ES full cluster restart requirement for major upgrade

[laurentiusoica]

Hello,

I plan an ES cluster that should not have downtime. I read that for major version upgrades there is a requirement to do a full cluster restart https://www.elastic.co/blog/multiple-elasticsearch-clusters.

I see that ES 6.x doesn't require Full cluster restart anymore, but it is not clear for me if this is only specific to 6.x or it is planned for all major releases that follows 6.x https://www.elastic.co/guide/en/elasticsearch/reference/current/setup-upgrade.html#upgrade-paths

What would be the best approach in order to support major upgrades without downtime at cluster level ?

Thanks

[warkolm (Mark Walkom)]

You need to be in 5.6.X to be able to do a rolling restart to 6.0.

[laurentiusoica]

The cluster is not deployed yet so I could go directly with latest. My concern is how will I handle an upgrade to 7.x let's say, without downtime.

[warkolm (Mark Walkom)]

Once you are on 5.6.X, you can then go to 6.X via rolling, and then continue that through to 7/8/9/whatever.

[rjernst (Ryan Ernst)]

To clarify, the current plan going forward is to allow rolling upgrades from the last minor version of the previous major. So, say that when 7.0 is released, 6.4 is the last release in 6.x. Then any 6.4 release should allow doing a rolling upgrade to any 7.x release.