Under what circumstance will a Kafka Partition (not consumer) lose its offset, and reset to 0? We are running Kafka 0.10.2.1
We have a Kafka topic with 8 evenly distributed partitions. Each partition's offset is relatively in line with each other. Partition Latest Offset Leader Replicas In Sync Replicas Preferred Leader? Under Replicated? 0 7,450 6165 <https://kafka.pinadmin.com/clusters/data06/brokers/6165> (6165,6168,6156) (6156,6165,6168) true false 1 7,611 6011 <https://kafka.pinadmin.com/clusters/data06/brokers/6011> (6011,6156,6157) (6011,6156,6157) true false 2 7,467 6026 <https://kafka.pinadmin.com/clusters/data06/brokers/6026> (6026,6157,6158) (6026,6157,6158) true false 3 7,511 6028 <https://kafka.pinadmin.com/clusters/data06/brokers/6028> (6028,6158,6159) (6028,6158,6159) true false 4 7,458 6128 <https://kafka.pinadmin.com/clusters/data06/brokers/6128> (6128,6159,6160) (6128,6159,6160) true false 5 0 6004 <https://kafka.pinadmin.com/clusters/data06/brokers/6004> (6004,6160,6161) (6004,6160,6161) true false 6 7,599 6148 <https://kafka.pinadmin.com/clusters/data06/brokers/6148> (6148,6161,6162) (6148,6161,6162) true false 7 7,511 6172 <https://kafka.pinadmin.com/clusters/data06/brokers/6172> (6172,6162,6163) (6162,6172,6163) true false Recent I noticed partition-5 has a latest offset = 0. We did not issue any offset resets. Can someone explain how this may occur? We did have a incident where we had replace multiple kafka brokers, and we have lost all replicas for partition 5. Since in Kafka 10, offset information is no longer stored in zookeeper, can such a reset happen if we lose all replicas for partition-5, and the controller doesn't get to update the replacement in time? Thank you.
