[squid-users] Going into hit-only-mode for 5 minutes

[Baird, Josh]

Hi,

We recently started having problems where our Squid 2.6 (squid-2.6.STABLE21-6.el5) proxy servers would stop serving requests.  In my cache.log, I see many of these:

2015/04/14 01:13:45| Failure Ratio at 26.15
2015/04/14 01:13:45| Going into hit-only-mode for 5 minutes...
2015/04/14 01:18:46| Failure Ratio at 3.55
2015/04/14 01:18:46| Going into hit-only-mode for 5 minutes...
2015/04/14 01:23:46| Failure Ratio at 1.02
2015/04/14 01:23:46| Going into hit-only-mode for 5 minutes...
...
2015/04/14 06:50:58| idnsSendQuery: Can't send query, no DNS socket!
2015/04/14 06:50:58| idnsSendQuery: Can't send query, no DNS socket!
2015/04/14 06:50:58| idnsSendQuery: Can't send query, no DNS socket!
2015/04/14 06:50:58| idnsSendQuery: Can't send query, no DNS socket!

I suspect this is the problem - the proxy is running out of DNS sockets.   I have already determined that there are not problems with the DNS servers that these proxies are using (in their /etc/resolv.conf).  Could this be caused by a bad user chewing up DNS sockets/children with invalid URL requests?

The "going into hit-only-mode" errors appear to be ICP related?  In this case, I believe we have ICP completely disabled:

# icp_access allow allowed_src_hosts
# icp_access deny all_src

Could anyone offer any suggestions or advice to help figure out what is causing these problems?

Thanks,

Josh