations(p: asyncpg.Pool) -> None:
async with p.acquire() as conn:
# Multiple API replicas start together; an advisory lock keeps migrations
# from racing each other instead of letting all replicas apply SQL at once.
await conn.execute("SELECT pg_advisory_lock($1)", MIGRATION_LOCK_KEY)
try:
await _apply_migrations(conn)
finally:
await conn.execute("SELECT pg_advisory_unlock($1)", MIGRATION_LOCK_KEY)
async def _apply_migrations(conn: asyncpg.Connection) -> None:
await conn.execute(
"""
CREATE TABLE IF NOT EXISTS schema_migrations (
version TEXT PRIMARY KEY,
applied_at TIMESTAMPTZ NOT NULL DEFAULT now()
)
"""
)
applied = {r["version"] for r in await conn.fetch("SELECT version FROM schema_migrations")}
for path in sorted(MIGRATIONS_DIR.glob("*.sql")):
if path.name in applied:
continue
sql = path.read_text(encoding="utf-8")
async with conn.transaction():
await conn.execute(sql)
await conn.execute(
"INSERT INTO schema_migrations (version) VALUES ($1)", path.name
)
logger.info("applied migration %s", path.name)
async def ping() -> bool:
try:
async with pool().acquire() as conn:
await conn.fetchval("SELECT 1")
return True
except Exception: # noqa: BLE001 - readiness probe, any failure means not ready
return False
+37
View File
@@ -0,0 +1,37 @@
"""Canonical error envelope: {"error": {"code", "message", "request_id"}}."""
class ApiError(Exception):
def __init__(self, status: int, code: str, message: str):
self.status = status
self.code = code
self.message = message
super().__init__(message)
def bad_request(message: str) -> ApiError:
return ApiError(400, "invalid_request", message)
def unauthorized(message: str = "authentication required") -> ApiError:
return ApiError(401, "unauthorized", message)
def not_found(message: str = "not found") -> ApiError:
return ApiError(404, "not_found", message)
def conflict(message: str) -> ApiError:
return ApiError(409, "conflict", message)
def unavailable(message: str = "dependency unavailable") -> ApiError:
return ApiError(503, "unavailable", message)
def payload_too_large(message: str) -> ApiError:
return ApiError(413, "payload_too_large", message)
def unsupported_media_type(message: str) -> ApiError:
return ApiError(415, "unsupported_media_type", message)
+89
View File
@@ -0,0 +1,89 @@
"""Shared (cross-replica) idempotency for POST message, scoped to (user, method, path, key)."""
import asyncio
import hashlib
import json
import time
import uuid
import asyncpg
IDEMPOTENCY_KEY_RE_MAXLEN = 128
WAIT_TIMEOUT_SECONDS = 5.0
POLL_INTERVAL_SECONDS = 0.05
class IdempotencyConflict(Exception):
"""Same key reused with a different request body."""
def request_hash(body: dict) -> str:
canonical = json.dumps(body, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
return hashlib.sha256(canonical.encode("utf-8")).hexdigest()
async def run_once(
conn: asyncpg.Connection,
user_id: uuid.UUID,
method: str,
path: str,
key: str,
body: dict,
perform,
):
"""Execute `perform()` exactly once per (user, method, path, key); replay stored result otherwise."""
digest = request_hash(body)
inserted = await conn.fetchrow(
"""
INSERT INTO idempotency_keys (user_id, method, path, idem_key, request_hash)
VALUES ($1, $2, $3, $4, $5)
ON CONFLICT (user_id, method, path, idem_key) DO NOTHING
RETURNING idem_key
""",
user_id, method, path, key, digest,
)
if inserted is None:
existing = await conn.fetchrow(
"SELECT request_hash, status, response_status, response_body FROM idempotency_keys "
"WHERE user_id=$1 AND method=$2 AND path=$3 AND idem_key=$4",
user_id, method, path, key,
)
if existing is None:
raise RuntimeError("idempotency row vanished")
if existing["request_hash"] != digest:
raise IdempotencyConflict()
return await _await_result(conn, user_id, method, path, key, existing)
try:
status_code, response_body = await perform()
except Exception:
# Free the slot so a genuine retry after a transient failure is not blocked forever.
await conn.execute(
"DELETE FROM idempotency_keys WHERE user_id=$1 AND method=$2 AND path=$3 AND idem_key=$4",
user_id, method, path, key,
)
raise
await conn.execute(
"""
UPDATE idempotency_keys
SET status='done', response_status=$5, response_body=$6
WHERE user_id=$1 AND method=$2 AND path=$3 AND idem_key=$4
""",
user_id, method, path, key, status_code, json.dumps(response_body),
)
return status_code, response_body
async def _await_result(conn, user_id, method, path, key, row):
if row["status"] == "done":
return row["response_status"], json.loads(row["response_body"])
deadline = time.monotonic() + WAIT_TIMEOUT_SECONDS
while time.monotonic() < deadline:
await asyncio.sleep(POLL_INTERVAL_SECONDS)
current = await conn.fetchrow(
"SELECT status, response_status, response_body FROM idempotency_keys "
"WHERE user_id=$1 AND method=$2 AND path=$3 AND idem_key=$4",
user_id, method, path, key,
)
if current and current["status"] == "done":
return current["response_status"], json.loads(current["response_body"])
raise TimeoutError("concurrent idempotent request did not finish in time")
+29
View File
@@ -0,0 +1,29 @@
"""Server-side image validation: real format by content, not by the client's claim."""
import io
from PIL import Image, UnidentifiedImageError
CONTENT_TYPE_TO_PIL_FORMAT = {"image/png": "PNG", "image/jpeg": "JPEG"}
class InvalidImage(Exception):
pass
def validate(data: bytes, declared_content_type: str, max_pixels: int) -> None:
"""Raise InvalidImage if the bytes are not a genuine, within-budget PNG/JPEG."""
expected_format = CONTENT_TYPE_TO_PIL_FORMAT.get(declared_content_type)
if expected_format is None:
raise InvalidImage("unsupported content type")
try:
with Image.open(io.BytesIO(data)) as img:
# Image.open() only parses the header; check the pixel budget
# BEFORE calling .load(), which is what actually decodes pixels.
if img.format != expected_format:
raise InvalidImage("declared content type does not match actual image format")
width, height = img.size
if width * height > max_pixels:
raise InvalidImage("image exceeds the decoded pixel budget")
img.load() # forces full decode; raises on truncated/corrupted data
except (UnidentifiedImageError, OSError, ValueError) as exc:
raise InvalidImage(str(exc)) from exc
+724
View File
@@ -0,0 +1,724 @@
"""Messenger API — lab 5: durable queue + async attachment processing."""
import asyncio
import base64
import binascii
import json
import os
import re
import socket
import uuid
from datetime import datetime, timezone
import asyncpg
from fastapi import FastAPI, Header, Request
from fastapi.responses import JSONResponse, Response
from pydantic import BaseModel, ConfigDict, Field, ValidationError
import hashlib
import os
from . import db, sessions, storage, images, broker, outbox
from . import idempotency
from .errors import (
ApiError, bad_request, conflict, not_found, payload_too_large,
unauthorized, unavailable, unsupported_media_type,
)
from .security import hash_password, verify_password
USERNAME_RE = re.compile(r"^[a-z0-9_]{3,32}$")
IDEMPOTENCY_KEY_RE = re.compile(r"^[A-Za-z0-9._:-]{1,128}$")
FILENAME_FORBIDDEN_RE = re.compile(r"[/\\\x00-\x1f]")
MAX_UPLOAD_BYTES = int(os.environ.get("MAX_UPLOAD_BYTES", "10485760"))
MAX_IMAGE_PIXELS = int(os.environ.get("MAX_IMAGE_PIXELS", "40000000"))
INSTANCE_ID = os.environ.get("INSTANCE_ID") or f"{socket.gethostname()}-{os.getpid()}"
app = FastAPI(title="MTUSI Messenger — lab 2")
# ---------------------------------------------------------------------------
# Request/response plumbing: request id, instance id, canonical error shape.
# ---------------------------------------------------------------------------
@app.middleware("http")
async def tagging_middleware(request: Request, call_next):
request_id = uuid.uuid4().hex
request.state.request_id = request_id
try:
response = await call_next(request)
except ApiError as exc:
response = _error_response(exc, request_id)
except Exception: # noqa: BLE001 - never leak internals
response = _error_response(ApiError(500, "internal_error", "internal error"), request_id)
response.headers["X-Request-Id"] = request_id
response.headers["X-Instance-Id"] = INSTANCE_ID
return response
def _error_response(exc: ApiError, request_id: str) -> JSONResponse:
headers = {}
if exc.status == 401:
path = "/api/v1/auth/sessions"
scheme = 'Basic realm="messenger", charset="UTF-8"' if _is_login_path(exc) else 'Bearer realm="messenger"'
headers["WWW-Authenticate"] = scheme
return JSONResponse(
status_code=exc.status,
content={"error": {"code": exc.code, "message": exc.message, "request_id": request_id}},
headers=headers,
)
def _is_login_path(exc: ApiError) -> bool:
return getattr(exc, "login_path", False)
@app.exception_handler(ApiError)
async def api_error_handler(request: Request, exc: ApiError):
return _error_response(exc, getattr(request.state, "request_id", uuid.uuid4().hex))
def iso(value: datetime) -> str:
return value.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
def user_json(record) -> dict:
return {
"id": str(record["id"]),
"username": record["username"],
"display_name": record["display_name"],
"created_at": iso(record["created_at"]),
}
def chat_json(record, member_ids: list[uuid.UUID]) -> dict:
return {
"id": str(record["id"]),
"title": record["title"],
"member_ids": [str(m) for m in member_ids],
"created_at": iso(record["created_at"]),
}
def message_json(record, attachment_ids: list[uuid.UUID]) -> dict:
return {
"id": str(record["id"]),
"chat_id": str(record["chat_id"]),
"sender_id": str(record["sender_id"]),
"text": record["text"],
"attachment_ids": [str(a) for a in attachment_ids],
"created_at": iso(record["created_at"]),
}
def attachment_json(record, variant_rows=()) -> dict:
return {
"id": str(record["id"]),
"chat_id": str(record["chat_id"]),
"owner_id": str(record["owner_id"]),
"filename": record["filename"],
"content_type": record["content_type"],
"size_bytes": record["size_bytes"],
"sha256": record["sha256"],
"status": record["status"],
"variants": [variant_json(v) for v in variant_rows],
"error_code": record["error_code"],
"created_at": iso(record["created_at"]),
"updated_at": iso(record["updated_at"]),
}
def variant_json(record) -> dict:
