What I usually do is follow the order here
https://druid.apache.org/docs/27.0.0/operations/rolling-updates
And upgrade service by service.
Example :
• Rolling upgrade historical
• When it’s done, Rolling upgrade middle managers
• Then Brokers
• Then coordinator-overlords
I have not had problems with that procedure.
Once I did not respect that and rolling updated all services at the same time, I had a bit of downtime but restarting coordinator-overlords fixed everything.