Upgrade for >= 3.6 #509
Description
We have changed the command line options for upgrades in ArangoDB. This was necessary because the old --database.auto-upgrade=true has some bad properties. From 3.6 on, there is a setting --cluster.upgrade=<VALUE> where <VALUE> can be auto, disable, force and online.
Since coordinator pods are created without persistent volume and so have no memory about previous versions, we had a problem since not all upgrade procedures were run properly in k8s.
To fix this, we should always add the command line option --cluster.upgrade=online to the start command for the coordinators, provided the version is >= 3.6.0. If the version is lower, this must not be done, since the arangod would error out since it does not know the option.
We should verify that in a normal upgrade procedure (patch level or minor or major), the coordinators are simply restarted and are never started with the option --database.auto-upgrade=true.