Release v1.6.0 · janbjorge/pgqueuer

v1.6.0

Run pgq upgrade before starting workers: this release adds a schema index.

Added

Partial index pgqueuer_picked_heartbeat_idx on (heartbeat) WHERE status = 'picked'. Dequeue's stale-job scan reads only timed-out rows instead of every picked job. Built with a plain CREATE INDEX, so writes block during the build.
enqueue(..., on_conflict="update_priority"): a duplicate dedupe_key raises the existing job's priority (never lowers it) and returns its id instead of raising DuplicateJobError.
pgqueuer.run() starts a worker from Python without the CLI.

Changed

PGQUEUER_* settings are read once per process and shared. DBSettings object names are validated before reaching SQL. The default channel now follows a PGQUEUER_PREFIX set after import pgqueuer.
Heartbeat loop awaits its task directly instead of asyncio.wait.

Fixed

Shutdown: in-flight jobs and schedules drain before the driver exits; scheduler fails fast on loop errors; cancellation propagates through task and listener health checks.
Completion watcher: polls until every waiter resolves, keeps one status query in flight under NOTIFY bursts, refresh_interval=None disables polling, no raise on exit after a cancelled waiter.
Retries: RetryRequested inside a TaskGroup retries the job; a failing retry_job marks the job failed; a retried job's context is released before re-queue.
Drivers: psycopg quotes the channel in LISTEN/UNLISTEN and delivers NOTIFYs only to their channel's callbacks; asyncpg and psycopg release the listener only on the last exit.
In-memory driver: log_jobs emits table_changed, statistics buckets accumulate, log_statistics(last=...) excludes the cutoff bucket.

Removed

PostgreSQL 13 (EOL). CI tests 14–18. Stay on v1.5.0 if you need 13.
pgqueuer-web and pgqueuer-prometheus Docker images. Build from tools/web/Dockerfile and tools/prometheus/Dockerfile.
dev and docs extras. Now uv dependency groups.

Full Changelog: v1.5.0...v1.6.0