You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Bob Barrett (Jira)" <ji...@apache.org> on 2020/03/30 23:12:00 UTC

[jira] [Created] (KAFKA-9788) Sensor name collision for group and transaction coordinator load metrics

Bob Barrett created KAFKA-9788:
----------------------------------

             Summary: Sensor name collision for group and transaction coordinator load metrics
                 Key: KAFKA-9788
                 URL: https://issues.apache.org/jira/browse/KAFKA-9788
             Project: Kafka
          Issue Type: Bug
            Reporter: Bob Barrett
            Assignee: Bob Barrett


Both the group coordinator and the transaction coordinator create a Sensor object on startup to track the time it takes to load partitions, and both name the Sensor "PartitionLoadTime":

[https://github.com/apache/kafka/blob/trunk/core/src/main/scala/kafka/coordinator/transaction/TransactionStateManager.scala#L98]

[https://github.com/apache/kafka/blob/trunk/core/src/main/scala/kafka/coordinator/group/GroupMetadataManager.scala#L92]

However, Sensor names are meant to be unique. This name collision means that there is actually only one underlying "PartitionLoadTime" Sensor per broker, which is marked for each partition loaded by either coordinator, resulting in the metrics for group and transaction partition loading to be identical, and based the combination of each data set. These should be renamed to allow distinguishing between the two coordinator types.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)