Troy Wilson wrote:
Wasn't there a known problem with killing listener threads in the worker mpm? Do you know off the top of your head about where that got fixed? 2.0.?? A co-worker has what seems to be that problem.
FWLIW, there are two conditions I know of in current code:
1) when child needs to go away and all worker threads in the process are busy, which means the listener thread is in pthread_cond_wait(), which doesn't return EINTR when a signal is delivered to the thread
2) when pthread accept mutex is used and the listener thread in the child process needing to go away is blocked in pthread_mutex_lock(), which doesn't return EINTR when a signal is delivered to the thread
I guess based on your description of when it happens that you're hitting the message for some other condition.
Does this cause real problems other than the senseless messages in the log? The process is on its way out with or without a cooperating listener thread. The severity of the listener-thread-didn't-go-away message needs to be adjusted downward IMHO.
