Unregister gen_server name in terminate

[michalmuskala]

Currently, for every registered process there's a potential race condition where when the process terminates, the supervisor can restart it before the registry processes the DOWN message.
This isn't possible to avoid in all situations, but for servers that trap_exit and process terminate calls, this is entirely possible.

This change implements this adding a call to gen:unregister_name just after users terminate callback is processed.

[github-actions]

CT Test Results

    2 files     97 suites   1h 4m 51s ⏱️
2 215 tests 2 163 ✅ 52 💤 0 ❌
2 601 runs  2 545 ✅ 56 💤 0 ❌

Results for commit 83e8109.

♻️ This comment has been updated with latest results.

To speed up review, make sure that you have read Contributing to Erlang/OTP and that all checks pass.

See the TESTING and DEVELOPMENT HowTo guides for details about how to run test locally.

Artifacts

• Complete CT logs (Download Logs)
• HTML Documentation (Download HTML Docs)
• No Windows Installer found

// Erlang/OTP Github Action Bot

[Maria-12648430]

@michalmuskala why only gen_servers, ie why not also gen_statems and gen_events?

There is one potential problem with doing it this way. Very contrived, though, so I'm just pointing it out.
What may happen is this:

1. a gen_server process A gets started with a registered name
2. the process is unregistered while it is running, either by itself or by another process
3. another gen_server (or other even) process B gets started registered (or registered while it is running) with the same name
4. process A terminates and thereby unregisters the name it was started with but no longer "owns", actually unregistering process B

As I said, this is contrived, but we don't know what exists out there in the wild. AFAICT, there is nothing in the documentation saying that sth like this should not be done.

[michalmuskala]

I agree this should be done for all standard behaviours. I published this as a sort-of proposal, happy to handle the other behaviours, if the OTP team agrees this is a good idea in principle.

[josevalim]

@michalmuskala should the registry instead check if the process is alive before returning it? That’s what we do for Elixir’s Registry.

[michalmuskala]

Does it also check that on register commands and then just clears the registration to avoid returning already_registered error?

[josevalim]

Yes, all operations that lookup in the registry check if it is alive before returning, although it is a local registry. Unsure about the trade-offs in a distributed one.

[michalmuskala]

I agree this is a better approach, I'm going to close this