[mpm-itk] apc cache ("answer" after long time)

Dzianis Kahanovich mahatma at bspu.unibel.by
Thu Dec 6 14:01:53 CET 2012


Steinar H. Gunderson пишет:
> On Wed, Dec 05, 2012 at 01:30:49PM +0300, Dzianis Kahanovich wrote:
>> Thanks, good. But after day of work with pthread mutexes I got multiple "child
>> died with signal 11" and even cron's restart.
> 
> If cron restarts, that's probably an issue with cron, not with pthreads. :-)

There are my English... ;)
In cron I restart apache if last N messages (not lookup now precise script
logic) is "child dies with signal 11". In other words, sometimes "child died" is
not fatal for whole system, but sometimes this is "avalanche". Now it was
"avalanche".

> If a child dies with signal 11, you should probably try to get a backtrace.
> The Apache wiki has more information:

May be, but I lazy prefer to use eaccelerator now. May be, once eaccelerator
will be outdated for current PHP again, I will try to use APC (again)... ;) With
eaccelerator I have absolutely no errors. Even I remember third-party tests -
eaccelerator is faster. Just APC is near "mainstream".

Just strange: people say have no this problem...

>   http://httpd.apache.org/dev/debugging.html
> 
>> But it may be result of "my vfork()" (CLONE_VFORK here).
> 
> Note that vfork() is really only ever intended to be followed by execve()
> (or _exit()).

vfork is not problem. Just it source of much more "child died", but without it
it happened too. Early just with vfork it was completely not working, but
without - just sometimes (once per week?) apache was restarting via cron script
and everytime "child died" message, but not too much.


-- 
WBR, Dzianis Kahanovich AKA Denis Kaganovich, http://mahatma.bspu.unibel.by/



More information about the mpm-itk mailing list