Need help: Varnish unexpectedly flushing cache weekly

[Jordi Salvat]

Hey Josh! Long while indeed!

Thanks for your help.

We're using FILE (20GB) , ulimt -n is 1024 (is that what you're referring
to?) -- right now the process only has 122 files open, so that doesn't seem
to be the issue... but I'll watch out for next Wednesday.

I'll also look at SESS_WORKSPACE.

-- 
Jordi Salvat i Alabart
CTO & Project Manager
Salir.com
+34 676 49 86 00


2011/8/25 Joshua Levine <jlevine at iwin.com>

>
>
> I sent that reply explicit to crashes, crashes that actually appear
> addressed in your version, so I should also have asked, based on your
> references to a somewhat regular schedule, if you are using MALLOC or FILE,
> and also what you are allowing for NFILES… perhaps you are exceeding those
> resource allocations?
>
> Cheers,
> Joshua
>
> On Aug 24, 2011, at 9:09 AM, Jordi Salvat wrote:
>
> > Hi,
> >
> > we restarted our Varnish instance on Sunday 7th around 19:00.
> >
> > On Wednesday 10th 9:45 the child process restarted (causes unknown) --
> loosing the existing cache.
> >
> > ... the cache slowly grew again ....
> >
> > On Wednesday 17th around 2:00, the cache was somehow flushed, loosing
> about 75% of its objects.
> >
> > ... the cache slowly grew again ....
> >
> > On Wednesday 24th at 2:00, the cache was flushed again, loosing about 85%
> of its objects.
> >
> > On each of these occasions our cache hit rate went down from ~90% to
> under 50%. Needless to say this is highly undesirable.
> >
> > What might be causing this behaviour? Any hints and directions helping
> research welcome.
> >
> > Thanks.
> >
> > More details:
> >
> > Varnish version: 2.1.5 (Debian lenny package 2.1.5-1~lenny4)
> > O.S.: Linux s229-161 2.6.38 #1 SMP Sat Mar 26 09:51:44 CET 2011 x86_64
> GNU/Linux
> >
> > MRTG daily graphs for cache size and objects in cache:
> >
> > <varnish_cache-day.png>
> >
> > <varnish_objects-day.png>
> >
> > Note how the two graphs are slightly different: size goes down between
> 2:00 and 15:30, object count goes down suddenly at 15:30.
> >
> > Main VCL file attached (long to read -- feel free to ask questions).
> >
> > --
> > Jordi Salvat i Alabart
> > CTO & Project Manager
> > Salir.com
> > +34 676 49 86 00
> > <salir.vcl>_______________________________________________
> > varnish-misc mailing list
> > varnish-misc at varnish-cache.org
> > https://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20110825/440572e0/attachment-0003.html>