We are trying to upgrade our 7.8.0-OSS cluster to 7.15.2 using an automated deployment method where all new nodes are added to the cluster before the old nodes are taken down 1-by-1.
We have noticed that during deployment the elected master node must be taken down last in order for the upgrade to finish successfully. If the elected master node is killed before all other 7.8.0 nodes we get this error node version [7.8.0] may not join a cluster comprising only nodes of version [7.15.2] or greater and the cluster goes into a red state. It's mentioned here that nodes cannot join a cluster with a newer version master however this issue has never occurred before when carrying out rolling upgrades between OSS versions.
What is different in the basic version that is causing this to break? Is there anything that we can configure in order to keep the old functionality?
Old cluster details
3 nodes with data and master roles
Java 11
v7.8.0-OSS
New cluster details
3 nodes with data and master roles
Java 11
v7.15.2
Thanks in advance for your help.