Chris Woodfield wrote:
Hi,
Reading the squid FAQ, it's obvious to me that putting cache_dirs on a
RAID (particularly RAID5) has serious performance penalties and is
highly discouraged. However, what's not as clear is how squid deals with
single-disk failures and whether or not it handles failures gracefully
enough to obviate the need for RAID.
If I have a squid running multiple cache_dirs on single disks, and one
disk suffers a failure, how does squid respond? Will it simply stop
using that cache_dir and soldier on, or can this cause an application
crash?
Probably crash. Unfortunately Restart itself and crash again. Repeat...
Also, when starting up squid, what is the effect of an unavailable
cache_dir? I'm thinking of a situation where squid is restarted before a
bad disk can be replaced.
If cache_dir is completely absent/missing squid will throw up a message
about needing -z option to create the cache and exit.
It should not be too much work to make the code ignore individual
cache_dir missing and accept if at least one is present.
If you are interested in sponsoring it let us know.
If squid does have problems here, could using pairs of RAID1 partitions
be an acceptable compromise, with the cost of reduced total storage?
It's the disk- write-duplication that slows the HDD and thus squid down
on all/most TCP_MISS objects. RAID without the duplication is much less
of a penalty, but still leaves you with the partially corrupt cache_dir
problem of possible crashes.
Amos
--
Please use Squid 2.6STABLE17+ or 3.0STABLE1+
There are serious security advisories out on all earlier releases.
