There are some significant scaling issues with ActiveMQ 5.8 and MCollective, especially around SSL connections. I’ve been working with some clients and solved many of these problems by using ActiveMQ 5.9.1 as the middleware.
I’ve created an RPM for Active 5.9.1 on RHEL/CentOS that matches the one published on the Puppet Labs EL6 dependencies repository, with the following changes:
1. A revised activemq.xml file which mirrors changes in the original Puppet Labs-provided activemq file.
2. Extracts the wrapper config, init script, apache config, and info provider files from the previously used wlcg.tgz file and includes them separately, discarding the obsolete changes in that tgz file.
3. Uses a patch against the Apache-provided log4j.properties to enable the changes previously provided in wlcg.tgz.
4. Removes the checkUser query on ‘status’ which causes the bug reported in Puppet Labs bug CPR-17.
*** Changes in behavior:
As the activemq config file supplied in the community repository is specific to MCollective, the following changes were made to bring it in line with the recommendations specified in the Puppet Labs documentation: http://docs.puppetlabs.com/mcollective/deploy/middleware/activemq.html
5. Disables flow control as recommended in the configuration documentation.
6. Disables UseDedicatedTaskRunner in the wrapper.conf as recommended in the configuration documentation.
7. Re-enables the default admin logins for the jetty configuration, as the previous password-less logins made the Jetty console available to anyone.