Clarification about OpenSearch Serverless Search Replicas

Hello,

When an index gets created. It gets created with x number of shards. There are multiple ways to support high TPS (query type, data size) for an index with x number of shards.

1. Span out replicas on more nodes: Since AOSS keeps 2 replica. there will be 2x number of replicas for an index. If the search OCU you configure it more than 2x, the number of workers can scale out up to 2x. And all workers can host 1 replica to support high search TPS. This approach can't do anything beyond 2x workers as it can not spread out 2x replicas on more than 2x workers.

Example: A search collection Index uses 8 shards. Since AOSS keeps 2 replica. there will be 16 number of replicas for an index. If the search OCU customer has configured to 16, the number of workers can scale out up to 16. And all workers can host 1 replica to support high search TPS. This approach can't do anything beyond 16 workers as it can not spread out 16 replicas on more than 16 workers.

2. Add more replicas: Beyond 2x worker/OCU is where adding more replica would help. This is not yet supported. But the team is working on this and will be supported soon.

If you have any further queries or need any assistance with your workload. Feel free to reachout to us through a support case as currently replicas does not scale and the team can help you with your requirements accordingly.