[3m1n3nc3]

Summary

ark queue:work can only work one queue, and a bare worker silently ignores jobs dispatched to any other queue. This caught Arkyc in production: its single PM2 worker drained only default while every job went to a named queue, so sessions stalled in processing and no webhook was ever delivered.

--queue now accepts:

ark queue:work                      # unchanged: the connection's default queue only
ark queue:work --queue=webhook,ocr  # priority list: ocr runs only while webhook is empty
ark queue:work --queue=*            # every queue on the connection
ark queue:work --queue=webhook,*    # webhook first, then anything

A bare queue:work is deliberately unchanged. Making it work every queue would break split deployments where a worker must not take jobs it can't run.

Changes

• Worker: runNextJob() and daemon({ queue }) take a string or a priority-ordered list; ANY_QUEUE ('*') is exported.
• QueueWorkCommand: parses --queue into a deduplicated list; the startup line says what it drains.
• QueueContract.popAny() backs the wildcard. The base implementation works the default queue, so custom drivers behave as before until they override it.
  • database: the oldest due job across all queues (the existing query minus the queue filter), FIFO by arrival, so a busy queue can't starve the rest. The reserved job keeps its own queue name.
  • redis: discovers queues with SCAN (never the blocking KEYS), including queues holding only delayed jobs, and works them round-robin. RedisClient gains scan, which ioredis already has.
• Docs: the queue guide and README cover all three --queue forms and the "every queue needs a worker" rule. The jobs guide warns that a bare worker ignores onQueue(...) jobs. The CHANGELOG has an Unreleased entry.
• A separate style: commit holds unrelated pnpm format output (6 files, formatting only).

Tests

12 new tests in packages/queue/tests/multi-queue.test.ts:

• Worker: priority order, skipping empty queues, never touching unlisted queues, * draining all, and webhook,* ordering.
• database popAny: run against a fake table that applies the real filters. Checks there's no queue filter and the reserved job keeps its own queue.
• redis popAny: run against a fake client. Checks discovery across two SCAN pages, round-robin order, and a queue holding only delayed jobs being found once due.

tests/MemoryQueue.ts now keeps jobs per queue (unnamed pushes and pops still use default), so the existing worker tests pass unchanged. All 24 queue tests pass.

[codecov]

Codecov Report

❌ Patch coverage is 87.01299% with 10 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
packages/queue/tests/MemoryQueue.ts	78.78%	7 Missing ⚠️
packages/queue/src/Contracts/QueueContract.ts	0.00%	2 Missing ⚠️
packages/console/src/prepare/BuildInterfaces.ts	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!