Michael Westerby created KAFKA-20947:
----------------------------------------
Summary: ZkWriteSnapshotTimeMs metric reports negative values
Key: KAFKA-20947
URL: https://issues.apache.org/jira/browse/KAFKA-20947
Project: Kafka
Issue Type: Bug
Components: kraft, migration
Affects Versions: 3.9.2
Reporter: Michael Westerby
The {{ZkWriteSnapshotTimeMs}} metric reports negative values when the KRaft
migration driver performs the full metadata sync from KRaft to ZooKeeper during
the migration.
This is due to an error in how the elapsed time is calculated in
{{SyncKRaftMetadataEvent}}, where this is computed as {{startTime - endTime}}
rather than {{endTime - startTime}}:
- {code:java}
controllerMetrics.updateZkWriteSnapshotTimeMs(NANOSECONDS.toMillis(startTime -
endTime));
{code}
-
https://github.com/apache/kafka/blob/5e9866f43ab8e7e41ef39e5584ac50019381328d/metadata/src/main/java/org/apache/kafka/metadata/migration/KRaftMigrationDriver.java#L846
The correct order is seen in the other call site to this method in
{{MetadataChangeEvent}}:
- {code:java}
controllerMetrics.updateZkWriteSnapshotTimeMs(NANOSECONDS.toMillis(time.nanoseconds()
- startTime));
{code}
-
https://github.com/apache/kafka/blob/5e9866f43ab8e7e41ef39e5584ac50019381328d/metadata/src/main/java/org/apache/kafka/metadata/migration/KRaftMigrationDriver.java#L572
The fix for this would be to correct the subtraction order where this is
invoked in {{SyncKRaftMetadataEvent}} to prevent these negative values.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
