Bug #63335
openNotification kafka: Persistent messages are removed even when the broker is down and no ack is received
0%
Description
This is the regression caused due to changes in https://github.com/ceph/ceph/pull/51864.
Currently the way Kafka processing works when the ACK level is set broker, is notification is sent to kafka using the rd_kafka_produce(), which currently DOES not return -1 if the BROKER is down.
But when the broker is down, the CALLBACK from kafka acknowledgement is never returned and hence this ack is propagated back to notification system which then does a retry and does not delete the persistent notification from queue.
But after the https://github.com/ceph/ceph/pull/51864, the connection was destroyed after idle_timeout and callbacks were invoked with SUCCESS, so even tho the notification were not delivered, they were removed from the queue because the callback was invoked with sucess.