Merging ES Clusters

Hi All,

We have two clusters:-

  • Cluster A we have 5 new nodes running ES v7.17 (All data nodes out of which two are master)

  • Cluster A Contains no data what so ever it was meant to be a backup cluster but now they want it to be active

  • Cluster B 10 active nodes Running ES v7.9.3(All data nodes, 2 master).

  • Cluster B Contains data that can be backed up using snapshot repository.

Can someone suggest how to merge these two cluster because we were asked to.

Need to meet below two conditions:

  • No data loss.
  • Minimal downtime.

My idea was to go about it by doing the following

  1. backup cluster B using snapshot
  2. redirect all data sources to cluster A
  3. update all cluster B ES to 7.17
  4. Vote out nodes one by one from Cluster B
  5. Generate a new UUID on said node or delete the /data directory which contains the current UUID
  6. setup ES conf on node to join cluster A
  7. After all nodes are moved backup the snapshot to the new cluster

I did not proceed with my takeaway because of l:

  1. I just don't know if this will work and how to proceed with step number 5 to get rid of the old UUID
  2. can i backup the snapshot to the new cluster to retrieve old data

Thanks in advance

I would recommend instead doing the following:

  1. Make another node in cluster A a master node. You should always have at least 3 master eligible nodes in any cluster. Having just 2 nodes does not allow one of the master eligible noides to go down without making the cluster unavailable.
  2. Perform a rolling upgrade of cluster A to Elasticsearch 7.17. Take a snapshot prior to the upgrade.
  3. Shut down all nodes in cluster B, clear the data path, change the config and restart them so they can join cluster A as data nodes.
1 Like

I apologize cluster A is actually configured to have 3 master-eligible nodes I learned from my mistakes.

I'll try to do as you said but just to be make it clear by clearing the data you mean all of the Elasticsearch.conf directive "path.data" content and folder am i right?

Yes, the nodes in cluster B need to join as completely new data nodes that have no knowledge of being part of any prior cluster.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.