Don’t Lose Your ets Tables explains the principles that Immortal.ETSTableManager is based on.

I tried this at some point, and it worked except in the cases it didn’t. Which in reality were my fault, but it would have taken some major refactoring to make it work. In other words, from my personal experience I found that this method actually caused some weird code coupling issues.

Now I would recommend a separate tracker that requires a completion acknowledgement along with a job timeout. When a job is finished then the worker acknowledges the completion and it’s removed from the tracker; when the job had a timeout, it is purged from the tracker and re-queued. This way you don’t have to keep track of the failure state or its progress through whatever job stages you may have.

If your jobs need to be ordered or aren’t idempotent then this won’t work and I wish the best of luck to you.

You can use immortal. Or add a stash server that will store state on terminate function of dying GenServer,