Replica shards are not involving while searching

[chanakagl (chanaka liyanarachchi)]

Hi,

I have 3 node cluster with time series indices(per day). I have enabled allocation awareness and all nodes in AWS. Node2 is having Index1 all replica shards on it while primaries are assigned to other two nodes. when I running the search queries, Node2(Where replicas in) not involving at all. Is replica shards not using at searching time? please let me know is there any setting needs to be tweaked?

[chanakagl (chanaka liyanarachchi)]

@here could anyone pls help me on this?

[dadoonet (David Pilato)]

Read this and specifically the "Also be patient" part.

It's fine to answer on your own thread after 2 or 3 days (not including weekends) if you don't have an answer.

[dadoonet (David Pilato)]

Replicas by default are also used to serve search requests. No specific settings required.

[chanakagl (chanaka liyanarachchi)]

@dadoonet thanks for the reply.

Now I can see replica is serving for the search requests slightly. I'm using 6.2.4 version. I'm having 3 Packetbeat indices and each one has 5 shards and 1 replica and each shard is 5gb of size.
Below are the CPU stats when executing search quires, I

Node 1 (Primary shard 1,3)

Node2 (All Replicas 0 - 4)

Node3 (Primary shard 0, 2, 4)

Is there any setting i can tweak to optimize the searches? I'am using some data histogram aggregation queries to get metrics of the Prod cluster.

[dadoonet (David Pilato)]

Few notes:

• You probably need only one shard instead of 5 per index.
• Which node is used by Kibana or your application to hold the search request? I mean that you could may be use a coordinating node if you wish in front of your cluster. May be that will remove some load on one of your data nodes.
• If really needed you can also use the preference parameter. See https://www.elastic.co/guide/en/elasticsearch/reference/current/search-request-body.html#request-body-search-preference

[chanakagl (chanaka liyanarachchi)]

@dadoonet thank you very much i'll try those also.