How to optimize cluster performance with profiles

charm
, ,
1

How to optimize cluster performance with profiles

Note: Profile configuration is currently only available in the edge channel.

The charmed OpenSearch operator provides a way to tune the performance of deployed OpenSearch clusters through the profile configuration parameter. Profiles define resource requirements and tuning performance related parameters to suit different use cases. One of the said parameters would be the JVM allocated heap size for example.

Currently, the OpenSearch charm supports two profiles:

  • testing – optimized for lightweight, development, or testing workloads.
  • production – optimized for production-grade, large-scale deployments.

By default:

  • The testing profile is applied by default when deploying OpenSearch version 2.
  • The production profile will become the default in the next stable release of OpenSearch version 3.

Deploy OpenSearch with the testing profile

juju deploy opensearch --channel=2/edge --config profile=testing

This command deploys OpenSearch with the testing profile, which applies the following requirements and constraints:

  • Cluster can run with a minimum of 1 node (with both ClusterManager and Data roles).
  • No strict memory requirements are enforced.
  • JVM heap size (-Xms / -Xmx) is set to 1 GB by default.

TLS is still required, as it is a first-class requirement for all OpenSearch deployments.

For example, to deploy a self-signed certificates operator and integrate it with OpenSearch:

juju deploy self-signed-certificates --channel=latest/stable
juju integrate self-signed-certificates opensearch

After deployment, the application status should look similar to:

App                       Version  Status  Scale  Charm                     Channel        Rev  Exposed  Message
opensearch                         active      1  opensearch                2/edge         274  no
self-signed-certificates           active      1  self-signed-certificates  latest/stable  264  no

If you inspect the logs using juju debug-log --include opensearch/0 --level WARNING you will see a warning indicating that the testing profile is in use:

unit-opensearch-0: 07:42:00 WARNING unit.opensearch/0.juju-log opensearch-peers:1: Testing profile is used. This profile is not suitable for production use and should only be used for testing purposes.

Deploy OpenSearch with the production profile

juju deploy opensearch --channel=2/edge --config profile=production
juju deploy self-signed-certificates --channel=latest/stable
juju integrate self-signed-certificates opensearch

When deploying with the production profile, OpenSearch enforces stricter constraints to ensure stability and performance. For example, if only a single node is deployed, the application will enter a blocked state:

App                       Version  Status   Scale  Charm                     Channel        Rev  Exposed  Message
opensearch                         blocked      1  opensearch                                 1  no       At least 3 cluster manager nodes are required. Found only 1. - At least 3 data nodes are required. Found only 1.

The production profile has the following requirements and optimizations:

  • Minimum of 3 cluster manager nodes.
  • Minimum of 3 data nodes.
  • Recommended memory: 8 GB.
  • JVM heap size (-Xms / -Xmx) is automatically set to 50% of available RAM, with a minimum of 4 GB and a maximum of 31 GB.

Important: If any of these requirements are not met, the charm will remain in a blocked state until corrected.

Update profile at runtime

The OpenSearch profile can be changed after deployment. The charm automatically detects the update and reconfigures the instance accordingly.

juju config opensearch profile=<profile name>testing
# or
juju config opensearch profile=production

Profile comparison

Note: Both profiles enforce the same system prerequisites regarding swap that must be disabled and vm.max_map_count >= 262144. The charm would guarantee the same high availability in both profiles.

Feature / Setting testing Profile production Profile
Cluster size Minimum 1 node (ClusterManager + Data) Minimum 3 cluster manager nodes and 3 data nodes
Memory requirement No enforced requirement Recommended: 8 GB
JVM Heap size Fixed at 1 GB 50% of RAM, min 4 GB, max 31 GB
Default use case Development / testing / lightweight loads Production workloads and large deployments