return {
"name": record["name"],
"content_type": record["content_type"],
"size_bytes": record["size_bytes"],
"width": record["width"],
"height": record["height"],
"sha256": record["sha256"],
}
# ---------------------------------------------------------------------------
# Cursors: opaque base64url(json) of the keyset tuple, scoped per resource.
# ---------------------------------------------------------------------------
def encode_cursor(payload: dict) -> str:
raw = json.dumps(payload, separators=(",", ":")).encode()
return base64.urlsafe_b64encode(raw).decode().rstrip("=")
def decode_cursor(value: str) -> dict:
try:
padded = value + "=" * (-len(value) % 4)
raw = base64.urlsafe_b64decode(padded.encode())
return json.loads(raw)
except (binascii.Error, ValueError, UnicodeDecodeError):
raise bad_request("malformed cursor")
# ---------------------------------------------------------------------------
# Auth: Basic only on POST /auth/sessions, Bearer everywhere else protected.
# ---------------------------------------------------------------------------
async def require_bearer(authorization: str | None) -> asyncpg.Record:
if not authorization or not authorization.startswith("Bearer "):
raise unauthorized()
token = authorization[len("Bearer "):].strip()
if not token:
raise unauthorized()
try:
user_id = await sessions.resolve(token)
except Exception: # noqa: BLE001 - dependency failure must not grant access
raise unavailable("session store unavailable")
if user_id is None:
raise unauthorized()
async with db.pool().acquire() as conn:
record = await conn.fetchrow("SELECT * FROM users WHERE id=$1", user_id)
if record is None:
raise unauthorized()
return record
def parse_basic(authorization: str | None) -> tuple[str, str]:
err = ApiError(401, "unauthorized", "Basic credentials required")
err.login_path = True
if not authorization or not authorization.startswith("Basic "):
raise err
try:
decoded = base64.b64decode(authorization[len("Basic "):].strip()).decode("utf-8")
except (binascii.Error, UnicodeDecodeError):
raise err
if ":" not in decoded:
raise err
username, password = decoded.split(":", 1)
return username, password
# ---------------------------------------------------------------------------
# Request bodies (additionalProperties: false in the contract -> extra="forbid").
# ---------------------------------------------------------------------------
class CreateUser(BaseModel):
model_config = ConfigDict(extra="forbid")
username: str
display_name: str = Field(min_length=1, max_length=100)
password: str = Field(min_length=12, max_length=128)
class CreateChat(BaseModel):
model_config = ConfigDict(extra="forbid")
title: str = Field(min_length=1, max_length=100)
member_ids: list[str] = Field(min_length=1, max_length=100)
class CreateMessage(BaseModel):
model_config = ConfigDict(extra="forbid")
text: str = Field(max_length=4000) # required key, "" allowed if attachments are present
attachment_ids: list[str] = Field(default_factory=list, max_length=10)
async def parse_body(request: Request, model):
try:
raw = await request.json()
except json.JSONDecodeError:
raise bad_request("malformed JSON")
try:
return model.model_validate(raw)
except ValidationError as exc:
raise bad_request(str(exc.errors()[0]["msg"]) if exc.errors() else "invalid body")
# ---------------------------------------------------------------------------
# Health
# ---------------------------------------------------------------------------
@app.get("/health/live")
async def health_live():
return {"status": "ok"}
@app.get("/health/ready")
async def health_ready():
if await db.ping() and await sessions.ping() and await storage.ping() and await broker.ping():
return {"status": "ok"}
raise unavailable("database, session store, object storage or broker not ready")
# ---------------------------------------------------------------------------
# Users
# ---------------------------------------------------------------------------
@app.post("/api/v1/users", status_code=201)
async def register_user(request: Request):
body = await parse_body(request, CreateUser)
if not USERNAME_RE.match(body.username):
raise bad_request("username must match ^[a-z0-9_]{3,32}$")
if not body.display_name.strip():
raise bad_request("display_name must not be blank")
password_hash = hash_password(body.password)
try:
async with db.pool().acquire() as conn:
record = await conn.fetchrow(
"""
INSERT INTO users (username, display_name, password_hash)
VALUES ($1, $2, $3)
RETURNING id, username, display_name, created_at
""",
body.username,
body.display_name,
password_hash,
)
except asyncpg.UniqueViolationError:
raise conflict("username already taken")
return JSONResponse(status_code=201, content=user_json(record))
@app.get("/api/v1/users/me")
async def current_user(authorization: str | None = Header(default=None)):
record = await require_bearer(authorization)
return user_json(record)
# ---------------------------------------------------------------------------
# Sessions
# ---------------------------------------------------------------------------
@app.post("/api/v1/auth/sessions", status_code=201)
async def create_session(authorization: str | None = Header(default=None)):
username, password = parse_basic(authorization)
async with db.pool().acquire() as conn:
record = await conn.fetchrow("SELECT * FROM users WHERE username=$1", username)
err = ApiError(401, "unauthorized", "invalid credentials")
err.login_path = True
if record is None or not verify_password(password, record["password_hash"]):
raise err
try:
token, expires_at = await sessions.create(record["id"])
except Exception: # noqa: BLE001
raise unavailable("session store unavailable")
return JSONResponse(
status_code=201,
content={
"token": token,
"token_type": "Bearer",
"expires_at": iso(expires_at),
"user": user_json(record),
},
)
@app.delete("/api/v1/auth/sessions/current", status_code=204)
async def revoke_session(authorization: str | None = Header(default=None)):
await require_bearer(authorization) # validates the token is currently live
token = authorization[len("Bearer "):].strip()
try:
await sessions.revoke(token)
except Exception: # noqa: BLE001
raise unavailable("session store unavailable")
return Response(status_code=204)
# ---------------------------------------------------------------------------
# Chats
# ---------------------------------------------------------------------------
@app.post("/api/v1/chats", status_code=201)
async def create_chat(request: Request, authorization: str | None = Header(default=None)):
actor = await require_bearer(authorization)
body = await parse_body(request, CreateChat)
try:
member_ids = [uuid.UUID(m) for m in body.member_ids]
except ValueError:
raise bad_request("member_ids must be UUIDs")
if len(set(member_ids)) != len(member_ids):
raise bad_request("member_ids must not contain duplicates")
if actor["id"] not in member_ids:
raise bad_request("caller must be a member of the chat")
async with db.pool().acquire() as conn:
async with conn.transaction():
existing = await conn.fetch(
"SELECT id FROM users WHERE id = ANY($1::uuid[])", member_ids
)
if len(existing) != len(member_ids):
raise bad_request("unknown member in member_ids")
chat = await conn.fetchrow(
"INSERT INTO chats (title) VALUES ($1) RETURNING id, title, created_at",
body.title,
)
await conn.executemany(
"INSERT INTO chat_members (chat_id, user_id) VALUES ($1, $2)",
[(chat["id"], m) for m in member_ids],
)
return JSONResponse(status_code=201, content=chat_json(chat, member_ids))
@app.get("/api/v1/chats")
async def list_chats(
authorization: str | None = Header(default=None),
limit: int = 50,
cursor: str | None = None,
):
actor = await require_bearer(authorization)
if not (1 <= limit <= 100):
raise bad_request("limit must be between 1 and 100")
after_created_at = after_id = None
if cursor:
payload = decode_cursor(cursor)
try:
after_created_at = datetime.fromisoformat(payload["created_at"])
after_id = uuid.UUID(payload["id"])
except (KeyError, ValueError):
raise bad_request("malformed cursor")
async with db.pool().acquire() as conn:
rows = await conn.fetch(
"""
SELECT c.id, c.title, c.created_at
FROM chats c
JOIN chat_members cm ON cm.chat_id = c.id
WHERE cm.user_id = $1
AND ($2::timestamptz IS NULL OR (c.created_at, c.id) > ($2, $3))
ORDER BY c.created_at ASC, c.id ASC
LIMIT $4
""",
actor["id"],
after_created_at,
after_id,
limit + 1,
)
has_more = len(rows) > limit
rows = rows[:limit]
items = []
for row in rows:
members = await conn.fetch("SELECT user_id FROM chat_members WHERE chat_id=$1", row["id"])
items.append(chat_json(row, [m["user_id"] for m in members]))
next_cursor = None
if has_more:
last = rows[-1]
next_cursor = encode_cursor({"created_at": iso(last["created_at"]), "id": str(last["id"])})
return {"items": items, "next_cursor": next_cursor}
async def _load_chat_if_member(conn, chat_id: uuid.UUID, user_id: uuid.UUID):
chat = await conn.fetchrow("SELECT id, title, created_at FROM chats WHERE id=$1", chat_id)
if chat is None:
raise not_found("chat not found")
membership = await conn.fetchrow(
"SELECT 1 FROM chat_members WHERE chat_id=$1 AND user_id=$2", chat_id, user_id
)
if membership is None:
raise not_found("chat not found")
return chat
@app.get("/api/v1/chats/{chat_id}")
async def get_chat(chat_id: str, authorization: str | None = Header(default=None)):
actor = await require_bearer(authorization)
try:
chat_uuid = uuid.UUID(chat_id)
except ValueError:
raise bad_request("chat_id must be a UUID")
async with db.pool().acquire() as conn:
chat = await _load_chat_if_member(conn, chat_uuid, actor["id"])
members = await conn.fetch("SELECT user_id FROM chat_members WHERE chat_id=$1", chat_uuid)
return chat_json(chat, [m["user_id"] for m in members])
# ---------------------------------------------------------------------------
# Messages
# ---------------------------------------------------------------------------
@app.post("/api/v1/chats/{chat_id}/messages", status_code=201)
async def create_message(
chat_id: str,
request: Request,
authorization: str | None = Header(default=None),
idempotency_key: str | None = Header(default=None, alias="Idempotency-Key"),
):
