RE: [squid-users] Overflowing filesystems

> -----Original Message-----
> From: Michael Puckett [mailto:Michael.Puckett@Sun.COM]
> Sent: Wednesday, November 23, 2005 6:26 PM
> To: squid-users
> Subject: Re: [squid-users] Overflowing filesystems
>
>
> Sorry if you see this again, I got a bounced mail from
> squid-cache.org
>

The mailing list doesn't allow HTML mail.

> Chris Robertson wrote:
>
>>>>> -----Original Message----- From: Michael Puckett
>>>>> [mailto:Michael.Puckett@Sun.COM] Sent: Wednesday, November
>>>>> 23, 2005 9:25 AM To: squid-users Subject: [squid-users]
>>>>> Overflowing filesystems
>>>>>
>>>>> I am running this version of squid:
>>>>>
>>>>> Squid Cache: Version 2.5.STABLE10 configure options:
>>>>> --enable-large-cache-files --disable-internal-dns
>>>>> --prefix=/opt/squid --enable-async-io --with-pthreads
>>>>> --with-aio --enable-icmp --enable-snmp
>>>>>
>>>
>>> I imagine you have some reason for disabling the internal
>>> DNS resolution. I'm a bit curious as to what it would be...
>>>
>>
> That is the way our admin set it up. This particular application is
> an internal (to the company) only caching system which (relatively)
> few users move (relatively) few VERY large, multi GB files from
> (relatively) few origins to (relatively) few destinations. We are not
> caching web pages.
>

Fair enough.

>>
>>>>> specifically enabled for large files. My cache_dir is 535GB and the
>>>>> cache_dir directive looks like this:
>>>>>
>>>>> cache_dir aufs /export/vol01/cache 400000 64 64
>>>>> cache_swap_low 97 cache_swap_high 99
>>>>>
>>>
>>> Aside from the unusually low number of directories for the
>>> amount of data, that all seems fine.
>>>

Obviously if all it's caching are really big files, you don't need many directories.

>>>>> Squid has consumed the entire partition:
>>>>>
>>>>> /dev/dsk/c1t1d0s7 537G 529G 2.2G 100%
>>>>> /export/vol01
>>>>>
>>>>> Not the 400GB expected in the cache_dir directive and is
>>>>> now giving write failures.
>>>>>
>>>>> Have I set something up wrong? Why has the cache_dir size
>>>>> directive been ignored and why isn't old cached content being
>>>>> released?
>>>>>
>>>
>>> Is Squid the only thing writing to this cache_dir? Is
>>> there only one instance of Squid running? Do you see a process like
>>> unlinkd running? Are there any errors in the cache_log? What OS are
>>> you running? Assuming (judging from your email address) it's
>>> Solaris, have you had a gander at the FAQ
>>> (http://www.squid-cache.org/Doc/FAQ/FAQ-14.html#ss14.1)?
>>>
>>
> Good call on the OS :) Yes, we are running a multiprocessor Solaris
> 10 system. There are no errors on the cache log other than the
> filesystem write failures as the filesystem fills up. The server is
> entirely dedicated to Squid as a cache server, the filesystem
> entirely dedicated to the cache.
>
> PS output shows: 0 S squid 20127 20121 0 40 20 ? 153
> ? Jul 15 ? 0:00 (unlinkd)
>
> with no runtime thus far. Yes, we have had a gander at the FAQ and
> have been running squid internally for a number of years now. This is
> the first time we have filled up so large a filesystem while running
> the largefile squid version however.
>
> -mikep
>

Huh... Well, I have no experience with either acceleration setups or the enable-large-cache-files compilation option, but I would have to advise enabling the cache_store_log and see if it gives any indication of what is going on here (clear the cache manually to get a fresh start). Just how big are the files you are caching? Would it be possible for the cache to be at 395.6GB (99% of 400) and a new object is requested that fills the partition (a staggering 140GB file)? Not sure how Squid would handle that. Then again, I find the possibility of a single 140GB file to be... Unlikely.

Chris
Received on Fri Dec 02 2005 - 11:08:13 MST