Jobs: durable task runner with fence + retry + force-takeover

Postgres-backed task framework via createTaskRunner. Per-attempt fence in Postgres + an optional Redis mirror via createRedisFence; an idempotency store (createIdempotencyStore) caches results so a retry with the same key returns the cached output instead of duplicating work. live.cron runs a 1Hz refresh tick that polls the table and publishes both the row list and a stats snapshot.

Pending

Running

Committed

Failed

Total

Recent tasks (0)

No tasks yet. Enqueue one above.

The page subscribes to demos:jobs:list (the row list, capped at 30) and demos:jobs:stats (status counts). Both publish via merge: 'set' - every event replaces the value, since the canonical state lives in Postgres and the demo doesn't try to track per-row crud events from inside the runner's state machine.

Force takeover bumps fence_expires_at back into the past; the runner's recovery sweep (recoveryInterval: 2000) reclaims the row and rearms with a new fence within ~2s. The original handler's heartbeat (heartbeatInterval: 1500) sees the fence drift and aborts via ctx.signal. With the Redis fence enabled, abort fires on the next heartbeat tick instead of waiting for Postgres time math.

Retry policy: handler exceptions trigger up to 3 attempts (linear 250ms backoff). The fail-once mode throws on attempt 1 and commits on attempt 2; fail-always exhausts retries and lands the row at failed.

Enqueue simulate-work

Recent tasks (0)