OpenSearchVectorStore Does Not Fully Apply Mapping Configuration During Index Initialization

When using Spring AI's (version: 1.0.0-M6) OpenSearchVectorStore, the custom mapping configuration is not fully applied during index initialization. Specifically:

loadCustomMapping：
{
"mappings": {
"properties": {
"id": {
"type": "keyword"
},
"content": {
"type": "text",
"analyzer": "standard"
},
"metadata": {
"type": "object"
},
"embedding": {
"type": "knn_vector",
"dimension": 1536,
"method": {
"name": "hnsw",
"space_type": "cosinesimil",
"engine": "lucene"
}
}
}
},
"settings": {
"index": {
"knn": true,
"number_of_shards": 1,
"number_of_replicas": 0
}
}
}

@Bean
public VectorStore vectorStore(OpenSearchClient openSearchClient, EmbeddingModel embeddingModel) {
return OpenSearchVectorStore.builder(openSearchClient, embeddingModel)
.index(indexName)
.mappingJson(loadCustomMapping())
.similarityFunction(similarityFunction)
.initializeSchema(initializeSchema)
.batchingStrategy(new TokenCountBatchingStrategy())
.build();
}

1. During initial index creation, only basic settings are applied:
   "settings": {
   "index": {
   "replication": {
   "type": "DOCUMENT"
   },
   "number_of_shards": "1",
   "provided_name": "mem,
   "knn": "true",
   "creation_date": "1746330816561",
   "number_of_replicas": "1",
   "uuid": "43LTNIlUT8Stt9dADOyqmA",
   "version": {
   "created": "136408027"
   }
   }
   }
2. After inserting the first document, the embedding field is automatically inferred as float type, instead of the knn_vector type specified in our mapping.json:

"embedding": {
  "type": "float"
}

3. Our expected configuration (mapping.json) is:

{
  "mappings": {
    "properties": {
      "id": {
        "type": "keyword"
      },
      "content": {
        "type": "text",
        "analyzer": "standard"
      },
      "metadata": {
        "type": "object"
      },
      "embedding": {
        "type": "knn_vector",
        "dimension": 1536,
        "method": {
          "name": "hnsw",
          "space_type": "cosinesimil",
          "engine": "lucene"
        }
      }
    }
  },
  "settings": {
    "index": {
      "knn": true,
      "number_of_shards": 1,
      "number_of_replicas": 0
    }
  }
}

The mapping.json configuration should be fully applied when creating the index, specifically:

1. The embedding field should be created as knn_vector type
2. All KNN-related settings should be applied
3. Shards and replicas should be set according to the configuration

Start at the OpenSearchVectorStore entry point and trace how mappingJson is handled during index initialization. Compare the requested mappings and settings with the created index and first document behavior. Done means the configured knn_vector mapping, KNN settings, shard count, and replica count are applied when the index is created.