6.3. Configuration mod_jk as Loadbalancer

This section describes Apache with mod_jk as a solution for load balancing in an OLAT cluster. You are free to use any other load balancer (hard- or software) as long as certain criterias are met (see below).

Refer to http://tomcat.apache.org/connectors-doc/generic_howto/workers.html for details

Criterias for the Load Balancer

Session stickyness: The OLAT Cluster requires session stickyness to be configured to make sure a particular user is always routed to the same tomcat instance.
Apache and one tomcat instance must run on same host: Static files (those under /olat/raw) can be served by Apache directly and do not need to take a detour via tomcat. Hence apache needs to be configured to serve /olat/raw directly from a tomcat instance on the same host. In order to do this you must install an OLAT node at least once on the host where you run apache - it does not necessary have to be included in the load balancing but it must be started once in order to populate the /olat/raw/* directories.

Example configuration of Apache with mod_jk

mod_jk.conf:

  <IfModule mod_jk.c>
    JkRequestLogFormat "%w %V %T %r"
    JkWorkersFile      "/modjkdirectory/workers.properties"
    JkLogFile          "/logdirectory/mod_jk.log"
    JkShmFile          "/modjkdirectory/mod_jk-runtime-status"
    JkLogLevel         error 
    JkOptions          +ForwardURICompatUnparsed
    JkMount            /olat/* loadbalancerworker
    JkMount            /olat loadbalancerworker
    JkUnMount          /olat/raw/* loadbalancerworker

   ...
   </IfModule>

Make sure to JkMount the loadbalancerworker under /olat
Make sure to JkUnMount the raw directory /olat/raw

workers.properties:

workers.tomcat_home=/usr/local/opt/tomcat-6.0
workers.java_home=/usr/local/opt/java-1.6
ps=/

worker.list=loadbalancerworker

worker.loadbalancerworker.type=lb
worker.loadbalancerworker.balance_workers=tomcatolat02,tomcatolat01
worker.loadbalancerworker.sticky_session=true
worker.loadbalancerworker.sticky_session_force=false
worker.loadbalancerworker.method=s
worker.loadbalancerworker.connect_timeout=15000
worker.loadbalancerworker.prepost_timeout=15000

worker.basic.type=ajp13
worker.basic.socket_timeout=1200
worker.basic.cachesize=1
worker.basic.cache_timeout=1300
worker.basic.recycle_timeout=1300
worker.basic.port=8009
worker.basic.distance=10
worker.basic.lbfactor=10
worker.basic.connect_timeout=30000
worker.basic.prepost_timeout=30000

worker.tomcatolat01.host=localhost
worker.tomcatolat01.port=8019
worker.tomcatolat01.domain=tomcatolat01
worker.tomcatolat01.reference=worker.basic
worker.tomcatolat01.lbfactor=10

worker.tomcatolat02.host=localhost
worker.tomcatolat02.port=8029
worker.tomcatolat02.domain=tomcatolat02
worker.tomcatolat02.reference=worker.basic
worker.tomcatolat02.lbfactor=10

Add the loadbalancerworker under worker.list and all OLAT nodes under worker.loadbalancerworker.balance_workers
Configuring connect_timeout and prepost_timeout instructs mod_jk to detect a deadlock or crash in an OLAT node