Don't think this is possible, but wanted to check. There is a
performance hit when we remove a machine from the cluster due to shard
reallocation and cold shards coming online. I currently have shard
allocations throttled, with concurrent_recoveries: 1 and
Our response times in a steady state are 25-30ms with maxes at 1-2
seconds. During shard reallocation, this increases by 10x, which isn't
bad, but my clients have gotten spoiled.
What I'd like to do is send a command to the cluster to tell it not to
re-allocate any shards for some amount of time while a server is taken
offline for maint, since I know that any re-allocations will pretty
much be wasted work. Is this doable in 16.2 and if not would others
consider this a viable feature?
This would mitigate the performance hit when the machine is removed,
but I'd still expect some hit when it is added back.