I'm using ElasticSearch 1.7.4 and was restarting a node (for upgrading from ES 1.4.4).
I constantly index bulks against one big index, and was monitoring shard statuses during the initialization phase of the restarted node.
I noticed that if one replica shard of my big index was being initialized or relocated, the indexing continued peacefully.
However, if one primary shard of the index was being relocated, the indexing could not continue and would timeout. (I don't have the exact error at hand.)
Is there any setting that may help with this?
Does this also happen in ES 2.x?
If relocating a primary shard causes such a problem, why not promoting one replica as primary, move the desired shard, then promote the moved shard back to primary.
Or create a new temporary replica at the desired node then promote it primary and remove the origin shard.
I understand there are synchronization issues lying behind, but promoting a replica to primary is still more affordable that blocking writes and waiting for a big (9 GB in my case) copy to finish.
Thanks for your insights!