[
https://issues.apache.org/jira/browse/HIVE-13403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15433573#comment-15433573
]
Eugene Koifman commented on HIVE-13403:
---------------------------------------
# AbstractRecordWriter.newBatch() - what is the point of the loop that does
_updaters.add(bucket, null)_. Why not create the ArrayList of the right size?
# TestStreaming.testBucketing() - This comment seems wrong now... _// no data
for bucket 3 -- expect 0 length bucket file_
## perhaps assert that there is no bucket #2 in actual1 collection?
# TestStreaming.testFileDumpCorruptDataFiles() - maybe _} else if
(file.contains("bucket_00002")) {_ should now be removed or perhaps in this
branch assertFalse() to indicate we don't expect this bucket?
otherwise LGTM
> Make Streaming API not create empty buckets (at least as an option)
> -------------------------------------------------------------------
>
> Key: HIVE-13403
> URL: https://issues.apache.org/jira/browse/HIVE-13403
> Project: Hive
> Issue Type: Bug
> Components: HCatalog, Transactions
> Affects Versions: 1.3.0
> Reporter: Eugene Koifman
> Assignee: Wei Zheng
> Priority: Critical
> Attachments: HIVE-13403.1.patch, HIVE-13403.2.patch,
> HIVE-13403.3.patch, HIVE-13403.4.patch
>
>
> as of HIVE-11983, when a TransactionBatch is opened in StreamingAPI, a full
> compliment of bucket files (AbstractRecordWriter.createRecordUpdaters()) is
> created on disk even though some may end up receiving no data.
> It would be better to create them on demand and not clog the FS.
> Tez can handle missing (empty) buckets and on MR bucket join algorithms will
> check if all buckets are there and bail out if not.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
