On Wed, 30 Sep 2009 17:22:34 -0300, Marcus Kool <marcus.kool@xxxxxxxxxxxxxxx> wrote: > Luis Daniel Lucio Quiroz wrote: >> Le mercredi 30 septembre 2009 11:14:43, Marcus Kool a écrit : >>> What are the values for the parameters cache_swap_low and >>> cache_swap_high ? >>> For a large cache it is recommended to have them close to each other. >>> E.g. >>> cache_swap_low 90 >>> cache_swap_high 91 >>> >>> You can also add >>> refresh_pattern (cgi-bin|\?) 0 0% 0 please use: refresh_pattern -i (/cgi-bin/|\?) 0 0% 0 (the other pattern was tried by the devs and found to catch a few too many things). >>> since dynamic pages should not be cached. >>> >>> Marcus >>> >>> Luis Daniel Lucio Quiroz wrote: >>>> Hi all, >>>> >>>> Well, after implementing cache, in a heavy environment (with about 5k >>>> users) I'm seeing that our squid is not freeing far enough objects, our >>>> 100GB disk cache fills in 5 days. I wonder I misunderstood >>>> refresh_pattern options. refresh_pattern is all about *extending* the period for which an object is cached. increasing the cache size. It does not assist with reducing things. >>>> >>>> I have this: >>>> >>>> refresh_pattern -i \.png$ 2880 1% 5760 >>>> refresh_pattern -i \.zip$ 0 1% 1440 >>>> refresh_pattern . 0 20% 4320 >>>> >>>> >>>> I wonder to know how would be if I want to keep objects 4 days and in >>>> 5th >>>> object could be discarted. For specific objects you can't. Only the website operator can determine that kind of fine-grained object control. >>>> >>>> TIA >>>> >>>> LD >> cache_swap_low 95 >> cache_swap_high 97 >> >> is this enoght=? Maybe yes, maybe no... check the average object size (squidclient mgr:info or the info page of cachemgr.cgi). Calculate 2% of your cache_dir size and divide by the avg object size. This is how may disk erase operations are required of the CPU+disks when Squid needs to make more space. > > If the disk gets full.... > It gets full because there is too little space in the file system > or the space is reclaimed too slow by Squid. Because: * you have set cache_dir size larger than your disks can hold. * you are logging to the disk where cache_dir resides and logs are filling up more than their share * your system and other apps are using the same hard drive as squid cache_dir and need more space than you left available. The Q's Marcus asks will lead to your solution... > > What is the output of 'df' ? > What are the values for parameter cache_dir ? > > And I recommend to change the parameters: > cache_swap_low 93 > cache_swap_high 94 > > -Marcus Amos
