Handling lost replicas in preemptible instances

I have a set up which has two instance groups on GCP: One is just a normal instance group, and the other is preemptible. The preemptible instance group is autoscaled based on QPS.

At this moment, the setup is that of an 'anti-cluster' each ElasticSearch instance is responsible for it's own data, and is deployed with a sidecar proxy and load balancer for all of the instances.

We have 100k documents, that are ingested every time an instances starts, and periodically every hour. This means there is a lot of processing going on to index the data and some overhead for that

This is why we're considering joining all of the nodes into it's own cluster so ElasticSearch will handle replicating the data in hopes it will have less overhead than indexing it at start time.

Few questions arise:

  1. How to configure number of replicas/ shards for dynamic sized cluster?
  2. What if all of the replicas are on the preemptible instances and all of them are lost?

Thanks for suggestions

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