Thanks Amos, I was looking at the 3.0 page for cache_peer definition since I am running 3.0 STABLE14, so I never saw those monitor options. I am not running anything that requires the 3.0 branch so I could switch to 2.7 to solve this problem. I would like to know if there are plans to include these options under the 3.x branches in the future? As I would prefer my configuration doesn't depend on an option that will not be available in the foreseeable future.
Thanks,
Dean Weimer
Network Administrator
Orscheln Management Co
-----Original Message-----
From: Amos Jeffries [mailto:squid3_at_treenet.co.nz]
Sent: Thursday, April 30, 2009 11:39 PM
To: Dean Weimer
Cc: crobertson_at_gci.net; squid-users_at_squid-cache.org
Subject: Re: [squid-users] Parent Proxy's, Not Failing over when Primary Parent is Down.
Dean Weimer wrote:
> -----Original Message-----
> From: crobertson_at_gci.net [mailto:crobertson_at_gci.net]
> Sent: Thursday, April 30, 2009 2:13 PM
> To: squid-users_at_squid-cache.org
> Subject: Re: [squid-users] Parent Proxy's, Not Failing over when Primary
> Parent is Down.
>
> Dean Weimer wrote:
>> I have a current Parent child proxy configuration I have been testing,
> its working with the exception of some sites not failing over to second
> parent when primary parent goes down.
>> In the test scenario I have 2 parent proxies, and one child proxy
> server, the parents are each configured twice using an alias IP address.
> This is done to load balance using round robin for the majority of web
> traffic yet allow some sites that we have identified to not work
> correctly with load balancing to go out a single parent proxy.
>>
>
> Since Squid 2.6 there has been a parent selection method called
> "sourcehash", which will keep a client-to-parent-proxy relationship
> until the parent fails.
>
> I considered this, but was concerned that after a failed proxy server,
> the majority of my load would be on one server, and not taking advantage
> of both links when the problem is resolved.
>
>> The load balanced traffic works as expected, the dead parent is
> identified and ignored until it comes back online. The traffic that
> cannot be load balanced is all using HTTPS (not sure HTTPS has anything
> to do with the problem or not), when I stop the parent proxy 10.50.20.7
> (aka 10.52.20.7) the round-robin configuration is promptly marked as
> dead. However a website that has already been connected to that is in
> the NONBAL acl just returns the proxy error from the child giving a
> connect to (10.52.20.7) parent failed connection denied.
>
> Hmmm... You might have to disable server_persistent_connections, or
> lower the value of persistent_request_timeout to have a better response
> rate to a parent failure with your current setup.
>
> Also considered this, but figured it would break some sites that are
> working successfully with load balancing because they create a
> persistent connection, and making the request timeout to low would
> becoming annoying to the users. Also as the default is listed at 2
> minutes, I noticed that even after as much as 5 minutes that the
> connection would not fail over.
>
>> It will not mark the non load balanced parent dead, closing and
> restarting the browser doesn't help. It will change the status to dead
> however if I connect to another site in the NONBAL acl. Going back to
> the first site, I can then connect, even though I have to log in again,
> which is expected and why these sites cannot be load balanced.
>> Does anyone have any ideas short of writing some sort of test script
> that will cause the parent to be marked as dead, if it fails without any
> user intervention.
>> Here is the cache peer configuration from the child proxy. FYI, I
> added the 5 sec timeout to see if it had any effect, and it didn't with
> the exception of speeding up the detection of the dead load balanced
> proxy.
>> ## Define Parent Caches
>> # Cache Peer Timeout
>> peer_connect_timeout 5 seconds
>> # Round Robin Caches
>> cache_peer 10.50.20.7 parent 8080 8181 name=DSL2BAL round-robin
>> cache_peer 10.50.20.6 parent 8080 8181 name=DSL1BAL round-robin
>> # Non Load Balanced caches
>> cache_peer 10.52.20.7 parent 8080 8181 name=DSL2
>> cache_peer 10.52.20.6 parent 8080 8181 name=DSL1
>>
>> ## Define Parent Cache Access rules
>> # Access Control Lists
>> acl NONBAL dstdomain "/usr/local/squid/etc/nonbal.dns.list
>> # Rules for the Control Lists
>> cache_peer_access DSL2BAL allow !NONBAL
>> cache_peer_access DSL1BAL allow !NONBAL
>> cache_peer_access DSL2 allow NONBAL
>> cache_peer_access DSL1 allow NONBAL
>>
>> Thanks,
>> Dean Weimer
>> Network Administrator
>> Orscheln Management Co
>
> Chris
>
> I am currently doing some testing by creating access control lists for a
> couple nonexistent sub domains on our own domain. This then just
> accesses the error page from the parent proxy for nonexistent domain, so
> it shouldn't put an unnecessary load on the internet links testing.
> Then allowing each one through one of the non balanced parents. By
> accessing that page with my browser it causes the parent to be marked
> dead.
>
> I could look at writing a script to access these pages through the child
> proxy every so many seconds to cause the parent to be marked as dead.
> It's kind of a hacked solution, but hopefully it would keep the users
> from having too much down time in the event that one proxy goes down.
Since 2.6 there have also been a set of monitor* options to do this in
various ways when ICP feedback is insufficient or not available.
>
> It would probably be preferable though to query ICP directly and then do
> a reconfigure on the child squid to exclude that parent from its
> configuration. If anyone can tell me where to find the information on
> how to do an ICP query that would save me some time, and be greatly
> appreciated, in the mean time I will start searching or worse yet if
> that fails sniffing network traffic to write an application to mimic the
> squid query.
http://www.squid-cache.org/Versions/v2/2.7/cfgman/cache_peer.html
http://www.squid-cache.org/Doc/config/icp_port/
http://www.squid-cache.org/Doc/config/icp_access/
You may find other useful info in there for questions you have not yet
asked as well.
Amos
-- Please be using Current Stable Squid 2.7.STABLE6 or 3.0.STABLE14 Current Beta Squid 3.1.0.7Received on Fri May 01 2009 - 12:26:46 MDT
This archive was generated by hypermail 2.2.0 : Sat May 02 2009 - 12:00:01 MDT