[squid-users] Squid with Icap fails on every Nth request

From: Graeme Bisset <[email protected]>
Date: Fri, 15 Oct 2004 10:18:38 +0100

Hi,

I'm running the Webwasher Squid/Icap package -
http://www.webwasher.com/squid-icap/download/squid-2.5.STABLE5-icap-6-pr
e3.tar.gz

And the Example icap server -
http://www.i-cap.org/spec/icap-server10.tar.gz

Problem 1 - If I build them both and run them, they start to run but
very quickly, it seems to lock up until the connections time out. I can
work around this by enabling the CLOSECONN (close connections after
dealing with requests) option in the icap server (iserver.h)

Problem 2 - If I open a browser and browse to fairly simple site (it
seems to happen more with sites that load quickly i.e. ones on my local
network) and hit F5 to refresh several times then on every 10th refresh
or so an item on the page fails to load. I've managed to track this far
enough to find that whenever it fails, it is entering the following
section of code in icap_respmod.c

    if (!cbdataValid(icap)) {
            debug(81, 3) ("icapSendRespMod: failed to establish
connection?\n");
            return;
    }

If I then hit refresh again then every item on the page will load
successfully although it seems to be quite random when and where the
problem occurs. It's not always the same item that fails to load.

Here's my squid config:

visible_hostname blocker
request_header_max_size 13 KB
http_port 172.21.118.119:3128
icp_port 0
dns_nameservers x.x.x.x
cache_effective_user nobody
cache_effective_group nobody
cache_mem 16 MB
cache_swap_low 95
cache_swap_high 98
maximum_object_size 8192 KB
memory_pools off
acl all src 0.0.0.0/0.0.0.0
cache_dir ufs /cache 8857 16 256
http_access allow all
icap_enable on
icap_preview_enable off
icap_preview_size 400
icap_check_interval 10
icap_send_auth_user on
icap_auth_scheme %u
icap_send_client_ip on
icap_service test respmod_precache 0 icap://localhost:1344
icap_class filter_class test
icap_access filter_class allow all
cache_log /blocker/logs/cache.log
cache_store_log none
cache_access_log none
logfile_rotate 7
shutdown_lifetime 10 seconds
half_closed_clients off
cache_mgr webmaster
client_db off
ie_refresh on

Has anyone seen this problem before? Is a fix available? I'm thinking it
might be some sort of race condition in the squid Icap Client?

Thanks in advance for any help anyone can give!

Graeme
Received on Fri Oct 15 2004 - 03:28:44 MDT

This archive was generated by hypermail pre-2.1.9 : Mon Nov 01 2004 - 12:00:02 MST