Our Kafka indexing service is very weird. We have a datasource with 10 replicas. Sometimes when submitting the new supervisor, it took a rather longer time than usual. As it finally finished submitting, it should new 10 running tasks and graceful shutdown the exiting supervisor. But there were no new tasks, and one of the old tasks finished with status SUCCESS, and the others were running. When the taskDuration elapsed, the others 9 tasks never Handoff.
The druid used is 0.9.2-rc2. We have a lot of datasource, and it seems the problem randomly pick datasources.
Some supervisor spec is as follows:
"replicas":10, "taskCount":1, "taskDuration":"PT1H", "lateMessageRejectionPeriod":"PT48H", "completionTimeout":"PT24H", "buildV9Directly":true
Thank you very much,