The Composer REST server uses data sources to store user session data. This doesn't mean that it has to have an explicit data source configured; if no data source is configured, the Composer REST server uses an out-of-the box memory connector.
When using multiple instances of the Composer REST server for high availability or load balancing, the instances do not share memory, so a data source is required. Any data source that has a loopback connector available can be used. In our experience, MongoDB, Cloudant, and Redis work out of the box; we just need to install the connector and configure the environment variables by following the steps provided by the Hyperledger.