Throttling of recovery (both gateway recovery and peer node recovery)

[kimchy]

There are two recoveries going on with elasticsearch. The first is when the first even primary is created, and is then recovery from the gateway, and the second is when a shard gets created / relocated into a another node.

This can be a taxing operation on the machine, especially when several shards are recovering at the same time. For this, we have two new settings that allow to throttle it:

indices.recovery.throttler.concurrent_recoveries: The number of concurrent recoveries allowed at the same time on a node. Defaults to the number of cores.

indices.recovery.throttler.concurrent_streams: The number of concurrent streams transfers allowed during recovery. A stream basically maps to an Lucene index file, which is a file that exists within specific shard index. Defaults to the number of cores.

[kimchy]

Throttling of recovery (both gateway recovery and peer node recovery), closed by 216dda3.