Emanuele Sabellico
Emanuele Sabellico
It's correct to send the ListOffsets request to the preferred replica. The loop probably comes from this discovered bug: https://github.com/confluentinc/librdkafka/issues/4620 When enabling debug logs, could you check if it's receiving...
@kphelps The replica id should be set to -1 in clients, and to the broker id in followers, see the RPC definition https://github.com/apache/kafka/blob/2f401ff4c85f6797391b8a3dd57d651f4de3d6ad/clients/src/main/resources/common/message/ListOffsetsRequest.json#L42 The error `NOT_LEADER_OR_FOLLOWER `happens when the broker...
Thanks @kphelps I was checking this issue more in depth and understood the problem, it's different from what I linked and as you said could be solved in two ways,...
Cannot fix it by sending the request to the follower because there are some problems: if replica id was different from CONSUMER_REPLICA_ID (-1), the isolation level parameter would be ignored,...
@kphelps sorry, giving we're have having an issue with the public CI, I've created this internal branch with your changes. #4754
Hi @konlps, it can be that during deployment that broker is not the leader for that partition anymore. Does the problem disappear when the brokers have stabilized?