I'm using ES 1.7.3 with 4 client nodes, 16 data nodes and 5 masters. We have detected a bug in our java cliente application this bug apply re-map for each index adding a mapping massivelly.
The cluster pending-task increase more quickly and the amount of the young heap space is used and finally the master do gabagge collection with failures.
The only way to stop it has been closing the indexes affected or dropping the http requests to these indexes ( blocking the URIs /index_pattern)
- Maybe it will be useful some feature for close some index if the pending task for this index grow up more quickly.
- Another useful suggestion will be manage the amount of memory in masters to manage the pending tasks for each index
Obviouslly these are suggestions due a defect of our client, but I think it could be useful for system administrators.