[squid-users] TCP_NEGATIVE_HIT/200, Vary object loop

Hello,

just started squid 3.0 (stable2) in production. Three web servers,
each has squid in front, all three squids are set as siblings to each
other.

First of, I'm really surprised to get TCP_NEGATIVE_HIT/200. Lots of them:

# wc -l squid_access.log
1482641 squid_access.log
# grep -c 'TCP_NEGATIVE_HIT' squid_access.log
118348
# grep -c 'TCP_NEGATIVE_HIT/200' squid_access.log
118003

so 10% of all requests are negative hits

Here are "storeCheckCachable() Stats":
no.not_entry_cachable 736
no.wrong_content_length 1
no.negative_cached 379766
no.too_big 0
no.too_small 0
no.private_key 0
no.too_many_open_files 0
no.too_many_open_fds 0
yes.default 144468

Interesting enough, these are mostly images and js/css objects. Then I
though it could be related to:

acl really_static urlpath_regex -i
\.(jpg|jpeg|gif|png|tiff|tif|svg|swf|ico|css|js)$
acl nocache_cookie req_header Cookie NOCACHE\=1
cache allow really_static
cache deny nocache_cookie

And indeed - if I come with Cookie: NOCACHE=1, for objects matching
really_static acl I get one of:
TCP_NEGATIVE_HIT/200
CD_SIBLING_HIT/192.168.10.162
TCP_IMS_HIT/304

So all local hits are named TCP_NEGATIVE_HIT if they match cache
allow. Though I'm not sure if this is a feature or a bug.

The next question is about Vary header. I get absolutely amazing
amount of these errors in cache.log:

2008/03/14 10:46:54| clientProcessHit: Vary object loop!
2008/03/14 10:46:54| varyEvaluateMatch: Oops. Not a Vary match on
second attempt, 'http://some.url' 'accept-encoding'
2008/03/14 10:46:55| clientProcessHit: Vary object loop!
2008/03/14 10:46:55| varyEvaluateMatch: Oops. Not a Vary match on
second attempt, 'http://some.other.url'
'accept-encoding="gzip,%20deflate"'

A rough number:
# grep -c 'Vary object loop' cache.log && wc -l squid_access.log
244816
1842602 squid_access.log

Any idea what kind of loop that is and how to avoid it?

Thanks!

Aurimas
Received on Fri Mar 14 2008 - 08:58:27 MDT