Unregistering cache in registry happens after cache shuts down about nebulex HOT 2 CLOSED

Hey 👋 !! Yeah, like you said, there is no obvious way to address this. I agree that a different exception like Nebulex.RegistryLookupError may look a bit better, but the problem remains. Unregistering before shutdown I don't think is a suitable solution because, first of all, the cache supervisor can be explicitly stopped or it can just crash, and in that case, where are we going to unregister? Second of all, unregistering means deleting the context from the Registry which uses the :persistent_term, and that could be expensive, that's why it is handled by the registry itself monitoring the cache supervisor process, and when it dies, it unregistered (similar to Ecto for example). I will think about it, and see if there is something we can do about it, but I think it is important to understand your use case also, maybe I can help you more. For example, why the cache is crashing, or why you are stopping it? How your main supervision tree is configured, because, if you have other processes depending on the cache, perhaps you can use another strategy like :rest_for_one ensuring when the cache supervisor is restarted, their dependants are restarted too. anyway, it is just to throw some ideas, let me know your thoughts. Thanks!

from nebulex.

Registry can still listen to supervisor exit messages and unregister in case supervisor didn't do this!

I will think about it, see if there is something we can do about it, but I think it is important to understand your use case also, maybe I can help you more. For example, why the cache is crashing or why you are stopping it? How your main supervision tree is configured, because, if you have other processes depending on the cache, perhaps you can use another strategy like :rest_for_one ensuring when the cache supervisor is restarted, their dependants are restarted too. anyway, it is just to throw some ideas, let me know your thoughts

Our use case is nothing special: we have Cache process in our main Application supervisor tree and we see errors on the app shutdown, since there's still some residual traffic from client processes on other nodes (we use this adapter which broadcasts events to other nodes).

One possible way to go about this is to have special worker process among supervisor children which will unregister on exit. The same way Bootstrap process is used to leave cluster in Partitioned and Replicated adapters

from nebulex.