[
https://issues.apache.org/jira/browse/FLINK-10603?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16667287#comment-16667287
]
Piotr Nowojski commented on FLINK-10603:
----------------------------------------
I think the single largest overhead comes from This run time overhead comes
from this code:
{code:java}
// this means that this is either:
// (1) the first execution of this application
// (2) previous execution has failed before
first checkpoint completed
//
// in case of (2) we have to abort all previous
transactions
abortTransactions(transactionalIdsGenerator.generateIdsToAbort());
{code}
in
{{org.apache.flink.streaming.connectors.kafka.FlinkKafkaProducer011#initializeState}}
Commenting this line out single handedly speeds up the tests by 30-40%. Of
course we can not do that on the production, but maybe we could introduce
private/internal switch {{assumeFirstExecution}} or
{{forceNotAbortOldTransactions}} that we could use in all of the tests except
of the one test that is checking for this exact feature
{{FlinkKafkaProducer011ITCase#testScaleDownBeforeFirstCheckpoint}}.
But this wouldn't be a healthy "hack"...
Besides that one would have to time profile the most time consuming tests more
and try to figure out if we can speed them up.
> Reduce kafka test duration
> --------------------------
>
> Key: FLINK-10603
> URL: https://issues.apache.org/jira/browse/FLINK-10603
> Project: Flink
> Issue Type: Sub-task
> Components: Kafka Connector, Tests
> Affects Versions: 1.7.0
> Reporter: Chesnay Schepler
> Assignee: vinoyang
> Priority: Major
> Labels: pull-request-available
>
> The tests for the modern kafka connector take more than 10 minutes which is
> simply unacceptable.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