# Session/access are re-checked on every call, including idempotent replays.
actor = await require_bearer(authorization)
try:
chat_uuid = uuid.UUID(chat_id)
except ValueError:
raise bad_request("chat_id must be a UUID")
if not idempotency_key or not IDEMPOTENCY_KEY_RE.match(idempotency_key):
raise bad_request("Idempotency-Key header is required and must match ^[A-Za-z0-9._:-]{1,128}$")
body = await parse_body(request, CreateMessage)
try:
attachment_uuids = [uuid.UUID(a) for a in body.attachment_ids]
except ValueError:
raise bad_request("attachment_ids must be UUIDs")
if len(set(attachment_uuids)) != len(attachment_uuids):
raise bad_request("attachment_ids must not contain duplicates")
if not body.text.strip() and not attachment_uuids:
raise bad_request("text must not be blank unless attachments are present")
async def perform():
record = await conn.fetchrow(
"""
INSERT INTO messages (chat_id, sender_id, text)
VALUES ($1, $2, $3)
RETURNING id, chat_id, sender_id, text, created_at
""",
chat_uuid,
actor["id"],
body.text,
)
if attachment_uuids:
owned = await conn.fetch(
"SELECT id, status FROM attachments WHERE chat_id=$1 AND id = ANY($2::uuid[])",
chat_uuid, attachment_uuids,
)
if len(owned) != len(attachment_uuids):
raise not_found("one or more attachment_ids are not available in this chat")
if any(r["status"] == "failed" for r in owned):
raise conflict("one or more attachment_ids reference a failed attachment")
await conn.executemany(
"INSERT INTO message_attachments (message_id, attachment_id) VALUES ($1, $2)",
[(record["id"], a) for a in attachment_uuids],
)
return 201, message_json(record, attachment_uuids)
async with db.pool().acquire() as conn:
await _load_chat_if_member(conn, chat_uuid, actor["id"])
async with conn.transaction():
try:
status_code, payload = await idempotency.run_once(
conn,
actor["id"],
"POST",
request.url.path,
idempotency_key,
{"text": body.text, "attachment_ids": sorted(body.attachment_ids)},
perform,
)
except idempotency.IdempotencyConflict:
raise conflict("Idempotency-Key reused with a different request body")
return JSONResponse(status_code=status_code, content=payload)
@app.get("/api/v1/chats/{chat_id}/messages")
async def list_messages(
chat_id: str,
authorization: str | None = Header(default=None),
limit: int = 50,
cursor: str | None = None,
):
actor = await require_bearer(authorization)
try:
chat_uuid = uuid.UUID(chat_id)
except ValueError:
raise bad_request("chat_id must be a UUID")
if not (1 <= limit <= 100):
raise bad_request("limit must be between 1 and 100")
after_created_at = after_id = None
if cursor:
payload = decode_cursor(cursor)
if payload.get("chat_id") != chat_id:
raise bad_request("cursor does not belong to this chat")
try:
after_created_at = datetime.fromisoformat(payload["created_at"])
after_id = uuid.UUID(payload["id"])
except (KeyError, ValueError):
raise bad_request("malformed cursor")
async with db.pool().acquire() as conn:
await _load_chat_if_member(conn, chat_uuid, actor["id"])
rows = await conn.fetch(
"""
SELECT id, chat_id, sender_id, text, created_at
FROM messages
WHERE chat_id = $1
AND ($2::timestamptz IS NULL OR (created_at, id) > ($2, $3))
ORDER BY created_at ASC, id ASC
LIMIT $4
""",
chat_uuid,
after_created_at,
after_id,
limit + 1,
)
has_more = len(rows) > limit
rows = rows[:limit]
items = []
for r in rows:
attached = await conn.fetch(
"SELECT attachment_id FROM message_attachments WHERE message_id=$1", r["id"]
)
items.append(message_json(r, [a["attachment_id"] for a in attached]))
next_cursor = None
if has_more:
last = rows[-1]
next_cursor = encode_cursor(
{"chat_id": chat_id, "created_at": iso(last["created_at"]), "id": str(last["id"])}
)
return {"items": items, "next_cursor": next_cursor}
# ---------------------------------------------------------------------------
# Attachments
# ---------------------------------------------------------------------------
@app.post("/api/v1/chats/{chat_id}/attachments", status_code=201)
async def upload_attachment(
chat_id: str,
request: Request,
filename: str,
authorization: str | None = Header(default=None),
):
actor = await require_bearer(authorization)
try:
chat_uuid = uuid.UUID(chat_id)
except ValueError:
raise bad_request("chat_id must be a UUID")
if not (1 <= len(filename) <= 255) or FILENAME_FORBIDDEN_RE.search(filename):
raise bad_request("filename must be 1-255 characters with no path separators or control characters")
content_type = request.headers.get("content-type", "")
if content_type not in ("image/png", "image/jpeg"):
raise unsupported_media_type("only image/png and image/jpeg are accepted")
# Enforce the byte budget BEFORE reading an unbounded body into memory.
data = bytearray()
async for chunk in request.stream():
data.extend(chunk)
if len(data) > MAX_UPLOAD_BYTES:
raise payload_too_large(f"upload exceeds {MAX_UPLOAD_BYTES} bytes")
data = bytes(data)
if not data:
raise bad_request("empty upload body")
try:
images.validate(data, content_type, MAX_IMAGE_PIXELS)
except images.InvalidImage as exc:
raise unsupported_media_type(str(exc))
digest = hashlib.sha256(data).hexdigest()
# Staging IS the final S3 prefix here (same bucket/prefix the worker later
# reads from) -- there is no separate move step, which is an honest
# simplification the contract explicitly allows when staging is already S3.
object_key = storage.new_object_key()
pool = db.pool()
async with pool.acquire() as conn:
await _load_chat_if_member(conn, chat_uuid, actor["id"])
try:
await storage.put_object(object_key, data, content_type)
except Exception: # noqa: BLE001
raise unavailable("object storage unavailable")
async with conn.transaction():
record = await conn.fetchrow(
"""
INSERT INTO attachments (chat_id, owner_id, filename, content_type, size_bytes, sha256, status, s3_key)
VALUES ($1, $2, $3, $4, $5, $6, 'queued', $7)
RETURNING id, chat_id, owner_id, filename, content_type, size_bytes, sha256, status, error_code, created_at, updated_at
""",
chat_uuid, actor["id"], filename, content_type, len(data), digest, object_key,
)
job = await outbox.enqueue(conn, record["id"])
# Best-effort immediate publish; if this process dies right here, the
# outbox relay (any replica) still publishes it from the committed row.
await outbox.publish_best_effort(pool, job)
return JSONResponse(status_code=202, content=attachment_json(record))
async def _load_attachment_if_member(conn, attachment_id: uuid.UUID, user_id: uuid.UUID):
record = await conn.fetchrow(
"""
SELECT a.* FROM attachments a
JOIN chat_members cm ON cm.chat_id = a.chat_id
WHERE a.id = $1 AND cm.user_id = $2
""",
attachment_id, user_id,
)
if record is None:
raise not_found("attachment not found")
return record
@app.get("/api/v1/attachments/{attachment_id}")
async def get_attachment(attachment_id: str, authorization: str | None = Header(default=None)):
actor = await require_bearer(authorization)
try:
att_uuid = uuid.UUID(attachment_id)
except ValueError:
raise bad_request("attachment_id must be a UUID")
async with db.pool().acquire() as conn:
record = await _load_attachment_if_member(conn, att_uuid, actor["id"])
variant_rows = await conn.fetch(
"SELECT name, content_type, size_bytes, width, height, sha256 FROM attachment_variants WHERE attachment_id=$1",
att_uuid,
)
return attachment_json(record, variant_rows)
@app.get("/api/v1/attachments/{attachment_id}/content")
async def get_attachment_content(
attachment_id: str,
authorization: str | None = Header(default=None),
variant: str = "original",
):
actor = await require_bearer(authorization)
try:
att_uuid = uuid.UUID(attachment_id)
except ValueError:
raise bad_request("attachment_id must be a UUID")
if variant not in ("original", "thumbnail", "crop", "optimized"):
raise bad_request("unknown variant")
async with db.pool().acquire() as conn:
record = await _load_attachment_if_member(conn, att_uuid, actor["id"])
if record["status"] in ("queued", "processing"):
raise conflict("attachment is not ready yet")
if record["status"] == "failed":
raise conflict("attachment processing failed")
if variant == "original":
s3_key, content_type = record["s3_key"], record["content_type"]
else:
variant_row = await conn.fetchrow(
"SELECT s3_key, content_type FROM attachment_variants WHERE attachment_id=$1 AND name=$2",
att_uuid, variant,
)
if variant_row is None:
raise not_found(f"variant {variant} is not available for this attachment")
s3_key, content_type = variant_row["s3_key"], variant_row["content_type"]
try:
data = await storage.get_object(s3_key)
except Exception: # noqa: BLE001
raise unavailable("object storage unavailable")
return Response(content=data, media_type=content_type)
# ---------------------------------------------------------------------------
# Lifecycle
# ---------------------------------------------------------------------------
@app.on_event("startup")
async def on_startup():
database_url = _read_secret("DATABASE_URL_FILE", "DATABASE_URL")
cache_url = _read_secret("CACHE_URL_FILE", "CACHE_URL")
await db.init_pool(database_url)
sessions.init(cache_url)
storage.init(
endpoint=os.environ["S3_ENDPOINT"],
region=os.environ.get("S3_REGION", "us-east-1"),
access_key=_read_secret("S3_ACCESS_KEY_ID_FILE", "S3_ACCESS_KEY_ID"),
secret_key=_read_secret("S3_SECRET_ACCESS_KEY_FILE", "S3_SECRET_ACCESS_KEY"),
bucket=os.environ.get("S3_BUCKET", "messenger-attachments"),
)
await storage.ensure_bucket()
broker.init(cache_url)
await broker.ensure_group()
app.state.outbox_relay_task = asyncio.create_task(outbox.relay_loop(db.pool()))
@app.on_event("shutdown")
async def on_shutdown():
app.state.outbox_relay_task.cancel()
await db.close_pool()
await sessions.close()
def _read_secret(file_env: str, inline_env: str) -> str:
path = os.environ.get(file_env)
if path and os.path.isfile(path):
return open(path, encoding="utf-8").read().strip()
value = os.environ.get(inline_env)
if not value:
raise RuntimeError(f"{file_env} or {inline_env} must be set")
return value
+91
View File
@@ -0,0 +1,91 @@
"""Transactional outbox: bridges the PostgreSQL COMMIT and the broker publish.
enqueue() writes the job row in the SAME transaction as the attachments
INSERT, so either both happen or neither does. The caller then makes a
best-effort immediate publish; if the process crashes between COMMIT and
that publish, the row is still there with published_at IS NULL, and the
background relay (any API replica) picks it up and publishes it later --
no job is silently lost, and none is invented without a committed row.
