Adaptive Replica Selection in OpenSearch

Adaptive replica selection is a mechanism designed to improve query response times and alleviate strain on overloaded OpenSearch nodes. It ensures that nodes experiencing delays due to issues like hardware, network, or configuration problems do not slow down the overall query process.

How It Works

Consider a scenario where one node in the cluster is underperforming. This might result from network congestion, hardware malfunctions, or misconfigurations, causing the response times for shards on that node to be significantly slower than those on other nodes.
When an OpenSearch cluster processes a query, it collects responses from shards across all relevant indices. Normally, OpenSearch uses a “round-robin” method to distribute shard requests across available nodes, including the struggling one. However, this approach can prolong query times when a distressed node is involved.
With adaptive replica selection enabled, OpenSearch prioritizes nodes with better response times. It avoids sending shard requests to the struggling node unless no other replicas are available. This reduces the load on problematic nodes, improving overall cluster efficiency and query performance.

Enabling Adaptive Replica Selection

PUT /_cluster/settings
{
  "transient": {
    "cluster.routing.use_adaptive_replica_selection": true
  }
}

By configuring this setting, you ensure that the cluster dynamically selects replicas based on their responsiveness, resulting in faster query execution and a more balanced system load.