# Collection types
<a name="mts-collection-types"></a>

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 | 
| --- | --- | --- | 
|  `SEARCH`  | 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. | 
|  `TIMESERIES`  | 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 `_id` on this target type. | 
|  `VECTORSEARCH`  | Server-generated IDs (source document IDs are not preserved). | Vector and semantic search workloads. `knn_vector` field mappings are automatically converted to the Faiss HNSW engine for serverless compatibility, and `model_id` references are removed because Amazon OpenSearch Serverless NextGen does not support training APIs. | 

**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.