We are using 4.1.12 standalone on RedHat Linux 7.3 servers and having rare HTTP requests delayed on their way into the Coyote HTTP/1.1 connectors. We have packet traces that indicate the request is delayed by exactly 100 seconds, but otherwise is received and responded to as one would expect. Other requests immediately before and after the problematic ones are handled without any significant delay. We are wondering if others have noticed this problem or similar ones that sound like it?
In the coyote connector code, it appears that the request sockets are set by default to have 100 second SO_LINGER timeouts for socket close() calls. Of course, this looks suspicious given our problem, but we have not been able to identify any way the blocked close() operation could affect incoming accept() requests. It is clear that running out of processor threads in the thread pool could cause such a delay, but we are not running under loads where this would happen, especially for 100 seconds. Any ideas out there? Randy Watler Finali Corporation
