[imp] Re: multiple IMP front ends?

Sheeri Kritzer sheeri.kritzer@tufts.edu
Mon Nov 4 16:33:46 2002


We have a similar implementation (3 load balanced servers behind a sticky load 
balancer) but our problem is this:

we have a mysql preferences database on a separate machine.  Mysql has no high 
availability options (it has redundancy, but it needs manual intervention).  
Therefore, your single point of failure is now from IMP/Horde/Apache server to 
your mysql server.  as discussed recently, if the preference database fails, 
IMP fails.

-Sheeri Kritzer
Systems Administrator
University Systems Group
Tufts University
617-627-3925
sheeri.kritzer@tufts.edu


Quoting Steven Premeau <premeau@uwp.edu>:

> >
> >
> >Has anyone implemented IMP on multiple front ends to distribute load
> >> (and also for redundancy). We already have it running on one machine
> >> pointing it to IMAP mail folders on another machine.
> >
> >
> >I know of several people doing it, and I'm sure there are many more.
> >
> 
>     The University of Wisconsin - Parkside is running IMP load-balanced 
> across two "front-end" machines.
> 
> >HEAD has a session handling API built in so that you can share a sessions
> db
> >between machines, but you don't need much more, and you can do that with
> >RELENG_* yourself quite easily.
> >
> We use the LBNamed module to balance usage between the nodes.
> 
> Setup wasn't that difficult -- here's the tips and traps we noticed.
> 
> Our prefs database is stored in a MySQL database on our "big" IMAP 
> server.  By configuring Horde/IMP to connect via IP to the database, all 
> preferences are shared.
> 
>     We have an SSL certificate issued to the load balancing DNS name 
> (best.email.uwp.edu a/k/a RangerMail.uwp.edu) the web server on both 
> node refers to itself as best.email.uwp.edu.
> 
>     We changed the session cookie domain from it's default to 
> ".email.uwp.edu" -- even though both systems have the same name, it just 
> made sessions much more stable.
> 
>     One gotcha that we haven't solved.  If a user attaches a file to a 
> message composition and then "dawdles" before sending the message (by 
> editing the composition or just delaying) we have had the system change 
> "front-end" hosts during that delay -- because the attached files are 
> placed in /tmp during composition, they are not shared between machines 
> and you get lots of errors when the system goes looking for them.  
> 
>     We have worked around that by educating our users to do the 
> attachment part last -- if they attach and then send, it hasn't every 
> changed hosts on us.
> 
> If you would like more information on the setup at UW-Parkside, please 
> let me know.
> 
> Steve.
> 
> 
> 
> -- 
> IMP mailing list
> Frequently Asked Questions: http://horde.org/faq/
> To unsubscribe, mail: imp-unsubscribe@lists.horde.org
> 



More information about the imp mailing list