[
https://issues.apache.org/jira/browse/APEXMALHAR-2493?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16015556#comment-16015556
]
ASF GitHub Bot commented on APEXMALHAR-2493:
--------------------------------------------
GitHub user chaithu14 opened a pull request:
https://github.com/apache/apex-malhar/pull/622
APEXMALHAR-2493 Fixed the issue of KafkaSinglePortExactlyOnceOutputOperator
going to the blocked state during recovery
@sandeshh @tushargosavi Please review and merge.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/chaithu14/incubator-apex-malhar
APEXMALHAR-2493-KafkaExactlyCBBug
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/apex-malhar/pull/622.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 #622
----
commit c784f4da46d1cf594aa4156135b9c196aa66d931
Author: chaitanya <chai...@apache.org>
Date: 2017-05-18T10:37:52Z
APEXMALHAR-2493 Fixed the issue of KafkaSinglePortExactlyOnceOutputOperator
going to the blocked state during recovery
----
> KafkaSinglePortExactlyOnceOutputOperator going to the blocked state during
> recovery
> -----------------------------------------------------------------------------------
>
> Key: APEXMALHAR-2493
> URL: https://issues.apache.org/jira/browse/APEXMALHAR-2493
> Project: Apache Apex Malhar
> Issue Type: Bug
> Reporter: Chaitanya
> Assignee: Chaitanya
>
> Steps to reproduce the issue:
> ---------------------------------------
> - Created the Kafka topic with single partition.
> - Created the application with the following DAG:
> BatchSequenceGenerator -> KafkaSinglePortExactlyOnceOutputOperator
> # of partitions of KafkaSinglePortExactlyOnceOutputOperator = 2.
> Let's say KO1, KO2 are the two instances.
> - Launched the app, after some time, manually killed the one of the instance
> of "KafkaSinglePortExactlyOnceOutputOperator" operator(KO2).
> - During recovery, the instance comes up and after some time, it goes to the
> blocked state. App master killed this instance.
> Observation:
> ----------------
> * There is an infinite while loop in rebuildPartialWindow() method.
> * While loop will break on the below 2 conditions:
> a) # of trails for "polled records from Kafka is empty" = 10
> b) Crossed boundary (consumerRecord.offset() >= currentOffset)
> In this scenario, KO1 keeps on writing the data to Kafka. So, the first
> condition will not satisfy.
> Operator is not checking the 2nd condition because of the below continue
> statement:
> if (!doesKeyBelongsToThisInstance(operatorId,
> consumerRecord.key())) {
> continue;
> }
> Solution: First check the cross boundary condition and then check the
> doesKeyBelongsToThisInstance(..).
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
