This is a Windows specific issue where a virtual thread attempting to establish a connection appears to hang, and the test eventually times out. It initially looked like an issue on older releases of Windows but it turns out to be an issue that arises when a SOCKET handle is recycled quickly. Thanks to Daniel Jeliński for spending time to find a way to reproduce this quickly, up to now it was too intermittent and needed tens of thousands of runs to having some chance of reproducing.
When a SOCKET has been polled, the event is disabled rather than removed from wepoll / AFD poll. When the SOCKET is closed and immediately recycled then there is a race with the close event and registration the "new" socket for events. The change that we've tested is to remove the event after polling. This is done in the client thread rather than the poller to avoid racing with an async close. As part of this change I've removed a temporary change to the Windows Pipe implementation and also remove a connect timeout that went into a few tests to workaround the issue. ------------- Commit messages: - Fix AIX - Initial commit Changes: https://git.openjdk.org/jdk/pull/16742/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=16742&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8282726 Stats: 42 lines in 8 files changed: 9 ins; 11 del; 22 mod Patch: https://git.openjdk.org/jdk/pull/16742.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/16742/head:pull/16742 PR: https://git.openjdk.org/jdk/pull/16742