"""
import asyncio
import datetime as dt
import logging
import uuid
import asyncpg
from . import broker
logger = logging.getLogger("app.outbox")
RELAY_INTERVAL_SECONDS = 2
RELAY_GRACE_SECONDS = 1 # don't race the enqueuing request's own immediate publish
async def enqueue(conn: asyncpg.Connection, attachment_id: uuid.UUID) -> dict:
job_id = uuid.uuid4()
await conn.execute(
"INSERT INTO job_outbox (id, attachment_id, job_id, pipeline_version, attempt) "
"VALUES (gen_random_uuid(), $1, $2, 1, 1)",
attachment_id, job_id,
)
return {
"schema_version": 1,
"job_id": str(job_id),
"type": "attachment.process.v1",
"attachment_id": str(attachment_id),
"pipeline_version": 1,
"attempt": 1,
"created_at": dt.datetime.now(dt.timezone.utc).isoformat().replace("+00:00", "Z"),
}
async def publish_best_effort(pool: asyncpg.Pool, job: dict) -> None:
"""Try to publish right away; on failure, leave it for the relay. Never
raises -- a broker hiccup here must not fail the already-committed upload."""
try:
await broker.publish(job)
async with pool.acquire() as conn:
await conn.execute(
"UPDATE job_outbox SET published_at=now() WHERE job_id=$1", uuid.UUID(job["job_id"])
)
except Exception: # noqa: BLE001
logger.warning("immediate publish failed for job %s; relay will retry", job["job_id"])
async def relay_once(pool: asyncpg.Pool) -> int:
async with pool.acquire() as conn:
rows = await conn.fetch(
"""
SELECT id, attachment_id, job_id, pipeline_version, attempt, created_at
FROM job_outbox
WHERE published_at IS NULL AND created_at < now() - ($1 * interval '1 second')
ORDER BY created_at
LIMIT 50
""",
RELAY_GRACE_SECONDS,
)
for row in rows:
job = {
"schema_version": 1,
"job_id": str(row["job_id"]),
"type": "attachment.process.v1",
"attachment_id": str(row["attachment_id"]),
"pipeline_version": row["pipeline_version"],
"attempt": row["attempt"],
"created_at": row["created_at"].astimezone(dt.timezone.utc).isoformat().replace("+00:00", "Z"),
}
await broker.publish(job)
await conn.execute("UPDATE job_outbox SET published_at=now() WHERE id=$1", row["id"])
return len(rows)
async def relay_loop(pool: asyncpg.Pool) -> None:
while True:
try:
await relay_once(pool)
except asyncio.CancelledError:
raise
except Exception: # noqa: BLE001 - the relay must survive transient DB/broker hiccups
logger.exception("outbox relay tick failed")
await asyncio.sleep(RELAY_INTERVAL_SECONDS)
+57
View File
@@ -0,0 +1,57 @@
"""Worker-side variant generation: thumbnail (grade 3), crop + optimized (grade 4)."""
import io
from PIL import Image, ImageOps
def _normalize(img: Image.Image) -> Image.Image:
"""Apply EXIF orientation, then flatten any alpha onto white."""
img = ImageOps.exif_transpose(img)
if img.mode in ("RGBA", "LA", "P"):
rgba = img.convert("RGBA")
background = Image.new("RGB", rgba.size, (255, 255, 255))
background.paste(rgba, mask=rgba.split()[-1])
return background
return img.convert("RGB")
def thumbnail(data: bytes) -> tuple[bytes, int, int]:
"""JPEG, contained within 256x256, never upscaled, rounded down, min 1px."""
with Image.open(io.BytesIO(data)) as raw:
img = _normalize(raw)
scale = min(256 / img.width, 256 / img.height, 1.0)
width = max(1, int(img.width * scale))
height = max(1, int(img.height * scale))
resized = img.resize((width, height), Image.LANCZOS)
buf = io.BytesIO()
resized.save(buf, "JPEG", quality=85)
return buf.getvalue(), width, height
def crop(data: bytes) -> tuple[bytes, int, int]:
"""JPEG, centered 128x128 square; small sources may be upscaled."""
with Image.open(io.BytesIO(data)) as raw:
img = _normalize(raw)
scale = 128 / min(img.width, img.height)
width = max(128, round(img.width * scale))
height = max(128, round(img.height * scale))
resized = img.resize((width, height), Image.LANCZOS)
left = (width - 128) // 2
top = (height - 128) // 2
cropped = resized.crop((left, top, left + 128, top + 128))
buf = io.BytesIO()
cropped.save(buf, "JPEG", quality=85)
return buf.getvalue(), 128, 128
def optimized(data: bytes) -> tuple[bytes, int, int]:
"""WebP, contained within 1280x1280, never upscaled, EXIF stripped."""
with Image.open(io.BytesIO(data)) as raw:
img = _normalize(raw)
scale = min(1280 / img.width, 1280 / img.height, 1.0)
width = max(1, int(img.width * scale))
height = max(1, int(img.height * scale))
resized = img.resize((width, height), Image.LANCZOS)
buf = io.BytesIO()
resized.save(buf, "WEBP", quality=80) # save() never copies source EXIF for WEBP
return buf.getvalue(), width, height
+16
View File
@@ -0,0 +1,16 @@
"""Password KDF (Argon2id) — never store or compare plaintext passwords."""
from argon2 import PasswordHasher
from argon2.exceptions import VerifyMismatchError, InvalidHashError
_hasher = PasswordHasher()
def hash_password(password: str) -> str:
return _hasher.hash(password)
def verify_password(password: str, password_hash: str) -> bool:
try:
return _hasher.verify(password_hash, password)
except (VerifyMismatchError, InvalidHashError):
return False
+54
View File
@@ -0,0 +1,54 @@
"""Opaque Bearer sessions backed by Redis/Valkey, TTL-bound, AOF-persistent."""
import json
import os
import secrets
import uuid
from datetime import datetime, timedelta, timezone
import redis.asyncio as redis
_client: redis.Redis | None = None
TTL_SECONDS = int(os.environ.get("SESSION_TTL_SECONDS", "3600"))
def init(url: str) -> None:
global _client
_client = redis.from_url(url, decode_responses=True, socket_timeout=3, socket_connect_timeout=3)
async def close() -> None:
if _client is not None:
await _client.close()
def client() -> redis.Redis:
if _client is None:
raise RuntimeError("session store is not initialized")
return _client
async def ping() -> bool:
try:
return bool(await client().ping())
except Exception: # noqa: BLE001 - readiness probe
return False
async def create(user_id: uuid.UUID) -> tuple[str, datetime]:
token = secrets.token_urlsafe(32) # 256 bits of entropy, opaque and non-guessable
expires_at = datetime.now(timezone.utc) + timedelta(seconds=TTL_SECONDS)
payload = json.dumps({"user_id": str(user_id), "expires_at": expires_at.isoformat()})
await client().set(f"session:{token}", payload, ex=TTL_SECONDS)
return token, expires_at
async def resolve(token: str) -> uuid.UUID | None:
raw = await client().get(f"session:{token}")
if raw is None:
return None
data = json.loads(raw)
return uuid.UUID(data["user_id"])
async def revoke(token: str) -> None:
await client().delete(f"session:{token}")
+63
View File
@@ -0,0 +1,63 @@
"""Private S3-compatible object storage for attachment bytes (MinIO behind the scenes).
boto3 is synchronous; calls are pushed to a thread so the event loop is not
blocked by S3 I/O. Object keys are server-generated UUIDs, never the
user-supplied filename (filename is display-only, see contracts/README.md).
"""
import asyncio
import io
import uuid
import boto3
from botocore.client import Config
from botocore.exceptions import ClientError
_client = None
_bucket: str | None = None
def init(endpoint: str, region: str, access_key: str, secret_key: str, bucket: str) -> None:
global _client, _bucket
_bucket = bucket
_client = boto3.client(
"s3",
endpoint_url=endpoint,
region_name=region,
aws_access_key_id=access_key,
aws_secret_access_key=secret_key,
config=Config(signature_version="s3v4", retries={"max_attempts": 2}),
)
async def ensure_bucket() -> None:
"""The API identity only has Read/Write/List on this one bucket, not
bucket-admin rights, so CreateBucket is not an option. SeaweedFS (like S3)
auto-creates the bucket on the first successful PutObject, so writing an
empty marker object both creates the bucket and proves write access."""
await asyncio.to_thread(_client.put_object, Bucket=_bucket, Key=".bootstrap", Body=b"")
async def ping() -> bool:
try:
await asyncio.to_thread(_client.head_object, Bucket=_bucket, Key=".bootstrap")
return True
except Exception: # noqa: BLE001 - readiness probe
return False
def new_object_key() -> str:
return f"attachments/{uuid.uuid4().hex}"
async def put_object(key: str, data: bytes, content_type: str) -> None:
await asyncio.to_thread(
_client.put_object, Bucket=_bucket, Key=key, Body=data, ContentType=content_type
)
async def get_object(key: str) -> bytes:
def _get():
resp = _client.get_object(Bucket=_bucket, Key=key)
return resp["Body"].read()
return await asyncio.to_thread(_get)
+201
View File
@@ -0,0 +1,201 @@
"""Standalone worker process: consumes attachment.process.v1 jobs, generates
variants, writes results idempotently. Runs as its own container/command,
separate from the API (contracts/attachment-states.md requires this split).
