Re: monit deadlock

[Martin Pala]

Martin Pala wrote:

> Jan-Henrik Haukeland wrote:
>
> > Martin Pala <address@hidden> writes:
> >
> >  
> >
> > > (gdb) set stopped=1
> > > (gdb) print stopped
> > > $3 = 1
> > > (gdb) continue
> > > Continuing.
> > > Program exited normally.
> > > (gdb)
> > >   
> >
> >
> > Hmm, I'm not sure. We use lots of signal blocks (which is good), but
> > here the TERM signal must have been blocked so do_destroy could not be
> > called. Try to move the 'do_signal_block();' from main() into
> > do_destroy *after* Run.stopped= TRUE;
> >
> > Does it help?
> >
> >  
> I tried it - it didn't helped. From the gdb output it seems that 
> signals works:
>
> #0  0x4002a354 in __pthread_sigsuspend () from /lib/libpthread.so.0
> #1  0x4002a118 in __pthread_wait_for_restart_signal () from 
> /lib/libpthread.so.0
> #2  0x400274dc in pthread_join () from /lib/libpthread.so.0
> #3  0x0804e7d7 in monit_http (action=-4) at monit_http.c:131
> #4  0x0804f17e in do_destroy (sig=15) at monitor.c:512
> #5  0x4002d685 in __pthread_sighandler () from /lib/libpthread.so.0
> #6  <signal handler called>
> #7  0x4002d91b in read () from /lib/libpthread.so.0
> #8  0x00000006 in ?? ()
> #9  0x00000038 in ?? ()
> #10 0x0805c651 in read_proc_file (
>   buf=0xbfffe880 "10233 (mozilla-bin) S 10232 690 690 0 -1 64 23 0 5 0 
> 141 117 0 0 9 0 0 0 1843919 89362432 14890 4294967295 134512640 
> 134720822 3221224560 3212834796 1080776422 0 0 4098 17453 3222590806 0 
> 0 33 0\n0 0 0"..., buf_size=4096, name=0xc3 <Address 0xc3 out of bounds>,
>   pid=10233) at process/common.c:98
> ...
>
> On line #10 was monit doing its normal work, on line #6 it was 
> interrupted by signal which was handled by do_destroy. Monit should 
> call monit_http(STOP_HTTP) which will set global httpd stopped flag to 
> true. For some reason the stopped flag is false => monit httpd don't 
> know that it should stop. I see no reason why stopped flag is still 
> false. The only thing which looks strange to me is on line #3:
>
> monit_http (action=-4)
>
> I usually  don't use debuggers => i'm not sure whether this is correct 
> (STOP_HTTP value is defined as 2 in monitor.h)
>
> Martin
>
>
>
> _______________________________________________
> monit-dev mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/monit-dev

It is solved. It seems that the problem was caused by race condition 
between monit_http(START_HTTP) and monit_http(STOP_HTTP). It seems that 
in the case that the signal was delivered in the meantime between the 
httpd thread_wrapper() reached from set_signal_block() to start_httpd(), 
stopped flag was set by do_destroy() signal handler to true and 
do_destroy() started to wait for httpd thread to terminate. As soon as 
monit returned from signal handler, it resumed by start_httpd(), which 
reseted stopped flag to false.

Martin