Queue priority across multiple containers

[rade-tomovic]

Hello,

If I have multiple containers running GoodJob, and these containers have 1 or more queues, I'm wondering how can I enforce global "priority" for certain jobs/queues.

I know what when running command I can use +queue1, queue2 to keep them in order, but if I have:
container1: queue1, queue2
container2: queue3
container3: queue4

How can I prioritize in a way that it's for example queue4 executed first, then 1 and 2 and then 3.

By container, I mean independent Docker based worker, but all of them pointing to the same DB.

Update: if this is better to do from the perspective of jobs, does it make sense to do:
self.priority = N for every job, and then just don't care for queues? I'm using version 3.30.1, so I know that there are differences between v4.

[bensheldon]

Could you explain what the goal is?

"Priority" only ever comes into play when there is a lack of capacity (eg more jobs are present than there are resources to process it). The best practice for performance is not to rely on priority at all: https://github.com/bensheldon/good_job?tab=readme-ov-file#doing-your-best-job-with-goodjob

[rade-tomovic]

Hey,

Thank you for the timely response. Situation is this:

Job 1: Important, let's say 5s latency
Job 2: Slightly less important, also 5s latency
Job 3: Not important, 500ms latency, but a 100x more frequent

You upload some file where every line triggers Job 1, then Job 2 and then a lot of Job 3.

My goal is to control execution in a way that whenever Job 1 enters execution, it has more priority to execute than anything else - basically like a priority queue data structure.

Also, current setup (and that can change if you think it's wrong) is that all of these jobs are in separate queues and even separate infrastructure. Infrastructure is mostly because of the safety in case that if something fails in less important part, it's not breaking whole functionality.

[bensheldon]

My goal is to control execution in a way that whenever Job 1 enters execution, it has more priority to execute than anything else.

You can do that with Active Job's numeric priority (MyJob.set(priority: -10).perform later).

The problem with priority is you can't preempt executing jobs. So if all your execution capacity is already running jobs, your newly created higher priority job won't get executed until one of those executors is finished with its current job. The only way around this is through isolation.

That link I shared is how I recommend setting up queues: use queues to group by latency, use labels to group by dependencies.

[rade-tomovic]

That's completely fine, my goal is not to have 10k non-important jobs overflowing resources. Even on a separate queue, it's still using the same database, so I'll implement both improvements:

• separating jobs between queues based on latency level, but also:
• adding job priorities based on importance for the system - jobs in execution are not the problem, only enqueueing more priority jobs before others