"""
import asyncio
import hashlib
import logging
import os
import random
import socket
import uuid
from . import broker, db, pipeline, storage
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
logger = logging.getLogger("worker")
CONSUMER_NAME = os.environ.get("INSTANCE_ID") or f"{socket.gethostname()}-{os.getpid()}"
MAX_ATTEMPTS = int(os.environ.get("JOB_MAX_ATTEMPTS", "5"))
JOB_TIMEOUT_SECONDS = int(os.environ.get("JOB_TIMEOUT_SECONDS", "60"))
RETRY_SCAN_INTERVAL_SECONDS = 1
def read_secret(file_env: str, inline_env: str) -> str:
path = os.environ.get(file_env)
if path and os.path.isfile(path):
return open(path, encoding="utf-8").read().strip()
value = os.environ.get(inline_env)
if not value:
raise RuntimeError(f"{file_env} or {inline_env} must be set")
return value
async def process_job(job: dict) -> None:
attachment_id = uuid.UUID(job["attachment_id"])
job_id = uuid.UUID(job["job_id"])
attempt = job["attempt"]
async with db.pool().acquire() as conn:
record = await conn.fetchrow("SELECT * FROM attachments WHERE id=$1", attachment_id)
if record is None:
logger.warning("job %s references missing attachment %s; dropping", job_id, attachment_id)
return
if record["status"] in ("ready", "failed"):
# Redelivery of an already-finished job, or a stale duplicate racing
# a newer attempt: never roll a finished result backwards.
logger.info("attachment %s already %s; no-op", attachment_id, record["status"])
return
async with db.pool().acquire() as conn:
await conn.execute(
"UPDATE attachments SET status='processing', updated_at=now() WHERE id=$1 AND status != 'ready'",
attachment_id,
)
content_types = {"thumbnail": "image/jpeg", "crop": "image/jpeg", "optimized": "image/webp"}
try:
original = await storage.get_object(record["s3_key"])
generated = {
"thumbnail": pipeline.thumbnail(original),
"crop": pipeline.crop(original),
"optimized": pipeline.optimized(original),
}
uploaded = []
for name, (data, width, height) in generated.items():
key = f"attachments/{attachment_id}/{name}"
await storage.put_object(key, data, content_types[name]) # network call: can fail transiently too
uploaded.append((name, key, data, width, height))
except Exception: # noqa: BLE001 - covers S3 reachability (get+put) and image decode, not just decode
logger.exception("processing failed for attachment %s", attachment_id)
await _handle_failure(job, "processing_failed")
return
async with db.pool().acquire() as conn:
async with conn.transaction():
for name, key, data, width, height in uploaded:
digest = hashlib.sha256(data).hexdigest()
await conn.execute(
"""
INSERT INTO attachment_variants (attachment_id, name, content_type, size_bytes, width, height, sha256, s3_key)
VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
ON CONFLICT (attachment_id, name) DO UPDATE SET
content_type=EXCLUDED.content_type, size_bytes=EXCLUDED.size_bytes,
width=EXCLUDED.width, height=EXCLUDED.height, sha256=EXCLUDED.sha256, s3_key=EXCLUDED.s3_key
""",
attachment_id, name, content_types[name], len(data), width, height, digest, key,
)
await conn.execute(
"UPDATE attachments SET status='ready', error_code=NULL, updated_at=now() WHERE id=$1 AND status != 'ready'",
attachment_id,
)
logger.info("attachment %s ready (attempt %d)", attachment_id, attempt)
async def _handle_failure(job: dict, error_code: str) -> None:
attachment_id = uuid.UUID(job["attachment_id"])
job_id = uuid.UUID(job["job_id"])
attempt = job["attempt"]
async with db.pool().acquire() as conn:
if attempt >= MAX_ATTEMPTS:
async with conn.transaction():
await conn.execute(
"UPDATE attachments SET status='failed', error_code=$2, updated_at=now() WHERE id=$1",
attachment_id, error_code,
)
await conn.execute(
"INSERT INTO job_failures (attachment_id, job_id, attempt, error_code) VALUES ($1,$2,$3,$4) "
"ON CONFLICT (attachment_id, job_id) DO NOTHING",
attachment_id, job_id, attempt, error_code,
)
logger.warning("attachment %s permanently failed after %d attempts: %s", attachment_id, attempt, error_code)
return
# Transient: schedule a durable delayed retry instead of sleeping in-process.
delay = min(30, 2 ** (attempt - 1)) + random.uniform(0, 1)
await conn.execute(
"""
INSERT INTO job_retry_schedule (attachment_id, job_id, attempt, next_attempt_at)
VALUES ($1, $2, $3, now() + ($4 * interval '1 second'))
ON CONFLICT (attachment_id, job_id) DO UPDATE SET
attempt=EXCLUDED.attempt, next_attempt_at=EXCLUDED.next_attempt_at
""",
attachment_id, job_id, attempt + 1, delay,
)
await conn.execute(
"UPDATE attachments SET status='queued', updated_at=now() WHERE id=$1 AND status != 'ready'",
attachment_id,
)
logger.info("attachment %s scheduled for retry #%d in %.1fs", attachment_id, attempt + 1, delay)
async def retry_scheduler_loop() -> None:
while True:
try:
async with db.pool().acquire() as conn:
due = await conn.fetch(
"SELECT attachment_id, job_id, attempt FROM job_retry_schedule WHERE next_attempt_at <= now() LIMIT 50"
)
for row in due:
job = {
"schema_version": 1,
"job_id": str(row["job_id"]),
"type": "attachment.process.v1",
"attachment_id": str(row["attachment_id"]),
"pipeline_version": 1,
"attempt": row["attempt"],
"created_at": "1970-01-01T00:00:00Z", # informational only; not used for ordering
}
await broker.publish(job)
await conn.execute(
"DELETE FROM job_retry_schedule WHERE attachment_id=$1 AND job_id=$2",
row["attachment_id"], row["job_id"],
)
except Exception: # noqa: BLE001 - scheduler must survive a transient hiccup
logger.exception("retry scheduler tick failed")
await asyncio.sleep(RETRY_SCAN_INTERVAL_SECONDS)
async def reclaim_loop() -> None:
"""Hand this worker any job another (likely dead) worker left unacked
past JOB_TIMEOUT_SECONDS -- lease recovery for a crashed worker."""
while True:
try:
for message_id, job in await broker.reclaim_stale(CONSUMER_NAME, JOB_TIMEOUT_SECONDS * 1000):
logger.info("reclaimed stale job %s (message %s)", job["job_id"], message_id)
await process_job(job)
await broker.ack(message_id)
except Exception: # noqa: BLE001
logger.exception("reclaim tick failed")
await asyncio.sleep(JOB_TIMEOUT_SECONDS / 2)
async def consume_loop() -> None:
while True:
try:
for message_id, job in await broker.read(CONSUMER_NAME, count=1, block_ms=2000):
await process_job(job)
await broker.ack(message_id)
except Exception: # noqa: BLE001 - keep the loop alive across transient errors
logger.exception("consume tick failed")
await asyncio.sleep(1)
async def main() -> None:
database_url = read_secret("DATABASE_URL_FILE", "DATABASE_URL")
cache_url = read_secret("CACHE_URL_FILE", "CACHE_URL")
await db.init_pool(database_url)
broker.init(cache_url)
await broker.ensure_group()
storage.init(
endpoint=os.environ["S3_ENDPOINT"],
region=os.environ.get("S3_REGION", "us-east-1"),
access_key=read_secret("S3_ACCESS_KEY_ID_FILE", "S3_ACCESS_KEY_ID"),
secret_key=read_secret("S3_SECRET_ACCESS_KEY_FILE", "S3_SECRET_ACCESS_KEY"),
bucket=os.environ.get("S3_BUCKET", "messenger-attachments"),
)
logger.info("worker %s starting", CONSUMER_NAME)
await asyncio.gather(consume_loop(), reclaim_loop(), retry_scheduler_loop())
if __name__ == "__main__":
asyncio.run(main())
+173
View File
@@ -0,0 +1,173 @@
name: mtusi2026lab5
services:
lb:
image: nginx:1.27.2
ports:
- "8080:8080"
volumes:
- ./infra/nginx/nginx.conf:/etc/nginx/nginx.conf:ro
depends_on:
api1:
condition: service_healthy
api2:
condition: service_healthy
deploy:
resources:
limits:
cpus: "0.5"
memory: 128M
api1: &api
build: .
image: mtusi2026lab5-app:local
environment: &api-env
APP_HOST: 0.0.0.0
APP_PORT: 8080
APP_ENV: lab
SESSION_TTL_SECONDS: ${SESSION_TTL_SECONDS:-3600}
DATABASE_URL_FILE: /run/secrets/database_url
CACHE_URL_FILE: /run/secrets/cache_url
S3_ENDPOINT: http://seaweedfs:8333
S3_REGION: us-east-1
S3_BUCKET: messenger-attachments
S3_ACCESS_KEY_ID_FILE: /run/secrets/s3_access_key_id
S3_SECRET_ACCESS_KEY_FILE: /run/secrets/s3_secret_access_key
MAX_UPLOAD_BYTES: ${MAX_UPLOAD_BYTES:-10485760}
MAX_IMAGE_PIXELS: ${MAX_IMAGE_PIXELS:-40000000}
INSTANCE_ID: api1
secrets: &api-secrets
- database_url
- cache_url
- s3_access_key_id
- s3_secret_access_key
depends_on:
postgres:
condition: service_healthy
redis:
condition: service_healthy
seaweedfs:
condition: service_healthy
healthcheck: &api-healthcheck
test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8080/health/ready', timeout=2)"]
interval: 5s
timeout: 3s
retries: 5
start_period: 10s
deploy:
resources:
limits:
cpus: "1.0"
memory: 256M
api2:
<<: *api
environment:
<<: *api-env
INSTANCE_ID: api2
worker:
build: .
