Thanks, this looks good!
Can you explain a bit about the 64 seconds interval (e.g. why '4'?)

Thanks, this looks good! Can you explain a bit about the 64 seconds interval (e.g. why '4'?)

Yeah, don't know what I was thinking there 😅. Changed it to sleep the proper amount with the remaining time.

@isaacdonaldson thanks for this!
I'm almost finishing a rewrite of the queue subsystem:

• Added postgres based queue
• "seamless" interface to pick between redis (sidekiq) and postgres queue
• ability to "compile out" irrelevant queue providers
• only a single trait is now needed: BackgroundWorker

Custom queues: In the original sidekiq it was used mostly for ordering. You ordered the critical queues first:

- critical
- system_messages
- default

And sidekiq would iterate in-series over these, it would only pull from the next queue if the preceding one was empty -- creating a kind of a priority mechanism.

I'm wondering if you experience the same and/or if you have other uses for custom queues?

PS: every worker is defined as its own separate, statically typed job, so we do have separation between jobs, they just go into the same queue in redis (the same physical list: default)

@jondot - Is there somewhere where I can check out your queue implementation? Would love to take a look :)

I agree that a better defined trait like BackgroundWorker would be a good idea when choosing between Postgres, Redis, or tokio 👍 .

I don't think that extra/custom queues are a necessity, but I do think storing a unique 'queue label' with the job (in Postgres and Redis) could be a good idea for debugging/observability/human readability. That is what I was wanting to use them for, but it could also be optional.

I also think that having a priority mechanism is quite important. I think using the same queue (Redis list/Postgres table) could be fine, as long as there is the ability to specify priority.

I don't mind giving some feedback or expanding on what I mean once I can see the implementation :)

Hi @isaacdonaldson - the final implementation is now merged, I hope you can take a look at the new code -- see if there's anything to improve from this PR into the new implementation