[5.3] let the worker sleep 1 second more when app is down for maintenance

[halaei]

If the queue driver, like sqs, supports long polling, it is better to set --sleep=0 when running queue:work console command. But if it is done so, the CPU will reach 100% when application is down by artisan app:down.

If you find this PR valid, please decide if it is backward compatible or not, and if it should be applied to earlier versions as well.

Thanks.

[GrahamCampbell]

👎 Don't use zero for sleeping. Makes no sense.

[taylorotwell]

@halaei can you explain more why you use sleep=0?

[halaei]

Short answer

long polling

Long answer

Sleeping is one of the worst practical solutions to wait for job when there is nothing in the queue. Because you don't want your jobs to be delayed because your workers are currently sleeping.
You only want your jobs to be delayed when all the workers are really busy.

One better solution is called long polling.

Amazon SQS supports long polling.

I also wrote my own Redis-based queue driver that supports long polling.

With long polling option available, it is sleeping that makes no sense any more. Because the queue server does the waiting for you in a better way.