Rolling Upgrade on a big cluster

[aramey]

With hundreds of indexer nodes in a cluster, it might take more than one business day to upgrade them all during a rolling upgrade. And thus indexers would run on different versions for a short while (while documentation states that indexers within a cluster should all be the same version.)

Is it possible to split the rolling upgrade process between two business days?
If yes, should the cluster master be kept into 'rolling upgrade' mode during the night between the two rollouts?
Thanks

[hmallett · Accepted solution]

Yes, you can take as long as you need for the rolling upgrade, you just accept that you can't do any other cluster maintenance (including bundle pushes) during the upgrade.
I would personally keep the cluster master in rolling upgrade mode overnight. I don't know what the implications of taking it out of that mode while the indexers are mixed versions would be though.