[
https://issues.apache.org/jira/browse/TEZ-2454?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Siddharth Seth updated TEZ-2454:
--------------------------------
Attachment: TEZ-2454.3.txt
Thanks for the review. Updating revised patch with fixes.
bq. FetcherOrderedGrouped - MapHost would not be null in fetchNext()?
Fixed.
bq. "count" in ShuffleSchedulerCallable should be incremented in inner loop?,
otherwise it might end up scheduling continuously.
Good catch. Fixed and added a unit test to verify this.
bq. FetcherOrderedGrouped- httpConnection need not be volatile?
bq. HttpConnection - connection, DataInputStream, connectionSucceeed need not
be volatile?
Cleanup / close can be called from a different thread, which is why these need
to be volatile.
Added the findbugs warning to the ignore list. It isn't related since
isShutdown state change happens via an AtomicBoolean.
> Change FetcherOrderedGroup to work as Callables instead of blocking threads
> ---------------------------------------------------------------------------
>
> Key: TEZ-2454
> URL: https://issues.apache.org/jira/browse/TEZ-2454
> Project: Apache Tez
> Issue Type: Improvement
> Reporter: Siddharth Seth
> Assignee: Siddharth Seth
> Attachments: TEZ-2454.1.wip.txt, TEZ-2454.2.txt, TEZ-2454.3.txt
>
>
> The Fetcher threads for Ordered Input currently run and block till merge
> completes, which makes it difficult to use them via ThreadPools.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
