Implement multi-tenant search with Amazon OpenSearch Serverless next generation

This article explains how to implement cost-effective multi-tenant search using Amazon OpenSearch Serverless next-generation architecture with collection-per-tenant model and per-account regional endpoints.

• Scale-to-zero compute reduces costs by charging only during active indexing or searching, not idle periods
• Per-account regional endpoint (account-id.aoss.region.on.aws) routes all collections through single hostname using headers
• Collection-per-tenant model provides full data and workload isolation between tenants with independent scaling
• Single OpenSearch client manages all collections using x-amz-aoss-collection-name header for routing
• Hash tenant IDs to collection names for privacy, staying within 64-character limits, and supporting schema migrations
• Cold start latency (~10 seconds) occurs when collection group scales to zero; use warmup queries for latency-sensitive tenants

The next-generation architecture makes collection-per-tenant practical by combining scale-to-zero economics with operational simplicity through unified endpoint management.