Does this attached patch help? This is to verify an hypothesis where
killed keepalive connections are still in the defer_linger_chain
without a worker to handle them.
I don't see "interrupted, but continuing" triggering in your log but
possibly it happened earlier..
On Wed, Jul 7, 2021 at 5:51 PM Stefan Eissing
<[email protected]> wrote:
>
> The things I do for you...
>
> [Wed Jul 07 15:48:52.591983 2021] [mpm_event:debug] [pid 48747:tid
> 4385197568] event.c(599): AH: wakeup_listener: start
> [Wed Jul 07 15:48:52.592000 2021] [mpm_event:debug] [pid 48747:tid
> 4385197568] event.c(603): AH: wakeup_listener: apr_pollset_wakeup
> [Wed Jul 07 15:48:52.592046 2021] [mpm_event:debug] [pid 48747:tid
> 4385197568] event.c(610): AH: wakeup_listener: ap_queue_info_term
> [Wed Jul 07 15:48:52.592188 2021] [mpm_event:debug] [pid 48747:tid
> 4385197568] event.c(629): AH: wakeup_listener: pthread_kill
> [Wed Jul 07 15:48:52.592166 2021] [mpm_event:debug] [pid 48747:tid
> 123145502474240] event.c(1985): (4)Interrupted system call: AH: pollset
> returned listener_may_exit=1 connection_count=2
> [Wed Jul 07 15:48:52.592523 2021] [mpm_event:debug] [pid 48747:tid
> 4385197568] event.c(635): AH: wakeup_listener: end
> [Wed Jul 07 15:48:52.592727 2021] [mpm_event:trace1] [pid 48747:tid
> 4385197568] event.c(2942): graceful termination received, join workers
> [Wed Jul 07 15:48:52.592515 2021] [mpm_event:trace1] [pid 48747:tid
> 123145502474240] event.c(1809): All workers are busy or dying, will close 2
> keep-alive connections
> [Wed Jul 07 15:48:52.593381 2021] [mpm_event:debug] [pid 48747:tid
> 123145502474240] event.c(1985): AH: pollset returned listener_may_exit=1
> connection_count=2
> [Wed Jul 07 15:48:52.593478 2021] [mpm_event:debug] [pid 48747:tid
> 123145502474240] event.c(1985): (4)Interrupted system call: AH: pollset
> returned listener_may_exit=1 connection_count=1
> [Wed Jul 07 15:48:57.552623 2021] [mpm_event:debug] [pid 48747:tid
> 123145502474240] event.c(1985): (70007)The timeout specified has expired: AH:
> pollset returned listener_may_exit=1 connection_count=1
> [Wed Jul 07 15:48:57.552658 2021] [mpm_event:trace1] [pid 48747:tid
> 123145502474240] event.c(1809): All workers are busy or dying, will close 0
> keep-alive connections
>
>
>
> Btw. I opened a bottle of nice wine - this is entertaining. ;-)
>
> - Stefan
>
> > Am 07.07.2021 um 17:48 schrieb Yann Ylavic <[email protected]>:
> >
> > On Wed, Jul 7, 2021 at 5:45 PM Stefan Eissing
> > <[email protected]> wrote:
> >>
> >>> Am 07.07.2021 um 17:39 schrieb Ruediger Pluem <[email protected]>:
> >>>
> >>> Hm. The ap_log_error statically writes listener_may_exit=0. Can you put
> >>> the actual value of listener_may_exit in the log message?
> >>
> >> I put it statically since that was the value in the "if ()" 6 lines above.
> >> it should have been visible then and there to work.
> >
> > The above if () does not always break.
> >
> >
> > Cheers;
> > Yann.
>
Index: server/mpm/event/event.c
===================================================================
--- server/mpm/event/event.c (revision 1891217)
+++ server/mpm/event/event.c (working copy)
@@ -1940,15 +1940,6 @@ static void * APR_THREAD_FUNC listener_thread(apr_
rc = apr_pollset_poll(event_pollset, timeout_interval, &num, &out_pfd);
if (rc != APR_SUCCESS) {
if (APR_STATUS_IS_EINTR(rc)) {
- /* Woken up, if we are exiting or listeners are disabled we
- * must fall through to kill kept-alive connections or test
- * whether listeners should be re-enabled. Otherwise we only
- * need to update timeouts (logic is above, so simply restart
- * the loop).
- */
- if (!listener_may_exit && !listeners_disabled()) {
- continue;
- }
timeout_time = 0;
}
else if (!APR_STATUS_IS_TIMEUP(rc)) {