image: mtusi2026lab5-app:local
command: ["python3", "-m", "app.worker"]
environment:
DATABASE_URL_FILE: /run/secrets/database_url
CACHE_URL_FILE: /run/secrets/cache_url
S3_ENDPOINT: http://seaweedfs:8333
S3_REGION: us-east-1
S3_BUCKET: messenger-attachments
S3_ACCESS_KEY_ID_FILE: /run/secrets/s3_access_key_id
S3_SECRET_ACCESS_KEY_FILE: /run/secrets/s3_secret_access_key
JOB_MAX_ATTEMPTS: ${JOB_MAX_ATTEMPTS:-5}
JOB_TIMEOUT_SECONDS: ${JOB_TIMEOUT_SECONDS:-60}
INSTANCE_ID: worker1
secrets: *api-secrets
depends_on:
postgres:
condition: service_healthy
redis:
condition: service_healthy
seaweedfs:
condition: service_healthy
deploy:
resources:
limits:
cpus: "1.0"
memory: 384M
postgres:
image: postgres:16.4
environment:
POSTGRES_DB: messenger
POSTGRES_USER: messenger
POSTGRES_PASSWORD_FILE: /run/secrets/postgres_password
secrets:
- postgres_password
volumes:
- pgdata:/var/lib/postgresql/data
healthcheck:
test: ["CMD-SHELL", "pg_isready -U messenger -d messenger"]
interval: 5s
timeout: 3s
retries: 10
deploy:
resources:
limits:
cpus: "1.0"
memory: 512M
redis:
image: redis:7.4.1
command: ["redis-server", "--appendonly", "yes", "--appendfsync", "everysec"]
volumes:
- redisdata:/data
healthcheck:
test: ["CMD", "redis-cli", "ping"]
interval: 5s
timeout: 3s
retries: 10
deploy:
resources:
limits:
cpus: "0.5"
memory: 256M
seaweedfs:
image: chrislusf/seaweedfs:3.73
command:
- server
- -dir=/data
- -s3
- -s3.port=8333
- -s3.config=/etc/seaweedfs/s3.json
volumes:
- seaweed_data:/data
- ./secrets/s3.json:/etc/seaweedfs/s3.json:ro
healthcheck:
test: ["CMD", "nc", "-z", "127.0.0.1", "8333"]
interval: 5s
timeout: 3s
retries: 10
start_period: 10s
deploy:
resources:
limits:
cpus: "1.0"
memory: 512M
volumes:
pgdata:
redisdata:
seaweed_data:
secrets:
database_url:
file: ./secrets/database_url.txt
cache_url:
file: ./secrets/cache_url.txt
postgres_password:
file: ./secrets/postgres_password.txt
s3_access_key_id:
file: ./secrets/s3_access_key_id.txt
s3_secret_access_key:
file: ./secrets/s3_secret_access_key.txt
+56
View File
@@ -0,0 +1,56 @@
attachment=5a9a4ec0-0586-44a6-bada-76d974507038; forcing attempt=JOB_MAX_ATTEMPTS (5) to go straight to failed/DLQ
attachment after exhausted attempts: {'status': 'failed', 'error_code': 'processing_failed'}
job_failures (DLQ) row: {'attempt': 5, 'error_code': 'processing_failed'}
--- content on failed attachment ---
{"error":{"code":"conflict","message":"attachment processing failed","request_id":"3cc35203fcd943fd952d443564f97794"}}
http_code: 409
--- explicit operator replay ---
python3: can't open file '/app/scripts/replay_failed.py': [Errno 2] No such file or directory
Container mtusi2026lab5-postgres-1 Running
Container mtusi2026lab5-redis-1 Running
Container mtusi2026lab5-seaweedfs-1 Running
Container mtusi2026lab5-redis-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Waiting
Container mtusi2026lab5-postgres-1 Waiting
Container mtusi2026lab5-postgres-1 Healthy
Container mtusi2026lab5-seaweedfs-1 Healthy
Container mtusi2026lab5-redis-1 Healthy
Container mtusi2026lab5-worker-run-b07e657c4d6b Creating
Container mtusi2026lab5-worker-run-b07e657c4d6b Created
python3: can't open file '/app/scripts/replay_failed.py': [Errno 2] No such file or directory
--- status after replay ---
{"id":"5a9a4ec0-0586-44a6-bada-76d974507038","chat_id":"e0f59339-338a-42b0-b20f-cbc743bde678","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"dlq.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"failed","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":5813,"width":256,"height":160,"sha256":"b8e59b0cad429b1c49c27c4a1c79e52bfc23e2fab7c7238d18d5addd4df4b546"},{"name":"crop","content_type":"image/jpeg","size_bytes":2504,"width":128,"height":128,"sha256":"634c9c489b3bac5cb990e37300ca686174a47e14e8cfda2c9838eeeefd7252d6"},{"name":"optimized","content_type":"image/webp","size_bytes":2462,"width":320,"height":200,"sha256":"1090d84ed8411226010ae498b08e1b9e6d81c8eee02ee77143a0dc327b894beb"}],"error_code":"processing_failed","created_at":"2026-10-29T22:17:00.999540Z","updated_at":"2026-10-29T22:17:03.607703Z"}
Container mtusi2026lab5-seaweedfs-1 Running
Container mtusi2026lab5-postgres-1 Running
Container mtusi2026lab5-redis-1 Running
Container mtusi2026lab5-postgres-1 Waiting
Container mtusi2026lab5-redis-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Waiting
Container mtusi2026lab5-redis-1 Healthy
Container mtusi2026lab5-seaweedfs-1 Healthy
Container mtusi2026lab5-postgres-1 Healthy
Container mtusi2026lab5-worker-run-ccd421ee163e Creating
Container mtusi2026lab5-worker-run-ccd421ee163e Created
Traceback (most recent call last):
File "/scripts/replay_failed.py", line 18, in <module>
from app import broker # noqa: E402
^^^^^^^^^^^^^^^^^^^^^^
ImportError: cannot import name 'broker' from 'app' (unknown location)
--- status after replay ---
{"id":"5a9a4ec0-0586-44a6-bada-76d974507038","chat_id":"e0f59339-338a-42b0-b20f-cbc743bde678","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"dlq.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"failed","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":5813,"width":256,"height":160,"sha256":"b8e59b0cad429b1c49c27c4a1c79e52bfc23e2fab7c7238d18d5addd4df4b546"},{"name":"crop","content_type":"image/jpeg","size_bytes":2504,"width":128,"height":128,"sha256":"634c9c489b3bac5cb990e37300ca686174a47e14e8cfda2c9838eeeefd7252d6"},{"name":"optimized","content_type":"image/webp","size_bytes":2462,"width":320,"height":200,"sha256":"1090d84ed8411226010ae498b08e1b9e6d81c8eee02ee77143a0dc327b894beb"}],"error_code":"processing_failed","created_at":"2026-10-29T22:17:00.999540Z","updated_at":"2026-10-29T22:17:03.607703Z"}
Container mtusi2026lab5-seaweedfs-1 Running
Container mtusi2026lab5-postgres-1 Running
Container mtusi2026lab5-redis-1 Running
Container mtusi2026lab5-postgres-1 Waiting
Container mtusi2026lab5-redis-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Waiting
Container mtusi2026lab5-postgres-1 Healthy
Container mtusi2026lab5-redis-1 Healthy
Container mtusi2026lab5-seaweedfs-1 Healthy
Container mtusi2026lab5-worker-run-de066dce3f98 Creating
Container mtusi2026lab5-worker-run-de066dce3f98 Created
replayed attachment 5a9a4ec0-0586-44a6-bada-76d974507038 as job 1cd0e532-2cec-49e3-924d-07f532749946
--- status after replay ---
{"id":"5a9a4ec0-0586-44a6-bada-76d974507038","chat_id":"e0f59339-338a-42b0-b20f-cbc743bde678","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"dlq.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"ready","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":5813,"width":256,"height":160,"sha256":"b8e59b0cad429b1c49c27c4a1c79e52bfc23e2fab7c7238d18d5addd4df4b546"},{"name":"crop","content_type":"image/jpeg","size_bytes":2504,"width":128,"height":128,"sha256":"634c9c489b3bac5cb990e37300ca686174a47e14e8cfda2c9838eeeefd7252d6"},{"name":"optimized","content_type":"image/webp","size_bytes":2462,"width":320,"height":200,"sha256":"1090d84ed8411226010ae498b08e1b9e6d81c8eee02ee77143a0dc327b894beb"}],"error_code":null,"created_at":"2026-10-29T22:17:00.999540Z","updated_at":"2026-10-29T22:17:44.642487Z"}
+42
View File
@@ -0,0 +1,42 @@
=== outbox gap: insert outbox row without publishing, confirm relay publishes it ===
Container mtusi2026lab5-postgres-1 Running
Container mtusi2026lab5-redis-1 Running
Container mtusi2026lab5-seaweedfs-1 Running
Container mtusi2026lab5-postgres-1 Waiting
Container mtusi2026lab5-redis-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Healthy
Container mtusi2026lab5-postgres-1 Healthy
Container mtusi2026lab5-redis-1 Healthy
Container mtusi2026lab5-api1-run-6d13714828b2 Creating
Container mtusi2026lab5-api1-run-6d13714828b2 Created
Traceback (most recent call last):
File "<string>", line 21, in <module>
File "/usr/local/lib/python3.12/asyncio/runners.py", line 194, in run
return runner.run(main)
^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/asyncio/runners.py", line 118, in run
return self._loop.run_until_complete(task)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/asyncio/base_events.py", line 687, in run_until_complete
return future.result()
^^^^^^^^^^^^^^^
File "<string>", line 9, in main
FileNotFoundError: [Errno 2] No such file or directory: 'tests/fixtures/sample.png'
chat=5295f6da-2daa-4be9-bc29-93ca17000d7b
Container mtusi2026lab5-redis-1 Running
Container mtusi2026lab5-seaweedfs-1 Running
Container mtusi2026lab5-postgres-1 Running
Container mtusi2026lab5-postgres-1 Waiting
Container mtusi2026lab5-redis-1 Waiting
