Could you try starting from scratch again? The recent fix that we had may not be able to recover a cluster already in an inconsistent state.
Thanks, Jun On Thu, Dec 12, 2013 at 8:45 PM, David Birdsong <david.birds...@gmail.com>wrote: > I was running a 2-node kafka cluster off github trunnk at: > eedbea6526986783257ad0e025c451a8ee3d9095 > > ...for a few weeks with no issues. I recently downloaded the 0.8 stable > version, configured/started two new brokers with 0.8. > > I successfully reassigned all but 1 partition from the older pair to the > newer pair, but have 1 partition seemingly stuck on an the old leader. The > replicas, ISR, and leader are all the same--no extra nodes are replicating > this last partition--this was true before any changes. > > I came across this thread: > > http://mail-archives.apache.org/mod_mbox/kafka-users/201312.mbox/%3ccacnty1ddbjse1bxrj1ertrxi+zbz3wawyvjdevvjpootnyo...@mail.gmail.com%3E > > ..and unlike the poster, I'm free to play fast and loose, so I built off of > trunk at: dd58d753ce3ffb41776a6fa6322cb822f2222500 > > I first upgraded one of the desired target ISR's and after a few minutes > upgraded the existing leader and bounced it, temporarily breaking that > partition--no luck. > > I'm at a loss as to how to recover this partition's data; short of the data > being recovered, how to even regain use of the partition. The data's not > critical, this was just an exercise in gaining operation familiarity w/ > kafka. > > I can't find any docs on how to get out of this situation. >
