Oracle Cloud Infrastructure Documentation

Resizing a Search with OpenSearch Cluster

Resize an OpenSearch cluster to change the node configuration to meet performance and other requirements.

You can resize an OpenSearch cluster in the following ways:

Note

In-place horizontal resizing allows the cluster to continue accepting reads and writes. Other resize operations can temporarily place the cluster in read-only mode. Increasing data-node storage without changing other settings doesn't require read-only mode. Cluster performance and individual requests can be affected during resizing. For more information, see Cluster Patching and Resize Operations.

For information on node hardware settings, see OpenSearch Node Types.

Managing Searchable-Snapshot Indexes

You can't reduce the search-node count to zero while searchable-snapshot indexes remain in the cluster. Perform the following steps before removing the last search node.

  1. Find the searchable-snapshot indexes by checking index settings for index.store.type: remote_snapshot.
  2. If the data must remain searchable, ensure the data nodes have enough storage, then restore the backing snapshot as regular indexes using storage_type: local and different index names. Verify the restored data and update application references or aliases before removing the original searchable-snapshot indexes.
  3. Delete the searchable-snapshot indexes from the cluster after they're no longer needed. Keep the backing snapshots if they're required for retention or future restoration.
  4. Confirm that no searchable-snapshot indices remain, then horizontally resize the search-node count to zero.

Limitations and Considerations

The following table lists and describes limitations and considerations to consider when resizing OpenSearch clusters.

Scenario Limitation or consideration Recommended action
Improve availability or resiliency Adding capacity across nodes is the appropriate scaling approach. Horizontally resize the cluster.
Improve performance More resources per node might be preferable to adding nodes. Vertically resize the cluster.
Both horizontal and vertical scaling are needed Horizontal scaling might also require vertical scaling. Perform the vertical resize before the horizontal resize.
Add data nodes More data nodes can affect cluster performance. Assess the expected performance impact before increasing node count.
Any resize operation Cluster performance can be affected while resizing is in progress. Plan and monitor the resize accordingly.
Decrease memory or OCPUs Reducing resources can degrade performance. Review Search with OpenSearch metrics before choosing lower values.
Decrease data-node storage Data-node storage can't be reduced. Storage can only be increased, and only under supported resize conditions.
Minimum node configuration Node configurations can't be reduced less than the supported minimums. For all node configuration requirements, see OpenSearch Node Types.
Vertical downscaling The service validates recent metrics against target CPU and memory utilization before applying the new configuration. Use metrics to select specific values; if the check fails, the operation returns InvalidParameter with a cannot downscale message.
Uncertain downscaling values Trying several configurations without understanding the performance impact isn't recommended. Gather enough historical metrics or contact support for recommendations.