I am doing tests with cluster of 5 elasticsearch nodes and I see big difference in performance depending on how elasticsearch cluster is configured. When the index is configured with 8 shards and no replicas, I see it can process up to 60 000 msg per second. When I reconfigure that to 8 shards and 1 replica, the performace drops mnore than half of that. (15-20 000). But the cluster with replicas is much more stable, it runs even vhen node is down. This is not the case with 0 replicas, when 1 elasticsearch node is stoppped, it switches cluster ot red and it stops processing new messages.
Question- Is there any other config or other way, that I could keep no replicas but for the cluster to be a bit more stable in case of stop of one node. Some dynamic attribute to tell ignore those errors, just continue.. _-) ? Thanks.