3 Node Cluster - how many shards per index

Is my understanding right that if i have a cluster of three identical data-nodes an index should have three shards and one replica for best search performance (parrallel processing on each node) ?

And then when doing horizontal scaling to nine data-nodes. Do i get best performance by adding one additional replica (three shards and two replica) ?

It depends a lot:

  • data volume
  • usage read VS write
  • hadoop connector ?
  • search vs aggregations
  • how many search by seconds

There is no magic rule, the good approach is to start simple, monitor every thing (CPU, RAM, Java heap...) and find the good point for your usage.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.