Re: [squid-users] rock problem on each squid restart

On 10/25/2013 06:09 AM, Omid Kosari wrote:

> on each squid restart the drive which is rock have some problems .
> that drive is very very busy after several minutes . the following line
> shows iostat after 18 minutes
>
> Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
> sdb 3004.00 41324.00 0.00 41324 0

This is normal: Squid needs to load disk cache index into RAM. That part
is not yet fully optimized in Rock, so the loading times may be
significant, especially for slow disks. There is definitely more
optimization work to be done here, but "disk very busy" is not really a
problem on its own.

> cache.log includes a lot of following lines
> WARNING: cache_dir[0]: Ignoring malformed cache entry meta data at
> 51174042469

Reporting more than a few similar warnings is a Squid reporting bug.
Besides being very annoying, it can slow your Squid down. This will be
eventually fixed. Meanwhile, you can comment out the "Ignoring
malformed" debugging line in fs/rock/RockRebuild.cc.

Without a difficult triage, I cannot say whether the warnings are benign
or indicate some kind of a problem. FWIW, they have been reported by
others as well. I can speculate that you will see more of them if you
restart Squid often.

> if again i restart squid it takes same time . in that time squid performance
> decreases and sometimes could not serve http . is this a normal behavior of
> rock ?
>
> here is my squid.conf
> cache_dir rock /cache2 101000 min-size=0 max-size=32767

The above tells Squid to write disk cache misses and read disk cache
hits regardless of the performance consequences. If your Squid receives
more traffic than your disk can handle, doing so is usually a bad idea.
See Performance Tuning at http://wiki.squid-cache.org/Features/RockStore

HTH,

Alex.
Received on Fri Oct 25 2013 - 15:30:05 MDT