This topic outlines the configuration of a vector index within an existing OpenSearch retrieval engine edition instance and illustrates the use of the SDK for data retrieval through vector queries.
Purchase an OpenSearch retrieval engine edition instance
For more information, see Purchase an OpenSearch retrieval engine edition instance.
Configure the table
Upon purchasing a new instance, you will find its status on the product page as "Pending configuration." An empty cluster, matching the purchased query and data nodes in number and specifications, will be automatically deployed. Before initiating searches, configure the cluster's Table basic information> Data synchronization> Index schema> Index rebuilding.
To set up table basic information, customize the Table name, determine the Number of shards, and specify the Number of data update resources.
The shard count for a table is capped at 256, and it is advisable not to exceed three times the number of data nodes in the instance.
The default provision is two free data update resources. Exceeding this number incurs charges calculated as n-2, where n represents the total data update resources for a single table.
For data synchronization, configure the full data source, which includes supported types such as the MaxCompute data source, API data source, Object Storage Service (OSS), and Data Lake Formation (DLF) data source. Taking MaxCompute as an example, select "Add data source," choose "MaxCompute" as the type, and enter the project, accesskeyID, accesskeySecret, Table, partition key, and consider enabling automatic reindexing.
Following successful data source configuration, proceed to set up the Index schema.
Define fields, with at least two required: the Primary key field and the Vector field (the latter should be configured as a multi-value float type).
To include a vector with a category, add a category field, which must be set as either a single-value or multi-value integer type.
Property and field content compression:
For attribute fields, you have the option to compress them. In form mode, choose between uncompressed and compressed; in developer mode, select either no_compressor or file_compressor.
For field content, decide whether to apply compression. By default, it remains uncompressed. Multi-value and STRING types default to uniq compression, while single-value numeric types use equal compression.