Container mtusi2026lab5-seaweedfs-1 Waiting
Container mtusi2026lab5-redis-1 Healthy
Container mtusi2026lab5-postgres-1 Healthy
Container mtusi2026lab5-seaweedfs-1 Healthy
Container mtusi2026lab5-api1-run-e0ced838f8e2 Creating
Container mtusi2026lab5-api1-run-e0ced838f8e2 Created
attachment_id: 60d002c1-f03b-4b2e-b1e0-d0ca1cb258e1
--- after relay window (API's background relay_loop ticks every 2s) ---
{"id":"60d002c1-f03b-4b2e-b1e0-d0ca1cb258e1","chat_id":"5295f6da-2daa-4be9-bc29-93ca17000d7b","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"outbox-gap.png","content_type":"image/png","size_bytes":155,"sha256":"4a915d9ec56a94871ded760acfc6175103895fbad0b0b4edbd2e3bc8ecedb399","status":"ready","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":693,"width":64,"height":64,"sha256":"6c25cc6e23df5235488bcb0cf285ed9f6e60fcdcc5e169cd47a317a0d9464548"},{"name":"crop","content_type":"image/jpeg","size_bytes":885,"width":128,"height":128,"sha256":"49f3ccc99299919178a3f9c96e2f8722b93e14528eef439219b71380686288dc"},{"name":"optimized","content_type":"image/webp","size_bytes":90,"width":64,"height":64,"sha256":"c085686a2b9ba6244316d14af6c4b18978d24e18b18c94e22929372dd9e29b96"}],"error_code":null,"created_at":"2026-10-29T22:18:26.167466Z","updated_at":"2026-10-29T22:18:28.033253Z"}
duplicate delivery handled (see worker log: should say already ready; no-op)
+8
View File
@@ -0,0 +1,8 @@
attachment=94dfbf3f-6cfe-40b5-90e5-5024f1a2058e (already processed to ready normally); now simulating a transient failure on it
scheduled retry: {'attempt': 2, 'next_attempt_at': datetime.datetime(2026, 10, 29, 22, 16, 37, 151859, tzinfo=datetime.timezone.utc)}
attachment status after failure: queued
--- after retry window ---
{"id":"94dfbf3f-6cfe-40b5-90e5-5024f1a2058e","chat_id":"eb779693-acc7-4102-9104-6db7d30ca5b7","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"r6.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"ready","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":5813,"width":256,"height":160,"sha256":"b8e59b0cad429b1c49c27c4a1c79e52bfc23e2fab7c7238d18d5addd4df4b546"},{"name":"crop","content_type":"image/jpeg","size_bytes":2504,"width":128,"height":128,"sha256":"634c9c489b3bac5cb990e37300ca686174a47e14e8cfda2c9838eeeefd7252d6"},{"name":"optimized","content_type":"image/webp","size_bytes":2462,"width":320,"height":200,"sha256":"1090d84ed8411226010ae498b08e1b9e6d81c8eee02ee77143a0dc327b894beb"}],"error_code":null,"created_at":"2026-10-29T22:16:32.596966Z","updated_at":"2026-10-29T22:16:37.609583Z"}
worker-1 | 2026-10-29 22:16:15,344 INFO worker worker1 starting
worker-1 | 2026-10-29 22:16:32,751 INFO attachment 94dfbf3f-6cfe-40b5-90e5-5024f1a2058e ready (attempt 1)
worker-1 | 2026-10-29 22:16:37,627 INFO attachment 94dfbf3f-6cfe-40b5-90e5-5024f1a2058e ready (attempt 2)
+10
View File
@@ -0,0 +1,10 @@
=== worker stopped; token=eUcTv9xFDTJBWxuGqC0yAgDbVxsvczX5_YyH37X10ZQ chat=80ae8843-3d01-4bf9-81e5-3696e12ee5f4 ===
upload response (worker down): {"id":"68ff6c2e-2876-4d28-9627-c264b4266bdf","chat_id":"80ae8843-3d01-4bf9-81e5-3696e12ee5f4","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"w.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"queued","variants":[],"error_code":null,"created_at":"2026-10-29T22:09:58.309575Z","updated_at":"2026-10-29T22:09:58.309575Z"}
attachment id: 68ff6c2e-2876-4d28-9627-c264b4266bdf
{"id":"68ff6c2e-2876-4d28-9627-c264b4266bdf","chat_id":"80ae8843-3d01-4bf9-81e5-3696e12ee5f4","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"w.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"queued","variants":[],"error_code":null,"created_at":"2026-10-29T22:09:58.309575Z","updated_at":"2026-10-29T22:09:58.309575Z"}
=== now: restart API (worker still down) ===
{"id":"68ff6c2e-2876-4d28-9627-c264b4266bdf","chat_id":"80ae8843-3d01-4bf9-81e5-3696e12ee5f4","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"w.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"queued","variants":[],"error_code":null,"created_at":"2026-10-29T22:09:58.309575Z","updated_at":"2026-10-29T22:09:58.309575Z"}=== content while queued ===
{"error":{"code":"conflict","message":"attachment is not ready yet","request_id":"e1a1ba173d924ca6a399a8fec74f47dd"}}
http_code: 409
=== starting worker ===
{"id":"68ff6c2e-2876-4d28-9627-c264b4266bdf","chat_id":"80ae8843-3d01-4bf9-81e5-3696e12ee5f4","owner_id":"94efa629-eea4-42e4-a876-df57d3b60683","filename":"w.png","content_type":"image/png","size_bytes":111817,"sha256":"99ea7533c441f79cd630d15ce4d58d1ec4e06f6f0efb3b77a5c088305cd1bd75","status":"ready","variants":[{"name":"thumbnail","content_type":"image/jpeg","size_bytes":5813,"width":256,"height":160,"sha256":"b8e59b0cad429b1c49c27c4a1c79e52bfc23e2fab7c7238d18d5addd4df4b546"},{"name":"crop","content_type":"image/jpeg","size_bytes":2504,"width":128,"height":128,"sha256":"634c9c489b3bac5cb990e37300ca686174a47e14e8cfda2c9838eeeefd7252d6"},{"name":"optimized","content_type":"image/webp","size_bytes":2462,"width":320,"height":200,"sha256":"1090d84ed8411226010ae498b08e1b9e6d81c8eee02ee77143a0dc327b894beb"}],"error_code":null,"created_at":"2026-10-29T22:09:58.309575Z","updated_at":"2026-10-29T22:10:16.813333Z"}
+32
View File
@@ -0,0 +1,32 @@
worker_processes 1;
# A single worker keeps the round-robin counter shared and deterministic for
# the lab's small replica count; a production LB would instead rely on a
# shared/consistent balancing policy across workers.
events { worker_connections 1024; }
http {
upstream messenger_api {
# Passive health check: a backend failing 1 probe is removed for 3s,
# which keeps failover detection well under the 10s budget.
server api1:8080 max_fails=1 fail_timeout=3s;
server api2:8080 max_fails=1 fail_timeout=3s;
}
server {
listen 8080;
# Slightly above the app's own 10 MiB attachment limit so the API --
# not Nginx's default HTML error page -- produces the contract's
# JSON 413 envelope. The API still enforces the exact byte budget.
client_max_body_size 11M;
location / {
proxy_pass http://messenger_api;
proxy_connect_timeout 1s;
proxy_read_timeout 5s;
proxy_next_upstream error timeout http_502 http_503 http_504;
proxy_next_upstream_tries 2;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
}
}
}
+18
View File
@@ -0,0 +1,18 @@
{
"identities": [
{
"name": "messenger-api",
"credentials": [
{
"accessKey": "__S3_ACCESS_KEY_ID__",
"secretKey": "__S3_SECRET_ACCESS_KEY__"
}
],
"actions": [
"Read:messenger-attachments",
"Write:messenger-attachments",
"List:messenger-attachments"
]
}
]
}
+33
View File
@@ -0,0 +1,33 @@
-- Initial schema: users, chats, membership, messages.
CREATE TABLE IF NOT EXISTS users (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
username TEXT NOT NULL UNIQUE,
display_name TEXT NOT NULL,
password_hash TEXT NOT NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT now()
);
CREATE TABLE IF NOT EXISTS chats (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
title TEXT NOT NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT now()
);
CREATE TABLE IF NOT EXISTS chat_members (
chat_id UUID NOT NULL REFERENCES chats(id) ON DELETE CASCADE,
user_id UUID NOT NULL REFERENCES users(id) ON DELETE CASCADE,
PRIMARY KEY (chat_id, user_id)
);
CREATE INDEX IF NOT EXISTS chat_members_user_idx ON chat_members (user_id);
CREATE TABLE IF NOT EXISTS messages (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
chat_id UUID NOT NULL REFERENCES chats(id) ON DELETE CASCADE,
sender_id UUID NOT NULL REFERENCES users(id),
text TEXT NOT NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT now()
);
-- Keyset pagination over (chat_id, created_at, id) avoids a full scan per page.
CREATE INDEX IF NOT EXISTS messages_chat_keyset_idx ON messages (chat_id, created_at, id);
+15
View File
@@ -0,0 +1,15 @@
-- Idempotency store for POST message: scope = (user, method, path, key).
CREATE TABLE IF NOT EXISTS idempotency_keys (
user_id UUID NOT NULL,
method TEXT NOT NULL,
path TEXT NOT NULL,
idem_key TEXT NOT NULL,
request_hash TEXT NOT NULL,
status TEXT NOT NULL DEFAULT 'pending',
response_status INT,
response_body JSONB,
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
PRIMARY KEY (user_id, method, path, idem_key)
);
CREATE INDEX IF NOT EXISTS idempotency_keys_created_idx ON idempotency_keys (created_at);
+25
View File
@@ -0,0 +1,25 @@
-- Attachment metadata lives in Postgres; bytes live in private S3. s3_key is
-- server-generated and never derived from the user-supplied filename.
