Thanks everyone for the advice.  I'm working on installing and testing 2.6 
STABLE16.  I'll see where we're at then, but I suspect things will be looking 
better.

Thanks.

Matt Ruzicka
Sr. Systems Engineer
[EMAIL PROTECTED]
www.cisp.com
www.yocolo.com

419.724.5345 : tel
419.867.6913 : fax

-----Original Message-----
From: Chris Robertson [mailto:[EMAIL PROTECTED]
Sent: Thursday, November 01, 2007 3:46 PM
To: [email protected]
Subject: Re: [squid-users] squid_radius_auth

Matt Ruzicka wrote:
> We're in process of rebuilding a couple web filter boxes on Centos 4.5 
> running Squid 2.5.STABLE14 (latest from yum) using squid_radius_auth 1.09 for 
> authentication with the following config:
>
> auth_param basic program /usr/local/squid/libexec/squid_radius_auth -f 
> /usr/local/squid/etc/squid_radius_auth.conf
> auth_param basic children 30
> auth_param basic realm Filtered Web Service
> auth_param basic credentialsttl 4 hours
> auth_param basic casesensitive off
>
> We're seeing an odd issue where radius authentication will start failing 
> sometime after the machine has been in production for a variable amount of 
> time.  The issue appears to arise only after at least 6 hours, but sometimes 
> as long as 10 or 11 hours.  If the machine is not in production and is only 
> receiving test authentications the issues does not appear to arise.
>
> The failures show up as a long lag after correct credentials are issued and 
> an eventual re-request for credentials.  When this happens the squid access 
> logs show denies for web traffic from these IPs passing account names that 
> had previously authenticated.  This lag is the same behavior we see if the 
> radius server is unreachable, but I can log into the machine and manually run 
> squid_radius_auth from the command line and authenticate without issue while 
> the problem is occurring.  During these failures we do no see the 
> authentication requests hitting our Radius servers.
>
> However, if I issue a reconfig the problem goes away for another 6+ hours or 
> so.
>
> I feels like the child processes are wedging somehow, but I'm not sure how or 
> why.
>
> Additionally the old filter servers are running older versions of CentOS, 
> Squid and v106 of squid_radius_auth and they are not seeing the issue.
>
> * Has anyone else seen similar behavior?
>

Yes.  See
http://www.squid-cache.org/mail-archive/squid-users/200605/0494.html
Granted, this issue was appearing with Squid-2.5-Stable13 and
squid_radius_auth 1.08.

As stated, updating to Squid 2.6 is recommended.  You can compile the
Fedora SRPM, or the CentOS5 SRPM (which is based on Squid-2.6Stable6) or
grab the source, use squid -V on your current install and use that as a
guide for compiling.  The CentOSPlus repository doesn't seem to have an
updated RPM for Squid.

> * Is there any additional logging or debugging I can run to hopefully see 
> what is happening?
>

>From http://www.squid-cache.org/mail-archive/squid-users/200501/0554.html:

debug_options ALL,1 29,9 84,9

then see cache.log for details on the auth progress.

Be warned that your logs will contain usernames+passwords in plain text
when doing this.

> For now we have put in place an hourly cron to issue the reconfig, but this 
> is a pretty cludgy work around.
>
> Thank you in advance.
>
> Matt Ruzicka
> Sr. Systems Engineer
> [EMAIL PROTECTED]
> www.cisp.com
> www.yocolo.com
>
> 419.724.5300 : tel
> 419.867.6913 : fax
>

Chris

Reply via email to