You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@kafka.apache.org by "Stanislav Kozlovski (Jira)" <ji...@apache.org> on 2023/12/26 20:01:00 UTC

[jira] [Resolved] (KAFKA-16007) ZK migrations can be slow for large clusters

     [ https://issues.apache.org/jira/browse/KAFKA-16007?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Stanislav Kozlovski resolved KAFKA-16007.
-----------------------------------------
    Resolution: Fixed

Closing since it's merged

 

> ZK migrations can be slow for large clusters
> --------------------------------------------
>
>                 Key: KAFKA-16007
>                 URL: https://issues.apache.org/jira/browse/KAFKA-16007
>             Project: Kafka
>          Issue Type: Improvement
>          Components: controller, kraft
>            Reporter: David Arthur
>            Assignee: David Arthur
>            Priority: Minor
>             Fix For: 3.7.0, 3.6.2
>
>
> On a large cluster with many single-partition topics, the ZK to KRaft migration took nearly half an hour:
> {code}
> [KRaftMigrationDriver id=9990] Completed migration of metadata from ZooKeeper to KRaft. 157396 records were generated in 2245862 ms across 67132 batches. The record types were {TOPIC_RECORD=66282, PARTITION_RECORD=72067, CONFIG_RECORD=17116, PRODUCER_IDS_RECORD=1, ACCESS_CONTROL_ENTRY_RECORD=1930}. The current metadata offset is now 332267 with an epoch of 19. Saw 36 brokers in the migrated metadata [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35].
> {code}
> This is a result of how we generate batches of records when traversing the ZK tree. Since we now using metadata transactions for the migration, we can re-batch these without any consistency problems.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)