fix: handle tcp_closed/ssl_closed gracefully instead of crashing

[samwar]

Summary

When MongoDB closes a TCP connection (idle timeout, server restart, network issues), mc_worker currently crashes with reason tcp_closed, producing SASL CRASH REPORT output and error logs on every closure. This became noticeably more frequent after upgrading from MongoDB 5 to MongoDB 6.

This PR changes handle_info/2 to treat closure messages as an expected shutdown, so the supervisor can restart the worker cleanly without log noise, while still preserving the reason for observability.

Change

In src/connection/mc_worker.erl, the handle_info/2 clause matching {tcp_closed, _} / {ssl_closed, _} now returns:

{stop, {shutdown, NetR}, State}

instead of {stop, tcp_closed, State}, where NetR is the exact closure atom (tcp_closed or ssl_closed).

OTP treats {shutdown, _} as an expected termination and skips crash reports, while preserving the termination reason for supervisor observability and monitoring processes. An error_logger:info_msg/2 line was added before termination so operators still have visibility into each closure.

Error handling for tcp_error / ssl_error is unchanged — those still stop with the underlying error reason.

Testing

• Adds test/tcp_closure_bug_SUITE.erl covering:
  • {shutdown, tcp_closed} / {shutdown, ssl_closed} on socket closure
  • Preservation of tcp_error / ssl_error termination semantics
  • Preservation of data-message processing (tcp / ssl payload messages)
  • Explicit-stop behavior unchanged
• rebar3 compile passes.
• New suite: 7 cases, all passing.

Notes

• No API or behavior change for callers — connections still terminate and are restarted by the pool supervisor; the only difference is the absence of spurious crash reports.

[samwar]

@comtihon Any chance you could look at this PR?

[comtihon]

hi, can you please check why tests are failing?

[samwar]

@comtihon the failure looks unrelated to my my changes and due to a gpg key issue when setting up the test containers. Is this fix something you want to do, or should I just include the change in this PR, even though it's a bit unrelated?

[samwar]

@comtihon My original PR passed all of the tests, and the push I made only changed comments. I think the test is just flaky. Since you're an admin, you can re-run the failed jobs, and that should make the PR green.

[samwar]

@comtihon I made an empty commit which re-triggered the tests and they all pass now. Can you take a look at this again?

[comtihon]

sorry, was out in the wild for a week.
thank you for your contribution!