As someone who had to admin an on-prem installation of various Atlassian products, upgrades were always a nightmare: you could try multiple dry-runs to make sure data tables were updated, and plug-in upgrades were good, but invariably there would still be a good chance that something would still go wrong when you tried it with production.

The workflow we settled on was basically leave the old version alone, rsync over all the data to a completely new app and DB servers, and do the upgrades on the new systems. Any attempt at in-place upgrades was just asking for trouble.

Anyone who was prime on the software was very happy when The New Guy on the team arrived due to personnel churn and they got handed the baton for dealing with it.