在部署期间暂停所有 Laravel 队列

There is a window during most deploys where the code on disk has changed but the workers running against it have not restarted yet. A worker that reserves a job in that window deserializes a payload written by the old code and runs it through the new code. Most of the time nothing happens. Occasionally a renamed job class or a changed constructor signature turns it into a failed job you have to replay by hand.

The usual workaround is maintenance mode, which stops far more than the queue, or queue:restart , which is a request rather than a guarantee since a worker only sees it between jobs. Laravel 13.25 adds a third option: a switch that stops every worker on every connection from reserving new work, without touching HTTP traffic.

The Command

php

工匠

队列：暂停

- 全部

And to undo it:

php

工匠

queue:resume

- 全部

The queue argument on both commands is now optional, so --all is the only thing you pass. Without the flag they behave as before and take a connection:queue pair.

The same switch is on the Queue facade for deploy scripts written in PHP, or for a controller behind an admin panel:

使用

Illuminate\Support\Facades\Queue
;
队列
::
pauseAll
（）；
// ...
队列
::
resumeAll
（）；

What Pausing Actually Does

A paused queue stops workers from reserving new jobs. The worker process stays alive and keeps looping, it just sleeps instead of popping. A job already being processed when you pause runs to completion, so a queue:pause --all in a deploy script does not interrupt anything mid-flight. Producers are unaffected too: SomeJob::dispatch() keeps writing to Redis or the database, and those jobs sit there until you resume.

Mechanically it is one cache key, illuminate:queues:paused , written with forever() . Workers already read the cache once per loop to check for restart and per-queue pause signals, and the global key is fetched in the same many() call, so this adds no extra round trips. QueueManager::isPaused() 和 getPausedQueues() both report a queue as paused when either the global key or its own key is set.

The Two Switches Are Independent

pause() 和 pauseAll() write different keys, and they are deliberately not aware of each other. If a queue was paused individually, resumeAll() leaves it paused:

队列
::
暂停
（
'redis'
,
“进口”
（英文）：
// parked while we investigate a bad job
队列
::
pauseAll
（）；
// deploy starts
队列
::
resumeAll
（）；
// deploy finishes
队列
::
isPaused
（
'redis'
,
“进口”
（英文）：
// 真的

This is the behavior you want in a deploy script. Somebody parked the imports queue on purpose an hour ago, and a deploy running resumeAll() should not quietly undo that. Clearing an individual pause still takes queue:resume redis:imports 。

活动

Two events fire alongside the existing per-queue QueuePaused 和 QueueResumed :

使用

Illuminate\Queue\Events\QueuesPaused
;
使用

Illuminate\Queue\Events\QueuesResumed
;
事件
::
听
（
功能
（
QueuesPaused
$event) {

日志
::
警告
（
'All queues paused'
（英文）：
});