[horde] clustering horde with memcached

Kevin Konowalec webadmin at ualberta.ca
Tue Jan 2 14:13:13 PST 2007


Oh yeah... I suppose I should have mentioned.... I'm using Horde  
3.1.2.  If my docs are correct we're running memcached 2.0.4.

So, then, I should be able to run multiple memcache servers and it'll  
just figure it out itself?  Is there any cleanup of sessions or  
anything that I should be doing manually (from my inexperienced-with- 
memcached perspective it looks like the memory allocated to memcached  
just fills up and then it knocks old sessions off when new ones are  
added... but the memory usage never goes lower than the max allocated.

Kevin



On Jan 2, 2007, at 2:30 PM, Liam Hoekenga wrote:

> It depends on the version of Horde and the version of the PECL  
> memcached extension you're using.
>
> In Horde <= 3.1.1, the memcache sessionhandler has a function,  
> _getConForId(), that uses the sessionID and the list of memcache  
> servers to figure out which memcache server has the session data.   
> This was done because prior to PECL memcache 2.x, the memcache  
> extension didn't support connection pooling.   Since this is a  
> strictly PHP based solution, it should work with any release of the  
> PECL memcache extension.
>
> PECL memcache 2.x >= 2.x supports connection pooling, and with the  
> release of Horde >= 3.1.2, _getConForId() is dropped in favor of  
> using the connection pooling (and connection hashing) internal to  
> the memcache extension.
>
> I'm gonna guess that connection pooling and PECL memcache >= 2.x  
> works ok w/ Horde <= 3.1.1 (since the pooling is implemented PHP),   
> but to make it work with Horde >= 3.1.2, you'll need to use  
> memcached >= 2.0.
>
> Liam
>
>
> Kevin Konowalec wrote:
>> We're running a cluster of 12 machines for our 75,000+ user base.   
>> They're all set up in a round-robin and users can conceivably hop  
>> from one machine to another in the course of a session (due to  a  
>> number of factors like the server they're on going down or  
>> whatever).  We've only got one central server running memcached  
>> for all of them though.  Is there a better way to do it?  If I  
>> tell horde to use 1..N memcached servers will it then scan all the  
>> servers until it finds the session when a user flops from one  
>> server to the next?
>> --Horde mailing list - Join the hunt: http://horde.org/bounties/ 
>> #horde
>> Frequently Asked Questions: http://horde.org/faq/
>> To unsubscribe, mail: horde-unsubscribe at lists.horde.org
>



More information about the horde mailing list