We have a three-node cluster onprem, and during rolling upgrade of the individual Elastic-nodes, we tend to toggle 'cluster.routing.allocation.enable' between 'primaries' and null.
Is this necessary to do between each individual node upgrade, or is it enough to set the it to 'primaries' before the upgrade and then back to null after all three nodes are done and online?
If it is: Is it crucial to wait for unsassigned shards to reach 0 before setting .enable to 'primaries' and upgrade the next node?
It takes a noticable time for 'unassigned shards' to reach 0, so it'd be time efficent to just play the wait-game once.