At a few installations of squid 4.12 (patched for GREASE) on NetBSD 9, I'm seeing that occasionally one of the listening ports no longer accepts connections (it doesn't reject them, but a connection does not get established). The port appears random; it's not the same every time and isn't related to ports with SSL interception. A restart of squid fixes it. Looking through the logs, this appears to coincide with lines such as: 2020/10/14 22:32:16 kid1| ERROR: getsockname() failed to locate local-IP on local=[::] remote=10.0.106.147:61996 FD 25 flags=1: (22) Invalid argument 2020/10/14 22:32:16 kid1| BUG 3556: FD 25 is not an open socket. This looks similar to Alex Rousskov's recent observations: https://bugs.squid-cache.org/show_bug.cgi?id=3556#c15 However, we have also seen with at sites where there is no SSL interception (the above lines are from such an installation). Based on this, I have 4 questions; 1) Am I right that triggering BUG 3556 could lead to the described symptoms? 2) Is this a squid bug or triggered by a problem in the underlying OS? If the latter, where to start looking? 3) What workarounds are there? Given that a restart fixes it, in some respects it would be better for squid to quit so it can be restarted rather than continue to run in a half-working state. 4) Related to 3), are there any other ways to detect the problem when it is happening besides parsing logs or testing if all ports are accepting connections? This could be used to trigger an automated restart as a temporary workaround. -- Stephen _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
