Elasticsearch 6.8.x and two versions of OpenJDK

[lazyusername]

Hello,

we have some mid sized cluster around ~70 nodes running on Debian Stretch (9.x) with OpenJDK-8. We will need to add some extra nodes with Debian Buster, where is present OpenJDK-11.

Both versions of OpenJDK is ok with Elasticsearch 6.8.6 according to compatibiltiy matrix. But is it possible to run cluster with some nodes with OpenJDK-8 and some nodes with OpenJDK-11?

Thx for answer guys!

[lazyusername]

I surely could try to use backported OpenJDK-11 on Debian Stretch nodes as prerequisite. But still, I will enter the situation while I will have two parts of cluster:

1. with OpenJDK-8,
2. with OpenJDK-11.

I found a lot of info about compatibiltiy client a and server parts to use the same Java. But I am having really hard time to find any info about the various supported java versions on the nodes in the same cluster.

Do you have any experience with that guys?

[DavidTurner]

Running with mixed JVM versions isn't something we test as far as I can see, and the manual specifically says not to do this:

The same JVM version should be used on all Elasticsearch nodes and clients.

That's not to say that I know of any specific problems with running with mixed JDK versions. Ultimately it's your data and your call to make.

Newer versions (≥7.x) include the recommended JDK in the installation for this very reason.

[lazyusername]

Thx for reply and info, @DavidTurner!

I am aware about the bundled JDK in ES 7.x and that's pretty cool. But sadly we are not at the point of upgrade anytime soon

The same JVM version should be used on all Elasticsearch nodes and clients.

I noticed that, but somehow I read it as the main message is about conjunction of version on nodes AND clients especially. My bad.

OK, we will consider to do the upgrade old nodes to backported OpenJDK-11 before adding the new nodes. I was not even sure about the compatibility so we would be able to do the rolling restart at this point.