使用倒数计时，celery任务未出现在task.apply_async之后的队列中

Question

I Hope this question qualifies for stack overflow.

Working with celery 4.2.0 and Redis as broker and backend.

Having a task

@shared_task()
def add(a, b):
    return a+b

And while a worker is active , running the fallowing command:

add.apply_async(countdown=60)

Results in the task not being registered to the default celery queue, but still being executed after the period of time stated in countdown

Why is that, and how can I look for all pending tasks? Doing this would have worked if the task would be registered to the queue:

    with celery_app.pool.acquire(block=True) as conn:
        tasks = conn.default_channel.client.lrange('celery', 0, -1)

If I terminate the worker while task havent been started I get the fallowing:

[WARNING/MainProcess] Restoring 1 unacknowledged message(s)

This tells me the task is kept somewhere else other then the queue, but I cannot figure out where

Answer 1

Turns out this is actually the expected behavior as workers grab pending tasks even before execution to speed things up and thus remove them from queue.

To achieve what I wanted, which is disable workers from taking tasks from queues before actually starting to work on them, and losing track of that task, I had to use 2 celery settings together

task_acks_late = True
worker_prefetch_multiplier = 1

This way a task is removed from its original queue, but still exists in a queue called 'unacked', which allows me to monitor it. Optimizing celery

Be advised that using the 'acks_late' acks late setting has some side affects you need to be aware of, such that if a worker unexpectedly terminated, the task will be restored and re run next time the worker is revived.