Page MenuHomePhabricator

Evaluate mixed-cluster behavior of elasticsearch + opensearch
Closed, ResolvedPublic

Description

The migration from elasticsearch to opensearch is estimated to run for many days, potentially a week or more. We will be in a mixed-cluster state for that time period. Historically, during major version upgrades, there have been limitations around things moving backwards. For example once a shard primary was on the new version it could not move back to an old version. Some similar limitations may exist here. We should run some tests to understand what limitations might be imposed by this.

We might also want to try and understand if we it matters when we migrate the master-capable nodes. First, last, or at any arbitrary time.

Incomplete list of potential tests:

  • elastic can join cluster with opensearch master
  • replica shard can move from opensearch to elastic
  • primary shard can move from opensearch to elastic
  • master can move from opensearch to elastic
  • master can move from elastic to opensearch
  • new primary shards can be allocated to elastic by opensearch master
  • new primary shards can be allocated to opensearch by elastic master

Event Timeline

Ran some basic tests using docker-compose (configuration: P71049) and everything looks to move back and forth between the two with no issues.

Overall, it looks like this works similar to a minor version upgrade. Everything I tested, with both elastic and opensearch as current cluster master, works the same in the mixed cluster as it does in a homogeneous cluster. It looks like we may not need to do anything special to handle the mixed-cluster state on deployment.

Gehel claimed this task.