[
https://issues.apache.org/jira/browse/TEZ-3723?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kuhu Shukla updated TEZ-3723:
-----------------------------
Attachment: TEZ-3723.004.patch
bq. unless this code is trying to ensure the old behavior is still valid.
That is what I had in mind since the the change in
TestUnorderedPartitionedKVWriter will cover the changed scenario(rawlength = 0
and < 6). Currently Ordered case empty partitions are 6 bytes in length and
this test covers that original behavior(rawlength =6).
> TezIndexRecord#hasData() returns true for empty index record in the Unordered
> case
> ----------------------------------------------------------------------------------
>
> Key: TEZ-3723
> URL: https://issues.apache.org/jira/browse/TEZ-3723
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Kuhu Shukla
> Assignee: Kuhu Shukla
> Attachments: TEZ-3723.001.patch, TEZ-3723.002.patch,
> TEZ-3723.003.patch, TEZ-3723.004.patch
>
>
> In the case of Unordered writer, use of {{hasData()}} returns true for empty
> partitions since the rawlength is zero (as writer skips writing them out
> altogether) and the check in hasData will fail as {{rawlength ==
> header.length +2}} fails.
> This currently works for the ordered case(and the method is used here unlike
> in unordered writer) but after TEZ-3605, it will fail for both scenarios.
> Even if we don't use the method, the logic should be fixed.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
