I actually think that it is not that risky (although ymmv). Worker nodes are pretty independent from the scheduler/webserver. As long as the datamodel hasnt changed and nodes dont change their reporting (new statusses) to the db (that hasnt happened for a long time) you are probably okay.
So the proper way to do (test ;-)) it is scheduler first, webserver next, nodes. Bolke Verstuurd vanaf mijn iPad > Op 13 sep. 2019 om 12:49 heeft Driesprong, Fokko <[email protected]> het > volgende geschreven: > > Hi John, > > I've never tried it like you're suggesting. It feels a bit risky. I can't > tell what you will encounter if you run different versions of Airflow. How > are you running Airflow? > > Cheers, Fokko > > Op do 12 sep. 2019 om 20:29 schreef John Smodic <[email protected]>: > >> Hey all, >> >> I'm looking to find an upgrade path for upgrading Airflow from 1.10.2 to >> 1.10.5. >> >> But the problem is, the Airflow production cluster is pretty busy all the >> time. >> >> Would it be problematic to upgrade the Webserver and Scheduler to 1.10.5 >> and gradually update the nodes as they become free? This would leave a >> period where nodes are on 1.10.2 and other nodes are on 1.10.5, but I don't >> know if that is expected to cause any issues. >> >> For 1.9 -> 1.10 I did a full blue green deployment, but didn't necessarily >> want to go through with that for what seems like a relatively minor series >> of patches. >> >> Thanks! >>
