We typically recommend daily polling and weekly collection. If polling detects a change in either the config or the inventory, then a collection is forced. If not, you do not incur the overhead of a full collection.
As an alternative to polling, you could rely solely on syslog messages from devices. By default, RME is configured to listen for config change and inventory change syslog messages. In a stable network, these plus weekly collection may be all you need.
DCR master/slave only replicates DCR data. The RME data is NOT replicated. Typically for these kind of "HA" setups, we have recommended that customers turn down periodic polling on the secondary server. The secondary server should receive syslog messages, but the inventory and config change automated actions (previously mentioned) should be disabled. Discovery should NOT run on the secondary server unless there are devices which only it can reach. In that case, strict Discovery filters should be put in place to limit the scope.
If a disaster occurs, it is up to you to turn up polling, discovery, and the automated actions on the secondary server.
