[squid-users] rerequesting and recaching

I run a squid httpd accel cluster and I've noticed that squid will often
have several requests going for objects which are popular enough and old
enough to have them cached (to put it mildly).

I ran a store.log for a bit and it is, indeed, recaching existing items.
In the span of a few seconds, it dropped and swapped this file twice:

991334057.900 RELEASE 01 00001448 1CAF1CC31DF4D20555B78E535EA79A26 ?
    ? ? ? ?/? ?/? ? ?
991334061.945 SWAPOUT 02 000004A1 1CAF1CC31DF4D20555B78E535EA79A26 200
991334057 974918817 -1 text/plain 312147/312147 GET
http://www.zeldalegends.net/misc/comedy/pkkill.WMV
991334073.555 RELEASE 02 000004A1 1CAF1CC31DF4D20555B78E535EA79A26 200
991334057 974918817 -1 text/plain 312147/312147 GET
http://www.zeldalegends.net/misc/comedy/pkkill.WMV
991334073.835 SWAPOUT 02 000004A1 1CAF1CC31DF4D20555B78E535EA79A26 200
991334073 974918817 -1 text/plain 312147/312147 GET
http://www.zeldalegends.net/misc/comedy/pkkill.WMV

What is causing this tail-chasing? The cache_dir is only 60% full and the
Apache response doesn't change.

I notice this problem on the files I have set to refresh on every hit (for
various reasons), but that may just be because they're huge and readily
apparent.

My current assumption is squid gets too many files open and starts
bypassing the cache_dir to retrieve the file, but still tries to cache it
at the end (which, of course, just makes the problem even worse).

Any suggestions on how I can reduce or stop this recaching?

Thanks,
        -- Brian
Received on Thu May 31 2001 - 23:38:07 MDT