Collection types
When you create an Amazon OpenSearch Serverless NextGen collection, you choose a collection type. The collection type determines how documents are indexed and, importantly for migration, whether the source document ID is preserved or replaced with a server-generated ID. Migration Assistant adapts metadata migration to the collection type so that the migrated indexes are compatible with the target.
| Collection type | Document IDs | Migration behavior |
|---|---|---|
|
|
Source document IDs are preserved. |
Full-text search workloads. Index mappings and settings are migrated and adapted to the collection model. Because source IDs are preserved, backfill is idempotent and re-running it does not create duplicate documents. |
|
|
Server-generated IDs (source document IDs are not preserved). |
Time-series workloads. RFS auto-detects the collection type and enables server-generated IDs so backfill can write documents to the collection. Do not rely on capture and replay to reconcile updates or deletes by source |
|
|
Server-generated IDs (source document IDs are not preserved). |
Vector and semantic search workloads. |
Important
Document-ID preservation during capture and replay is only meaningful on a SEARCH collection. On TIMESERIES and VECTORSEARCH collections, the target assigns its own document IDs, so an update or delete captured against a specific source _id cannot be matched to the same document on the target. If your migration relies on capture and replay to reconcile updates and deletes by document ID, target a SEARCH collection.
