Search clusters need special consideration:
Sharding: Split index across nodes. Each shard handles subset of documents. Query hits all shards, results merged.
Replication: Each shard has replicas. Replicas serve reads, increasing throughput.
Routing: Hash document ID to determine shard. Ensures same document always goes to same shard.
Reindexing: Index changes require rebuilding. Blue-green deployment: build new index, switch alias when ready.
Elasticsearch defaults: shard per index since version (previously ), replica per shard. Tune shard count based on data size and query volume.