I've had two relatively recent and relatively painful investigations come down to this closure of keepalive connections on activity when an individual child process has no idle workers:
https://github.com/apache/httpd/blob/trunk/server/mpm/event/event.c#L2190 This is worse than a typical keepalive race because the closure is really in response to the poll seeing the activity. Does it warrant a WARNING level message? It is nice to at least get the connection details into the log. Or an option to queue/block for a worker? -- Eric Covener [email protected]
