Persistence requirements for cluster_manager nodes

OpenSearch
1

Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):
OpenSearch 2.17.0
OpenSearch operator 2.6.0

Describe the issue:
I use OpenSearch operator GitHub - opensearch-project/opensearch-k8s-operator at v2.6.0 and I am interested in what minimum persistence requirements must be satisfied in case of node pool with the only one cluster_manager role.

Configuration:

opensearch-k8s-operator/opensearch-operator/examples/2.x/opensearch-cluster.yaml at main · opensearch-project/opensearch-k8s-operator · GitHub

  nodePools:
    - component: masters
      replicas: 3
      roles:
        - "cluster_manager"
      persistence: #what would be enough to specify here?
    - component: data
      replicas: 3
      roles:
        - "data"
      persistence:
         emptyDir: {}
2

@SlavaUtesinov What do you mean by minimum persistence requirements?
Have you seen this OpenSearch Operator documentation regarding data persistence?

3

Hi, @pablo! Yes, I have seen this documentation, but it answers the question how to set persistence, not to what persistence I should set. Let me clarify my question, OpenSearch nodes have different roles/types, you can look at Creating a cluster - OpenSearch Documentation and my question is does some correlation exist between node role and persistence requirements. For example, maybe nodes with the only one cluster_manager role don’t need state/persistence at all (it is my question), nodes with data role obviously must have state/persistence and etc, something like a table with two columns: role - persistence requirements.

4

@SlavaUtesinov As far as I know, OpenSearch doesn’t care about persistence as it is not aware of it. This is a different layer that is isolated from OpenSearch.
You either want to keep the data or dispose of it once the pod is deleted.
In the production environment, I would recommend using any persistence except emtpyDir.
Also, persistence doesn’t improve anything but the underlying storage class and hardware that is backing it does.

5

You are talking about persistence implementation details, but I am interested in high level abstracts - does cluster_manager node save something on disk in general?

6

@SlavaUtesinov Yes, they will keep metadata of the indices, cluster settings and index templates.

7

Thanks