Retry and DLQ policy

Retry and DLQ policy (pricing)

BullMQ

• Main and measure queues use exponential backoff (see src/lib/pricing/queue-client.ts).
• Max attempts: 5 by default (PRICING_DEFAULT_MAX_ATTEMPTS in retry-policy.ts).
• After the final failure, jobs are forwarded to persisted Postgres dead_letter_jobs and optionally the BullMQ DLQ queue.

Idempotency

• Operator actions (POST /api/operator/actions) accept Idempotency-Key (or idempotency-key). The same key + actor + payload hash returns the stored prior response without re-executing the mutation (idempotency_records table, migration 008).
• DLQ replay uses deterministic BullMQ jobId patterns where possible to avoid duplicate side effects when operators double-click.

Classification

• retry-policy.ts lists message markers treated as likely non-retryable for tuning and observability (not automatically wired to BullMQ attempts per job type — BullMQ still retries until max attempts unless job options change).

Operations

• Monitor DLQ growth via /api/health/deep and control plane; set LEAD_OS_DLQ_ALERT_THRESHOLD to surface alerts in deep health when persisted DLQ rows exceed the threshold.