[
https://issues.apache.org/jira/browse/KAFKA-5241?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16010417#comment-16010417
]
ASF GitHub Bot commented on KAFKA-5241:
---------------------------------------
GitHub user twbecker opened a pull request:
https://github.com/apache/kafka/pull/3054
KAFKA-5241: GlobalKTable does not checkpoint offsets after restoring state
Ensure checkpointable offsets for GlobalKTables are always written on close.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/twbecker/kafka KAFKA-5241
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/3054.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #3054
----
commit 4f7263768ef73d18cf6b90894f4a0286bc79ea14
Author: Tommy Becker <tobec...@tivo.com>
Date: 2017-05-15T12:13:07Z
Fix KAFKA-5241.
Ensure checkpointable offsets for GlobalKTables are always written on close.
----
> GlobalKTable does not checkpoint offsets after restoring state
> --------------------------------------------------------------
>
> Key: KAFKA-5241
> URL: https://issues.apache.org/jira/browse/KAFKA-5241
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Affects Versions: 0.10.2.1
> Reporter: Tommy Becker
> Priority: Minor
>
> I'm experimenting with an application that uses a relatively large
> GlobalKTable, and noticed that streams was not checkpointing its offsets on
> close(). This is because although
> {{org.apache.kafka.streams.processor.internals.GlobalStateManagerImpl#restoreState}}
> updates the checkpoint map, the actual checkpointing itself is guarded by a
> check that the offsets passed from the {{GloablStateUpdateTask}} are not
> empty. This is frustrating because if the topic backing the global table is
> both large (therefore taking a long time to restore) and infrequently
> written, then streams rebuilds the table from scratch every time the
> application is started.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
