What is your squid fd limit and your system ulimit (ulimit -n) for squid
user ?
-- 
Best regards,
Loïc BLOT, 
UNIX systems, security and network expert
http://www.unix-experience.fr



Le mercredi 29 mai 2013 à 14:03 -0400, Ron Wheeler a écrit :
> Have you looked at garbage collection as a possible source of the problem?
> If you really have a 300 GB cache, that might take a long time to 
> process during GC.
> 
> You might want to post your GC settings to see if anyone has a 
> suggestion or can eliminate GC as the source of your problem.
> 
> The fact that everything works before, during and after except that your 
> CPU is too busy to talk to you, does make one suspect that it is a 
> normal occurrence as far as Squid knows and not worth logging or noting 
> for humans to read.
> 
> 
> Ron
> 
> On 29/05/2013 1:48 PM, Mike Mitchell wrote:
> > I've hit something similar.  I have four identically configured systems 
> > with 16K squid FD limit, 24 GB RAM, 300 GB cache directory.  I've seen the 
> > same failure randomly on all four systems.  During the day the squid 
> > process handles > 100 requests/second, with a peak FD usage around 8K FDs.  
> > In the evenings the load drops to about 20 requests/second, with an FD 
> > usage around 1K FDs.  CPU usage hovers less than 10% during this time.
> > Randomly one of the four systems will start increasing its CPU usage.  It 
> > takes about 4 hours to go from less than 10% to 100%.  During the four 
> > hours the FD usage stays at 1K and the request rate stays right around 20 
> > requests/second.  Once the CPU reaches 100% the squid service stops 
> > responding.  About 20 minutes later it starts responding again with CPU 
> > levels back down below 10%.  There is nothing in the cache log to indicate 
> > a problem.  The squid process did not core dump, nor did the parent restart 
> > a child.
> >
> > I have not seen the problem during the day, only after the load drops.  The 
> > hangs do not coincide with the scheduled log rotates.  The one last night 
> > recovered a half-hour before the log rotated at 2:00 AM.
> >
> > Every one of my hangs have been proceeded with a rise in CPU usage, and 
> > squid recovers on its own without logging anything.
> >
> > I have a script that does
> >    GET cache_object://localhost/info
> >    GET cache_object://localhost/counters
> > every five minutes and puts the interesting (to me) bits into RRD files.
> > Obviously the script fails during the 20 minutes the squid process is 
> > non-responsive.
> >
> > ________________________________________
> > From: Stuart Henderson [[email protected]]
> > Sent: Tuesday, May 28, 2013 12:01 PM
> > To: [email protected]
> > Subject: Re: Squid CPU 100% infinite loop
> >
> > On 2013-05-17, Alex Rousskov <[email protected]> wrote:
> >> On 05/17/2013 01:28 PM, Loïc BLOT wrote:
> >>
> >>> I have found the problem. In fact it's the problem mentionned on my
> >>> last mail, is right. Squid FD limit was reached, but squid doesn't
> >>> mentionned every time the freeze appear that it's a FD limit
> >>> problem, then the debug was so difficult.
> >> Squid should warn when it runs out of FDs. If it does not, it is a
> >> bug. If you can reproduce this, please open a bug report in bugzilla
> >> and post relevant logs there.
> >>
> >> FWIW, I cannot confirm or deny whether reaching FD limit causes what
> >> you call an infinite loop -- there was not enough information in your
> >> emails to do that. However, if reaching FD limit causes high CPU
> >> usage, it is a [minor] bug.
> > I've just hit this one, ktrace shows that it's in a tight loop doing
> > sched_yield(), I'll try and reproduce on a non-production system and open
> > a ticket if I get more details..
> >
> >
> >
> >
> >
> 
> 

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to