Amazon OpenSearch Service 101: T-shirt size your domain for e-commerce search

This article provides a comprehensive guide to sizing Amazon OpenSearch Service domains for e-commerce search workloads using T-shirt sizing methodology.

• T-shirt sizing (XS-XL) simplifies capacity planning based on catalog size and query concurrency
• OpenSearch uses scatter-gather mechanism to distribute search queries across primary and replica shards
• Storage requirement formula: Raw data × (1 + replicas) × 1.45 accounts for overhead
• Shard sizes should maintain 10-30GB range for optimal read-heavy e-commerce performance
• Instance types: General-purpose (M8g) for balanced workloads, compute-optimized (C8g) for high QPS, memory-optimized (R8g) for large catalogs
• T-shirt sizing table maps product counts (500K to 20M+) to recommended node counts and instance types
• Vertical scaling upgrades instance types; horizontal scaling adds data nodes for traffic growth
• Monitor CPU under 80%, JVM memory below 75%, P95 query latency, and thread pool rejections

The guide enables teams to right-size OpenSearch clusters efficiently, starting small and iterating as business scales through load testing and CloudWatch monitoring.