Hi fellow kafka users, Recently we tried and are stuck in the middle of a migration from kafka 3.2 to 3.9. We upgrade one broker (setting inter.broker.protocol.version=3.2-IV0 to ensure compat, not setting log.message.format.version cause it's supposed to be the same aka 3.0-IV1).
One important things is : - data were on a ephemeral storage, so broker restarted empty - we do not drain the cluster before (aka not moving off partition leadership) It should have done automatically but.. After that we encounter lot of but not everywhere producing, consuming and transactions errors (I discovered transactions at this occasion) from lot of our application. It take time to stabilize. My guess is everything came back to normal when I manually reassigned partition leadership from the migrated/rebooted broker. (particularly __consummer_offsests and __transactions_state). I don't understand why we ended in this state. If you have ideas :) So after dealing with data back filling (but that's another story) we are stuck in the middle of the migration. Things looks like stable. We still have a lot of this worrying warning on the new broker : 2024-12-12 16:26:29,819] WARN [AddPartitionsToTxnSenderThread-102]: AddPartitionsToTxnRequest failed for node XX with invalid version exception. This suggests verification is not supported.Continuing handling the produce request. (kafka.server.AddPartitionsToTxnManager) Is it safe to continue like this? Any help/ clues appreciated. Best -- Raphael Mazelier
