Throwing some stones around here.
We have an elastisearch in place, with about 120 indexes (two per customer)
that need to be updated with a daily feed. This means reindexing for every
index at least once a day, which is costly. I know there are other means of
optimizing this, such as bulk indexing.
Anyway, in some sharded/distributed systems, one classical optimization
step is -given you have a "master" and a "slave" database server, by
instance- is to use the "master" server for reading/writing (mostly through
a supposed admin interface), and to use the "slave" server for read access
only.
I was wondering if is there any way of achieving this with the
elasticsearch replicas, maybe setting a few as "read-only" and then
redirecting all search request to these??? Looks like this is not likely to
be possible, but it was worth the shot. 
--