We think that some of the slowness with the master server in elasticsearch is due to the number of indices and the number of shards in the cluster. We should write up some script to parse through elasticsearch's /_cat/indices API and determine which indexes could have fewer total shards assigned to them. In some of our documentation nik suggested that ~2GB is a good size for shards. We should also re-evaluate this number, we don't know that it's wrong, but we also don't know that it's right. There are tradeoff's in both directions.
This fits in with our Q3 goal of evaluating the current elasticsearch configuration and optimizing it as appropriate.