Rolling upgrades to new VMs

[griese]

Hello,

I have two upgrades that I need to do and I was wondering if this process would work.
1) I need to migrate off of some old hardware (physical servers and an old storage array) and on to some VMs. I planned on using the latest OVA download.
2) Those old servers are on v2.0.8 so I planned on migrating to v2.1.7.

I currently only have 3 nodes installed of a 4 node cluster license.

So my plan was to do the following.

A) Install a new v2.1.7 VM from OVA
B) Add it to the existing cluster
C) Wait for shards to sync up
D) Remove old v2.0.8 node
E) Confirm cluster health is Green
F) Repeat starting with Step A until all of the old nodes are gone

Does that sound possible? Any gotchas I should watch out for?

Thanks,
Troy

[cdienger]

There were some significant changes between 2.0.8 and 2.1.7 and I would expect some problems trying to cluster the two versions. I would suggest:

A) Upgrade the 2.0.8 node(s) to 2.1.7
B) Install a new 2.1.7 vm from ova
C) Add it to the existing cluster
D) Wait for shards to sync up*
E) Remove the 2.0.8 node
F) Wait for a green health status
G) Repeat

*For step D be aware that only open indices will be synced. If you have closed indices on the 2.0.8 machines that you want to sync to 2.1.7 then you'll need to open those and allow them time to sync. You can find a script to do this at https://github.com/elastic/elasticsearch/issues/12963.

[griese]

Thanks for the feedback.

I'm not sure I completely understand what you're saying so I'll try to clarify by adding a couple more words. Please verify the steps are correct.

One question about this process though. Should I upgrade all of them to 2.1.7 first before trying to replace them? I assume that would be best.

A) Upgrade existing physical 2.0.8 node(s) to 2.1.7 (e.g. server1, server2, server3)
i) wait for a green health status
B) Install a new 2.1.7 vm from ova (e.g. serverA)
C) Add it to the existing cluster
D) Wait for shards to sync up*
E) Remove the old physical 2.0.8 node(s) which is now version 2.1.7 (server1)
F) Wait for a green health status
G) Repeat from step B for remaining servers (server2, server3)

Thanks,
Troy

[cdienger]

Yes, that should do it.