This topic describes the how to change the configurations of your OpenSearch Vector Search Edition instance and provides precautions.
Usage Notes
Item | Description |
Number of Query Result Searcher (QRS) workers | You can increase or decrease the number of QRS workers. If you want to decrease the number of QRS workers, make sure that the number of QRS workers after scale-in is greater than or equal to the number of QRS workers that are being used. The console will verify the number and show prompts. For example, if your instance contains 10 QRS workers and four QRS workers are being used, you can decrease the number of QRS workers by six at most. |
Specifications of QRS workers | You can only upgrade the specifications of QRS workers. The number of CPU cores and memory capacity of an upgraded QRS worker must be no less than those of the original QRS worker. |
Number of Searcher workers | You can increase or decrease the number of Searcher workers. If you want to decrease the number of Searcher workers, make sure that the number of Searcher workers after scale-in is greater than or equal to the number of Searcher workers that are being used. The console will verify the number and show prompts. For example, if your instance contains 10 Searcher workers and six Searcher workers are being used, you can decrease the number of Searcher workers by four at most. |
Specifications of Searcher workers | You can only upgrade the specifications of Searcher workers. The number of CPU cores and memory capacity of an upgraded Searcher worker must be no less than those of the original Searcher worker. |
Table storage mode
Data storage and queries across various tables utilize the shared resources of Searcher workers, including CPU, memory, and disk space:
The diagram illustrates two vector tables: Table 1 with 4 shards and Table 2 with a single shard. Data for Table 1 is evenly distributed across its shards based on the index. Table 2, having only one shard, acts as a broadcast table, storing a full index data set on each data node.
Precautions
Scaling operations are immediate and billing is based on the new configuration.
Each operation should be either a scale-out or scale-in, not both at the same time.
For example, if you have two QRS and two Searcher workers, you cannot reduce QRS workers to one while increasing Searcher workers to three at the same time.
You may reduce both QRS and Searcher workers to one, or maintain the number of Searcher workers while reducing QRS workers to one.
Alternatively, you can increase both QRS and Searcher workers to three, and so on.
After modifying the number of Searcher workers, adjust the number of data node replicas or edit the table to alter the shard count and reallocate resources on the Instance Details page.
In a single cluster, the number of QRS workers = the number of QRS workers × the number of replicas; similarly, the number of Searcher workers = the number of replicas × the number of data shards. Specify an integer multiple of the shard count when scaling out, as a shard represents the smallest data block unit.
The instance will be unavailable during an upgrade from a single node to multiple nodes. Ensure no production query traffic occurs during this time.
Procedure
On the Instance page of the console, find the desired instance on the list, and click Scale In/Out in the Actions column:
On the by page, set the desired number and specifications for QRS and Searcher workers, agree to the Terms of Service, and click Buy Now:
