n8n_queue_sre
Ensure n8n queue mode reliability - idempotency, DLQ, outbox, worker scaling, timeout coherence.
SKILL.md.claude/skills/04_n8n_queue_sre/SKILL.md
GitHubn8n Queue SRE (RESTO BOT)
Current topology
- n8n-main (image: n8n:1.80.0): Webhook receiver, queue publisher
- Resources: 1 CPU, 1GB RAM
- Health:
wget http://127.0.0.1:5678/healthz
- n8n-worker: Queue consumer, execution engine
- Resources: 0.75 CPU, 768MB RAM
- Health:
pgrep -f 'n8n worker' - Concurrency:
QUEUE_BULL_MAX_CONCURRENCY(default: 2)
- Redis 7-alpine: Bull queue backend
- AOF persistence, 256MB max, allkeys-lru eviction
- Health:
redis-cli ping
Business logic env vars (n8n-main + worker)
- Multi-channel: WA_SEND_URL, IG_SEND_URL, MSG_SEND_URL + API tokens
- Outbox: OUTBOX_MAX_ATTEMPTS=7, BASE_DELAY=30s, MAX_DELAY=3600s
- Fraud: FRAUD_FLOOD_LIMIT_30S=6, HIGH_ORDER_THRESHOLD=3000000
- SLO: INBOUND_TO_OUTBOX_P95=2000ms, PENDING_AGE_MAX=600s, DLQ_RATE_MAX=0.05
Non-negotiables
- Webhooks respond quickly (ack fast, process async via queue)
- Idempotency gate near ingestion (dedupe key stored in DB)
- Retries with exponential backoff; cap retries at OUTBOX_MAX_ATTEMPTS
- Dead-letter path: error workflow + durable store
- Worker concurrency explicit and measured
- Redis durability: AOF enabled
Operational checks
- Monitor queue depth and failed executions
- Ensure worker restarts do not cause duplicate side effects
- Ensure timeouts are coherent: Traefik (30s) -> gateway (read_timeout) -> n8n (execution_timeout)
- Verify outbox pattern: pending -> processing -> sent/failed/DLQ
Key files
project/docker-compose.hostinger.prod.yml(n8n-main, n8n-worker, redis services)project/.env(QUEUE_*, OUTBOX_*, SLO_*, FRAUD_* vars)project/workflows/(54 workflow JSON files)project/scripts/n8n-worker-entrypoint.sh(worker startup)
Required output
- Queue reliability plan (env + settings)
- Incident playbook: worker crash loop, stuck queue, redis down
- Verification: simulate inbound event -> single DB write -> idempotent re-send