Storage behavior

The Optimized engine stores data in two structures, each optimized for its purpose.

Structure	Format	Purpose
Analytics storage	Apache Parquet (columnar, natively compressed)	Aggregations, joins, range scans, and trend analysis
Search index	Lucene inverted index	Phrase, fuzzy, and wildcard matching

Parquet replaces three of Lucene's on-disk structures (doc values, BKD trees, and stored fields) with a single compressed columnar format. The inverted index is retained for full-text search. This architecture significantly reduces storage requirements for typical observability workloads with search enabled.

Storage tiers

The Optimized engine uses the Multi-tier storage for Amazon OpenSearch Service architecture, which combines Amazon S3 with local instance storage and is powered by OpenSearch Optimized Instances.

Tier	Purpose	Instances
Hot	Active ingestion and frequent queries	OpenSearch Optimized Instances (OR1, OR2, or OM2)
Warm	Infrequent queries and extended retention (read-only)	OI2 instances with managed storage

Amazon S3 backs both Parquet files and inverted-index segments for durability. Use Index State Management (ISM) policies to transition data from hot to warm based on age, shard size, or other conditions.

Important

On the Optimized engine, the warm tier is read-only. Although the general multi-tier storage architecture supports writes to the warm tier, Optimized domains do not support write operations in the warm tier. Ingest data into the hot tier and use Index State Management (ISM) policies to transition it to warm.