Improving Elasticsearch Query Performance for Billion‑Scale Datasets

When interviewing, a common question is how to improve Elasticsearch query efficiency on billions of records.

Elasticsearch performance is limited; large data sets can cause 5‑10 seconds latency on first queries, which improves after caching.

The key optimization is to maximize the OS filesystem cache: allocate enough memory so that index segment files fit into cache, aiming for cache size at least half of total data.

Store only searchable fields in Elasticsearch and keep the rest in a secondary store such as HBase; this reduces index size and improves cache hit rate.

Implement data warm‑up by periodically querying hot data so it stays in the filesystem cache, and separate hot and cold data into different indices to prevent cold data from evicting hot data.

Design documents to avoid complex joins; perform joins in the application layer before indexing.

Deep pagination is costly; avoid it by limiting page depth, using the Scroll API or search_after for efficient forward‑only pagination.

These practices together can reduce query times from several seconds to tens of milliseconds even with terabytes of data.