Hey folks, I've been working on a consulting gig for a company hosting on SmartOS (which I'm admittedly pretty unfamiliar with), and have been running into an odd issue on mpm_event.
When doing a simple load test with ab, we keep running into this pattern: [Sun May 19 17:44:49.132658 2013] [mpm_event:notice] [pid 56832:tid 1] AH00489: Apache/2.4.2 (Unix) configured -- resuming normal operations [Sun May 19 17:44:49.132879 2013] [core:notice] [pid 56832:tid 1] AH00094: Command line: '/opt/local/sbin/httpd' [Sun May 19 17:45:10.182436 2013] [mpm_event:error] [pid 56832:tid 1] AH00484: server reached MaxRequestWorkers setting, consider raising the MaxRequestWorkers setting [Sun May 19 17:48:12.955763 2013] [mpm_event:crit] [pid 57728:tid 29] (81)File descriptor in bad state: apr_pollset_poll failed. Attempting to shutdown process gracefully [Sun May 19 17:48:13.955597 2013] [mpm_event:error] [pid 56832:tid 1] AH00485: scoreboard is full, not at MaxRequestWorkers [Sun May 19 17:50:35.796865 2013] [mpm_event:error] [pid 56832:tid 1] AH00485: scoreboard is full, not at MaxRequestWorkers (... it goes on with lots of "scoreboard is full, not at MaxRequestWorkers" and a nice sprinkling of "(81)File descriptor in bad state: apr_pollset_poll failed" until load goes away) While I am able to tune the MPM up to handle more, and that seems to be somewhat of a band-aid, I'm wondering what the root cause might be. I'm thinking that the "scoreboard is full, not at MaxRequestWorkers" is because the scoreboard isn't being cleared out yet (either because of a bug, or because the high load isn't letting Apache get to housecleaning), but clueless as to what's causing the apr_pollset_poll to fail in the first place and cause this downward spiral. Anyone with SmartOS or Solaris experience seen something like this?
