The discussion revolves around how to best queue a long-running job in Discourse. pfaffman initially tried using a once-off job and fork, but encountered timeouts due to the job exceeding the 60-second limit. Falco suggested using a regular job instead, which can handle longer durations. pfaffman confirmed the issue persisted even when using ./bin/rails s. Falco then pointed out that pfaffman was incorrectly calling the job directly with new.execute instead of using Jobs.enqueue, which is the proper method. pfaffman acknowledged the mistake and successfully resolved the issue, noting that the job now runs correctly. The conversation highlights the importance of using the correct job enqueuing mechanism in Discourse.
So I’ve got a plugin that kicks off Ansible to do a Discourse installation (I really, really, want to stop using WordPress to take installation orders!).
Initially, I was calling fork to kick it off, which worked fine in testing, but I thought that I should do it a more “Discourse Way”.
So then made a once-off job to run the process, but it gets killed like this:
I tried adding sidekiq_options queue: 'low' to the class, and then to the function that runs the job.
Then I skipped having the job run the process, and just ran it directly in the model with
Discourse::Utils.execute_command(*instructions)
It works just fine if I run it from rails c, but however I run it (this is a Ubuntu development environment, started with ./bin/unicorn, it gets killed.
Once-off jobs are jobs that will only execute, well, once. Looks like what you need is a regular job that will be scheduled by a trigger, like a request hitting a controller.
Regular jobs can be longer than 60s, as some do in normal usage.