Problem with node after restart no partitions?

[Thunder Stumpges <ts...@ntent.com>]

Hi group,

We've been trying to track down a problem this morning for a little while, and thought I'd ask here while we keep looking.

We have 3 nodes (rep-3) running 8.1.1. We attempted a rolling upgrade yesterday to 8.2.1, and on the first node, after restarting, a single topic (a samza intermediate topic) started throwing replica fetcher errors over and over ("NotLeaderForPartition"). There may or may not have been other things attempted at this time (not by me so I cannot say for sure). Anyway we ended up rolling back to 8.1.1 and ALL data had been DELETED from that node. It spent most of yesterday re-syncing, and came into sync last night, and a rebalance made everything run smoothly (*except for these damn replica fetcher errors for that one partition).

Today my colleague attempted the "unsupported" topic delete command for the "bad" partition, and bounced that one troublesome node.

Upon coming up, I can see in server.log that it is reading in all of the segments in, and then starts spitting out a samza topic fetch error, and through JMX the "ReplicaManager".LeaderCount is 0. It is not attempting to fetch or load any topics.

The other two brokers are showing under-replicated (obviously). What is going wrong? How can we get that samza topic really and truly gone? (if that is the cause of the broker not coming up)

Thanks,
Thunder