Fix _sock_recv infinite loop when StatusDB TCP connection drops

[TKeji]

Problem

When using --reruns with pytest-xdist, every test makes two blocking TCP calls to the StatusDB server (get_test_failures and set_test_reruns in pytest_runtest_protocol). If the server-side connection drops, _sock_recv enters an infinite loop:

def _sock_recv(self, conn) -> str:
    buf = b""
    while True:
        b = conn.recv(1)
        if b == self.delim:  # b"" != b"\n" → never breaks
            break
        buf += b
    return buf.decode()

recv(1) returns b"" (empty bytes) on a closed socket, but the code only checks for the newline delimiter. Since b"" != b"\n" is always True, the loop never exits.

This causes xdist workers to hang indefinitely at ~90% CPU, appearing stuck on a test that never completes ([pytest-xdist running] ...). The hang persists until the process is manually killed.

Fix

Add a check for empty bytes from recv(1) and raise ConnectionError:

b = conn.recv(1)
if not b:
    raise ConnectionError("StatusDB connection closed unexpectedly")

The ConnectionError propagates as an INTERNALERROR that xdist handles by replacing the worker — much better than hanging forever.

Reproduction

Minimal reproduction (proves the infinite loop on the unpatched version):

import socket
from pytest_rerunfailures import SocketDB

s1, s2 = socket.socketpair()
s2.close()  # recv on s1 will now return b""

db = SocketDB()
db._sock_recv(s1)  # hangs forever on unpatched, raises ConnectionError on patched

Full reproduction in a test run: monkey-patch ServerStatusDB.run_connection to close the server-side connection after a few requests, then run pytest --reruns=1 -n 1 --dist=loadgroup. The worker hangs on the next test's db.get_test_failures() call in _sock_recv.

Impact

Affects any xdist run with --reruns enabled. Without --reruns, the TCP protocol is never exercised (pytest_runtest_protocol returns early), so the bug doesn't manifest.

Checklist

• Changelog entry in CHANGES.rst
• Test added
• Pre-commit hooks pass

[Copilot]

Pull request overview

This PR fixes a hang in the StatusDB socket protocol used during pytest-xdist runs with --reruns by making _sock_recv detect closed TCP connections and fail fast instead of looping forever.

Changes:

• Update SocketDB._sock_recv to raise ConnectionError when recv(1) returns empty bytes (closed connection).
• Add a regression test covering the closed-connection behavior.
• Document the fix in the changelog.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/pytest_rerunfailures.py	Adds an EOF (b"") check in _sock_recv and raises ConnectionError to avoid infinite loops on dropped connections.
tests/test_pytest_rerunfailures.py	Adds a unit test using socket.socketpair() to ensure _sock_recv raises on closed connections.
CHANGES.rst	Adds a 16.2 (unreleased) changelog entry describing the fix and impact in xdist runs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[icemac]

Hi @TKeji, thank you for your PR. It looks legit to me. I let copliot review it. Could you please check its comments?
The broken tests against main are fixed in #324.

[icemac]

Hi @TKeji Are you still working on this PR or can I close it as abandoned?

[janmrow]

Hello @TKeji and @icemac, this PR overlaps with a review comment on #358.

The new StatusDB authentication handshake can trigger the same _sock_recv() EOF behavior during worker startup, which may hang the entire xdist session.

Since #322 already contains the global fix, maybe it would make sense to finish and merge this PR first, then rebase #358 on top of it. What do you think?

The remaining Copilot comments seem to be limited to the regression test rather than the fix itself. If helpful, I'm happy to take care of that test cleanup and prepare a small follow-up commit to be cherry-picked to this branch by a maintainer.

If you’d rather not continue with this PR, I can add the EOF fix to #358 instead.
I mainly would like to avoid duplicating or superseding the existing work here without checking first.

[icemac]

@janmrow This PR look abandoned, so please add the EOF fix to your branch, so we'll not lose it and then we can close the PR here.

[TKeji]

Sorry for the long silence. This completely dropped off my radar. I should've closed this out months ago.

@janmrow please go ahead with your fix. Thanks for checking in on the PR first.

@icemac, I'm happy for the PR to be closed whenever the fixes are merged in the other PR.