forked from mtusi2026/mtusi2026lab5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> broker: RabbitMQ 4.3.7 (тег+digest), quorum queue + DLX/DLQ из definitions, пользователи relay (только publish в exchange) и worker (только consume), без management UI и портов. relay: outbox -> publish с publisher confirm, reaper истёкших lease, страховка от потерянных сообщений. worker: claim + heartbeat lease, варианты, публикация с fencing, ack после COMMIT, graceful stop, уборка staging и мусора попыток. S3: два bucket и четыре identity (admin, api, worker, auditor) с проверкой модели доступа в s3-init. Отдельная сеть queue. Reconciler lab4 (трек прошлой лабы) удалён: его роль выполняет уборка по журналу попыток.
99 lines
4.1 KiB
Bash
99 lines
4.1 KiB
Bash
# Copy to .env for your LOCAL synthetic test environment; never commit .env.
|
|
# These are logical interface suggestions; map actual framework names in REPORT.md.
|
|
APP_HOST=0.0.0.0
|
|
APP_PORT=8080
|
|
APP_ENV=lab
|
|
SESSION_TTL_SECONDS=3600
|
|
# Choose credentials locally. File paths contain no secret values.
|
|
# The files are created by `python scripts/init_secrets.py` and mounted by compose as secrets.
|
|
DATABASE_URL_FILE=/run/secrets/database_url
|
|
# Optional on 3, required use on 4 in lab2; Redis OR Valkey:
|
|
# valkey = sessions live only in Valkey (AOF, appendfsync always); postgres = table `sessions`.
|
|
SESSION_STORE=valkey
|
|
CACHE_URL_FILE=/run/secrets/cache_url
|
|
# Number of interchangeable API replicas behind HAProxy (compose deploy.replicas).
|
|
API_REPLICAS=2
|
|
# Idempotency-Key results are kept at least this long (contract: >= 24 h).
|
|
IDEMPOTENCY_TTL_SECONDS=86400
|
|
# Write path of POST message: function = one SQL call (migration 0003), multi = lab3 baseline path.
|
|
MESSAGE_WRITE_PATH=function
|
|
# Lab 4: private S3 bucket (SeaweedFS service `s3`). The API gets ITS OWN key pair (Read+Write on
|
|
# this bucket only) as Docker secrets; admin/reconciler pairs go only to s3-init/reconciler.
|
|
S3_ENDPOINT=http://s3:8333
|
|
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
|
|
# Lab 5: private staging bucket for originals accepted before 202 (API: Write; worker: Read+Write).
|
|
S3_STAGING_BUCKET=messenger-staging
|
|
# Contract limits (may be lowered, never raised above 10485760 / 40000000).
|
|
MAX_UPLOAD_BYTES=10485760
|
|
MAX_IMAGE_PIXELS=40000000
|
|
# Per replica: bodies buffered at once (RAM bound), wait for a slot, body deadline, image checks at once.
|
|
UPLOAD_CONCURRENCY=4
|
|
UPLOAD_QUEUE_TIMEOUT_SECONDS=5
|
|
UPLOAD_READ_TIMEOUT_SECONDS=30
|
|
IMAGE_VERIFY_TIMEOUT_SECONDS=10
|
|
IMAGE_VERIFY_CONCURRENCY=1
|
|
JPEG_MAX_MEMORY_MB=256
|
|
S3_TIMEOUT_SECONDS=3
|
|
# Janitor: 'pending' intents older than this are failed uploads (object + row removed).
|
|
PENDING_GRACE_SECONDS=900
|
|
JANITOR_INTERVAL_SECONDS=60
|
|
# Lab 5: broker credentials are files too (relay and worker get their OWN RabbitMQ user/URL;
|
|
# secrets/broker_url_relay|worker are mounted as /run/secrets/broker_url). The API has none.
|
|
BROKER_URL_FILE=/run/secrets/broker_url
|
|
# Jobs processed at once by one worker = RabbitMQ prefetch; image child processes at once.
|
|
WORKER_CONCURRENCY=2
|
|
WORKER_REPLICAS=1
|
|
# Managed attempts (claims) per job, then failed/retry_exhausted; backoff = full jitter.
|
|
JOB_MAX_ATTEMPTS=5
|
|
JOB_RETRY_BASE_SECONDS=2
|
|
JOB_RETRY_MAX_SECONDS=60
|
|
# One attempt: hard time limit (child process killed) and the lease renewed by its heartbeat.
|
|
JOB_TIMEOUT_SECONDS=60
|
|
JOB_LEASE_SECONDS=30
|
|
# Address-space limit of the image child process (RLIMIT_AS).
|
|
IMAGE_MAX_MEMORY_MB=768
|
|
# Staging copy of a FAILED attachment is kept this long for a replay (ready: deleted at once).
|
|
STAGING_RETENTION_SECONDS=86400
|
|
# Objects of attempts that never published are deleted this long after the attempt ended.
|
|
ORPHAN_GRACE_SECONDS=600
|
|
# Queued job without a claim for this long gets a new outbox row (lost-message guard).
|
|
QUEUED_REPUBLISH_SECONDS=300
|
|
# Pipeline version of new uploads (track 2: versions 1, 2, 3 in src/messenger/pipeline.py).
|
|
PIPELINE_VERSION=1
|
|
# Failure experiments only, per service (see src/messenger/config.py FAULT_POINTS); keep empty.
|
|
API_FAULT_POINTS=
|
|
RELAY_FAULT_POINTS=
|
|
WORKER_FAULT_POINTS=
|
|
# Additional settings of this implementation (defaults shown; all optional):
|
|
LOG_LEVEL=info
|
|
DB_POOL_MIN=1
|
|
DB_POOL_MAX=10
|
|
DB_TIMEOUT_SECONDS=3
|
|
CACHE_TIMEOUT_SECONDS=0.5
|
|
PASSWORD_HASH_CONCURRENCY=2
|
|
DOCS_ENABLED=true
|
|
# Graceful stop of a replica: readiness 503 + drain, then up to SHUTDOWN for in-flight requests.
|
|
DRAIN_SECONDS=3
|
|
SHUTDOWN_TIMEOUT_SECONDS=8
|
|
# CPU/RAM limits (docker compose deploy.resources.limits); API_* are per replica.
|
|
API_CPUS=1.0
|
|
API_MEMORY=512m
|
|
S3_CPUS=1.0
|
|
S3_MEMORY=1g
|
|
PG_CPUS=2.0
|
|
PG_MEMORY=1g
|
|
VALKEY_CPUS=0.5
|
|
LB_CPUS=1.0
|
|
BROKER_CPUS=1.0
|
|
BROKER_MEMORY=512m
|
|
WORKER_CPUS=2.0
|
|
WORKER_MEMORY=1536m
|
|
RELAY_CPUS=0.5
|
|
PG_MAX_CONNECTIONS=100
|
|
# Host ports of the load balancer (published on 127.0.0.1 only; API replicas publish nothing)
|
|
APP_PUBLISH_PORT=8080
|
|
LB_STATS_PORT=8404
|