> -----Original Message----- > From: Michael Puckett [mailto:Michael.Puckett@xxxxxxx] > Sent: Wednesday, November 23, 2005 6:26 PM > To: squid-users > Subject: Re: 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@xxxxxxx] Sent: Wednesday, November >>>>> 23, 2005 9:25 AM To: squid-users Subject: >>>>> 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
