Merge pull request #284711 from HeidiSteen/heidist-august

[azure search] Robert's feedback on stored false effect on reindexing

Author: prmerger-automator[bot]

articles/search/search-howto-reindex.md

@@ -9,7 +9,7 @@ ms.author: heidist
 
 ms.service: cognitive-search
 ms.topic: how-to
-ms.date: 07/01/2024
+ms.date: 08/14/2024
 ---
 
 # Update or rebuild an index in Azure AI Search
@@ -43,10 +43,12 @@ Queries continue to run, but if you're updating or removing existing fields, you
 
 + [Indexers automate incremental indexing](search-indexer-overview.md). If you can use an indexer, and if the data source supports change tracking, you can run the indexer on a recurring schedule to add, update, or overwrite searchable content so that it's synchronized to your external data.
 
-+ If you're making index calls directly, use `mergeOrUpload` as the search action.
++ If you're making index calls directly through the [push API](search-what-is-data-import.md#pushing-data-to-an-index), use `mergeOrUpload` as the search action.
 
 + The payload must include the keys or identifiers of every document you want to add, update, or delete.
 
++ If your index includes vector fields and you set the [`stored` property to false](vector-search-how-to-configure-compression-storage.md#option-3-set-the-stored-property-to-remove-retrievable-storage), make sure you provide the vector in your partial document update, even if the value is unchanged. A side effect of setting `stored` to false is that vectors are dropped on a reindexing operation. Providing the vector in the documents payload prevents this from happening.
+
 + To update the contents of simple fields and subfields in complex types, list only the fields you want to change. For example, if you only need to update a description field, the payload should consist of the document key and the modified description. Omitting other fields retains their existing values.
 
 + To merge inline changes into string collection, provide the entire value. Recall the `tags` field example from the previous section. New values overwrite the old values for an entire field, and there's no merging within the content of a field.

articles/search/vector-search-how-to-configure-compression-storage.md

@@ -7,7 +7,7 @@ author: heidisteen
 ms.author: heidist
 ms.service: cognitive-search
 ms.topic: how-to
-ms.date: 08/05/2024
+ms.date: 08/14/2024
 ---
 
 # Reduce vector size through quantization, narrow data types, and storage options
@@ -40,7 +40,7 @@ Quantization applies to vector fields receiving float-type vectors. In the examp
 
 Two types of quantization are supported:
 
-- Scalar quantization compresses floats into narrower data types. AI Search currently supports int8, which is 8 bits, reducing vector index size fourfold.
+- Scalar quantization compresses float values into narrower data types. AI Search currently supports int8, which is 8 bits, reducing vector index size fourfold.
 
 - Binary quantization converts floats into binary bits, which takes up 1 bit. This results in up to 28 times reduced vector index size.
 
@@ -228,7 +228,11 @@ An easy way to reduce vector size is to store embeddings in a smaller data forma
 
 The `stored` property is a boolean on a vector field definition that determines whether storage is allocated for retrievable vector field content. The `stored` property is true by default. If you don't need vector content in a query response, you can save up to 50 percent storage per field by setting `stored` to false.
 
-When evaluating whether to set this property, consider whether you need vectors in the response. Because vectors aren't human readable, they can be omitted in a query response that's rendered on a search page. Keep them, however, if you're using vectors in downstream process that consumes vector content.
+Considerations for setting `stored` to false:
+
+- Because vectors aren't human readable, you can omit them from results sent to LLMs in RAG scenarios, and from results that are rendered on a search page. Keep them, however, if you're using vectors in a downstream process that consumes vector content.
+
+- However, if your indexing strategy includes [partial document updates](search-howto-reindex.md#update-content), such as "merge" or "mergeOrUpload" on a document, be aware that setting `stored` to false will cause vectors in the non-stored field to be omitted during the merge. On each "merge" or "mergeOrUpload" operation, you must provide the vector fields in addition to other nonvector fields that you're updating, or the vector will be dropped. 
 
 Remember that the `stored` attribution is irreversible. It's set during index creation on vector fields when physical data structures are created. If you want retrievable vector content later, you must drop and rebuild the index, or create and load a new field that has the new attribution.