Mjolnir is currently configured to get the list of indices from localhost:9200, with multi-instances per node in place this will no longer work.
Adding more ports to check won't work neither since not all clusters are running as a dedicated instance on every node.
Using LVS+port endpoints is not an option as well since the LVS hostname resolves to localhost.

1. We could add another set of configuration parameters using puppet by picking a list of "seed" nodes per cluster
2. We could use the cluster state available from localhost:9200 because it should have a list of seed hosts configured as a cross cluster search pointing to the two small clusters.

Options 2 seems easier to implement and seems to help to reduce the moving parts in the multi-instance setup.