[
https://issues.apache.org/jira/browse/PIG-4542?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14535769#comment-14535769
]
Mohit Sabharwal commented on PIG-4542:
--------------------------------------
FYI [~kellyzly]. [~praveenr019], [~xuefuz], this patch:
- Gets rid of the use of RDD.count() in CollectedGroupConverter and
StreamConverter.
- Adds an abstract method endInput() which is executed before we call
getNextTuple() for the last time.
- Deletes POStreamSpark since it was just handling the last record.
- Removes special code in POCollectedGroup to handle the last record.
- While I was here, renamed POOutputConsumerIterator to OutputConsumerIterator.
The "PO" prefix should only be used for physical operators.
> OutputConsumerIterator should flush buffered records
> ----------------------------------------------------
>
> Key: PIG-4542
> URL: https://issues.apache.org/jira/browse/PIG-4542
> Project: Pig
> Issue Type: Sub-task
> Components: spark
> Affects Versions: spark-branch
> Reporter: Mohit Sabharwal
> Assignee: Mohit Sabharwal
> Fix For: spark-branch
>
> Attachments: PIG-4542.patch
>
>
> Certain operators may buffer the output. We need to flush the last set of
> records from such operators, when we encounter the last input record, before
> calling getNextTuple() for the last time.
> Currently, to flush the last set of records, we compute RDD.count() and
> compare the count with a running counter to determine if we have reached the
> last record. This is an unnecessary and inefficient.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
