Hey Soren!
Thanks a lot for the answer, I appreciate it.
The first option, with writing a query, worked very well!

# Use the running config to get the node name.
conf = Oban.config()

Oban.Job
|> where([j], j.state == "executing")
|> where([j], fragment("?[1]", j.attempted_by) == ^conf.node)
|> Repo.all()

For curiosity’s sake, using

Oban.Met.latest(Oban, :exec_count, filters: [node: "my-node-name"])

always gives me an empty map as a result.

However,

Oban.Met.latest(Oban, :full_count, filters: [node: "my-node-name", state: "executing"])

always gives the correct result! Why would that be?
Does it happen that a Job has the “executing” state, but it does not appear in the exec_count gauge?

Edit:
I think I understand it. It’s due to a misunderstanding of what the job state means. A Job might currently have the “executing” state, but not be running, in fact, it can mean that the job will execute in the future, is this correct? Or it might just be my database being inconsistent.

That’s most likely what is happening. Jobs reported by exec_count have live processes running. It’s possible to have job records marked as executing in the database when the processes aren’t running anymore. Those are called orphaned jobs, and they typically happen during shutdown when jobs run longer than the queue shutdown period.

There’s more about that situation in the Ready for Production guide.

It doesn’t mean that the job may run in the future. Most likely, it was running in the past and the queue shut down.