Auto-enqueue or perpetual jobs

[robacarp]

Archive.fm has an endless source of work. Right now I have it tracking a periodic job which enqueues other jobs, and it checks to see how many enqueued jobs exist and fills it up to 10 each run.

This is fiddly:

• I have to track which objects have been enqueued manually
• if I’m running >10 workers, it’ll go dry and I have to think about the enqueued numbers and frequency of the enqueueing job
• Especially when workers crash, jobs are getting suspended in the queue, so the list of “ready” jobs is inaccurate.
• there is a component of dual-run sometimes — I need to make sure I’m not doing double work for some reason

I think there should be a way to build a job type which never stops, never stalls, and never needs to be enqueued. Perhaps a PerpetualJob? Perhaps this can be solved with an after run hook and a the observability API? How would that get kicked off when it stalls?