High Availability Considerations

MM can be deployed under High Availability (HA) clusters based on a configuration of Active/Passive MM servers. In this HA architecture, all user URL requests are connected to a centralized HA Management server (machine / node 1) which redirects all calls to the Active MM server (machine / node 2) , or the Passive MM server (machine / node 3) if the Active MM server (or the machine itself ) is down. In such switch-over case, the delay is limited to the time it takes to startup the Passive MM server which is usually a couple of minutes assuming the machine was already active.

In the above HA architecture, both the Active and Passive MM servers must be sharing the same data on a shared database server, and on a shared file system server. For more details, see Understanding the Data Locations. In order to achieve an overall High Availability (HA) of the total MM solution, these shared database and file system servers must also have their own HA deployments, therefore involving a lot more machines / nodes for each of these database and file system servers. Starting a MM server starts both the Tomcat web application server and the Lucene search engine, therefore file system sharing is very important for the user experience as the response time is heavily driven by the License search engine storing its indexes on that shared file system in $MM_HOME/data).

Note that the HA deployment cannot be achieved with the bundled PostgreSQL database $MM_HOME/postgresql available with the MM entry level edition on Windows. In fact, HA deployment is only available on the MM most advanced edition as the license stored in the shared database must be enabled to support both the Active and Passive servers (obviously on different host ids).