CREATE TABLE IF NOT EXISTS attachments (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
chat_id UUID NOT NULL REFERENCES chats(id) ON DELETE CASCADE,
owner_id UUID NOT NULL REFERENCES users(id),
filename TEXT NOT NULL,
content_type TEXT NOT NULL,
size_bytes BIGINT NOT NULL,
sha256 TEXT NOT NULL,
status TEXT NOT NULL DEFAULT 'ready',
error_code TEXT,
s3_key TEXT NOT NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
);
CREATE INDEX IF NOT EXISTS attachments_chat_idx ON attachments (chat_id);
-- Many-to-many: a message can reference up to 10 attachments of its own chat.
CREATE TABLE IF NOT EXISTS message_attachments (
message_id UUID NOT NULL REFERENCES messages(id) ON DELETE CASCADE,
attachment_id UUID NOT NULL REFERENCES attachments(id),
PRIMARY KEY (message_id, attachment_id)
);
+53
View File
@@ -0,0 +1,53 @@
-- Variant outputs (thumbnail/crop/optimized) produced by the worker.
CREATE TABLE IF NOT EXISTS attachment_variants (
attachment_id UUID NOT NULL REFERENCES attachments(id) ON DELETE CASCADE,
name TEXT NOT NULL,
content_type TEXT NOT NULL,
size_bytes BIGINT NOT NULL,
width INT NOT NULL,
height INT NOT NULL,
sha256 TEXT NOT NULL,
s3_key TEXT NOT NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
PRIMARY KEY (attachment_id, name)
);
-- Transactional outbox: written in the SAME transaction as the attachments
-- INSERT, so a crash between COMMIT and broker publish leaves a recoverable
-- row instead of a lost job. A relay (in-process background task) publishes
-- unpublished rows and marks them, so publish is retried, not re-created.
CREATE TABLE IF NOT EXISTS job_outbox (
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
attachment_id UUID NOT NULL REFERENCES attachments(id) ON DELETE CASCADE,
job_id UUID NOT NULL,
pipeline_version INT NOT NULL DEFAULT 1,
attempt INT NOT NULL DEFAULT 1,
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
published_at TIMESTAMPTZ
);
CREATE INDEX IF NOT EXISTS job_outbox_unpublished_idx ON job_outbox (created_at) WHERE published_at IS NULL;
-- Durable delayed-retry schedule (Redis Streams has no native delay): a
-- failed attempt is parked here with a next_attempt_at instead of sleeping
-- in-process, so a worker crash during the backoff window does not lose the
-- retry -- any worker's scheduler tick will still publish it on time.
CREATE TABLE IF NOT EXISTS job_retry_schedule (
attachment_id UUID NOT NULL REFERENCES attachments(id) ON DELETE CASCADE,
job_id UUID NOT NULL,
attempt INT NOT NULL,
next_attempt_at TIMESTAMPTZ NOT NULL,
PRIMARY KEY (attachment_id, job_id)
);
CREATE INDEX IF NOT EXISTS job_retry_schedule_due_idx ON job_retry_schedule (next_attempt_at);
-- Explicit failure record for DLQ-style operator visibility and replay.
CREATE TABLE IF NOT EXISTS job_failures (
attachment_id UUID NOT NULL REFERENCES attachments(id) ON DELETE CASCADE,
job_id UUID NOT NULL,
attempt INT NOT NULL,
error_code TEXT NOT NULL,
failed_at TIMESTAMPTZ NOT NULL DEFAULT now(),
PRIMARY KEY (attachment_id, job_id)
);
+8
View File
@@ -0,0 +1,8 @@
fastapi==0.115.6
uvicorn[standard]==0.32.1
asyncpg==0.30.0
redis==5.2.1
argon2-cffi==23.1.0
pydantic==2.10.3
boto3==1.35.74
Pillow==11.0.0
+72
View File
@@ -0,0 +1,72 @@
#!/usr/bin/env python3
"""Find S3 objects with no matching `attachments` row (e.g. PUT succeeded but
the PostgreSQL INSERT that followed it never committed -- a crash between the
two writes). Dry-run by default; --delete removes orphans older than the
grace period, never anything referenced in the database.
Usage:
python3 scripts/reconcile_attachments.py --grace-minutes 60
python3 scripts/reconcile_attachments.py --grace-minutes 60 --delete
"""
import argparse
import asyncio
import datetime as dt
import os
import sys
import asyncpg
import boto3
def read_secret(file_env, inline_env):
path = os.environ.get(file_env)
if path and os.path.isfile(path):
return open(path, encoding="utf-8").read().strip()
return os.environ[inline_env]
async def main():
p = argparse.ArgumentParser()
p.add_argument("--grace-minutes", type=int, default=60,
help="Only objects older than this are candidates (protects in-flight uploads).")
p.add_argument("--delete", action="store_true", help="Actually delete orphans (default: dry-run).")
args = p.parse_args()
database_url = read_secret("DATABASE_URL_FILE", "DATABASE_URL")
conn = await asyncpg.connect(database_url)
known_keys = {r["s3_key"] for r in await conn.fetch("SELECT s3_key FROM attachments")}
await conn.close()
s3 = boto3.client(
"s3",
endpoint_url=os.environ["S3_ENDPOINT"],
region_name=os.environ.get("S3_REGION", "us-east-1"),
aws_access_key_id=read_secret("S3_ACCESS_KEY_ID_FILE", "S3_ACCESS_KEY_ID"),
aws_secret_access_key=read_secret("S3_SECRET_ACCESS_KEY_FILE", "S3_SECRET_ACCESS_KEY"),
)
bucket = os.environ.get("S3_BUCKET", "messenger-attachments")
cutoff = dt.datetime.now(dt.timezone.utc) - dt.timedelta(minutes=args.grace_minutes)
orphans = []
paginator = s3.get_paginator("list_objects_v2")
for page in paginator.paginate(Bucket=bucket, Prefix="attachments/"):
for obj in page.get("Contents", []):
if obj["Key"] in known_keys:
continue
if obj["LastModified"] > cutoff:
continue # might still be an in-flight upload; not an orphan yet
orphans.append(obj["Key"])
print(f"{len(orphans)} orphaned object(s) older than {args.grace_minutes} minutes with no DB row.")
for key in orphans:
print(" ", key)
if args.delete:
s3.delete_object(Bucket=bucket, Key=key)
if orphans and args.delete:
print(f"Deleted {len(orphans)} orphan(s).")
elif orphans:
print("Dry-run: nothing deleted. Re-run with --delete to remove them.")
if __name__ == "__main__":
sys.exit(asyncio.run(main()) or 0)
+66
View File
@@ -0,0 +1,66 @@
#!/usr/bin/env python3
"""Explicit operator replay for a failed attachment: resets status to
'queued' and republishes a fresh job (new job_id, attempt=1). Never replays
an attachment that is already 'ready' -- the same protection the worker
itself applies against stale duplicates.
Usage: python3 scripts/replay_failed.py <attachment-id>
"""
import asyncio
import datetime as dt
import os
import sys
import uuid
import asyncpg
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from app import broker # noqa: E402
def read_secret(file_env, inline_env):
path = os.environ.get(file_env)
if path and os.path.isfile(path):
return open(path, encoding="utf-8").read().strip()
return os.environ[inline_env]
async def main():
if len(sys.argv) != 2:
print("usage: replay_failed.py <attachment-id>", file=sys.stderr)
return 2
attachment_id = uuid.UUID(sys.argv[1])
conn = await asyncpg.connect(read_secret("DATABASE_URL_FILE", "DATABASE_URL"))
record = await conn.fetchrow("SELECT status FROM attachments WHERE id=$1", attachment_id)
if record is None:
print("no such attachment", file=sys.stderr)
return 1
if record["status"] == "ready":
print("attachment is already ready; refusing to replay a finished job")
return 0
if record["status"] != "failed":
print(f"attachment status is '{record['status']}', not 'failed'; nothing to replay")
return 0
job_id = uuid.uuid4()
await conn.execute("UPDATE attachments SET status='queued', error_code=NULL, updated_at=now() WHERE id=$1", attachment_id)
await conn.close()
broker.init(read_secret("CACHE_URL_FILE", "CACHE_URL"))
job = {
"schema_version": 1,
"job_id": str(job_id),
"type": "attachment.process.v1",
"attachment_id": str(attachment_id),
"pipeline_version": 1,
"attempt": 1,
"created_at": dt.datetime.now(dt.timezone.utc).isoformat().replace("+00:00", "Z"),
}
await broker.publish(job)
print(f"replayed attachment {attachment_id} as job {job_id}")
return 0
if __name__ == "__main__":
sys.exit(asyncio.run(main()))
+26
View File
@@ -0,0 +1,26 @@
#!/bin/sh
set -eu
cd "$(dirname "$0")/.."
wait_healthy() {
for svc in "$@"; do
for _ in $(seq 1 60); do
status=$(docker compose ps --format '{{.Health}}' "$svc" 2>/dev/null || true)
[ "$status" = "healthy" ] && break
sleep 1
done
done
}
# Sequenced: API/worker eagerly connect to Postgres/Redis/S3 at startup and
# exit if unreachable; Nginx resolves upstream hostnames only at its own
# startup. Each stage must be actually healthy before the next one starts.
docker compose restart postgres redis seaweedfs
wait_healthy postgres redis seaweedfs
docker compose restart api1 api2
wait_healthy api1 api2
docker compose restart worker
docker compose restart lb
exit 0
+12
View File
@@ -0,0 +1,12 @@
#!/bin/sh
set -eu
# Stops exactly the API replica identified by TARGET_INSTANCE_ID (api1 or api2),
# leaving it stopped. The other replica keeps serving traffic through the LB.
cd "$(dirname "$0")/.."
: "${TARGET_INSTANCE_ID:?TARGET_INSTANCE_ID must be set by the test harness}"
case "$TARGET_INSTANCE_ID" in
api1|api2) ;;
*) echo "unknown instance id: $TARGET_INSTANCE_ID" >&2; exit 1 ;;
esac
docker compose stop "$TARGET_INSTANCE_ID"
exit 0