[
https://issues.apache.org/jira/browse/HIVE-5454?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Phabricator updated HIVE-5454:
------------------------------
Attachment: D13317.1.patch
QwertyManiac requested code review of "HIVE-5454 [jira] HCatalog runs a
partition listing with an empty filter".
Reviewers: JIRA
HIVE-5454. HCatalog runs a partition listing with an empty filter.
Modify the input format of HCat to load partitions always with a filter if
available.
This is a HCATALOG-527 caused regression, wherein the HCatLoader's way of
calling HCatInputFormat causes it to do 2x partition lookups - once without the
filter, and then again with the filter.
For tables with large number partitions (100000, say), the non-filter lookup
proves fatal both to the client ("Read timed out" errors from
ThriftMetaStoreClient cause the server doesn't respond) and to the server (too
much data loaded into the cache, OOME, or slowdown).
The fix would be to use a single call that also passes a partition filter
information, as was in the case of HCatalog 0.4 sources before HCATALOG-527.
(HCatalog-release-wise, this affects all 0.5.x users)
TEST PLAN
Built hcat module and ran all its existing tests
REVISION DETAIL
https://reviews.facebook.net/D13317
AFFECTED FILES
hcatalog/core/src/main/java/org/apache/hive/hcatalog/data/transfer/impl/HCatInputFormatReader.java
hcatalog/core/src/main/java/org/apache/hive/hcatalog/mapreduce/HCatInputFormat.java
hcatalog/core/src/test/java/org/apache/hive/hcatalog/mapreduce/HCatMapReduceTest.java
hcatalog/hcatalog-pig-adapter/src/main/java/org/apache/hive/hcatalog/pig/HCatLoader.java
MANAGE HERALD RULES
https://reviews.facebook.net/herald/view/differential/
WHY DID I GET THIS EMAIL?
https://reviews.facebook.net/herald/transcript/39321/
To: JIRA, QwertyManiac
> HCatalog runs a partition listing with an empty filter
> ------------------------------------------------------
>
> Key: HIVE-5454
> URL: https://issues.apache.org/jira/browse/HIVE-5454
> Project: Hive
> Issue Type: Bug
> Components: HCatalog
> Affects Versions: 0.12.0
> Reporter: Harsh J
> Attachments: D13317.1.patch
>
>
> This is a HCATALOG-527 caused regression, wherein the HCatLoader's way of
> calling HCatInputFormat causes it to do 2x partition lookups - once without
> the filter, and then again with the filter.
> For tables with large number partitions (100000, say), the non-filter lookup
> proves fatal both to the client ("Read timed out" errors from
> ThriftMetaStoreClient cause the server doesn't respond) and to the server
> (too much data loaded into the cache, OOME, or slowdown).
> The fix would be to use a single call that also passes a partition filter
> information, as was in the case of HCatalog 0.4 sources before HCATALOG-527.
> (HCatalog-release-wise, this affects all 0.5.x users)
--
This message was sent by Atlassian JIRA
(v6.1#6144)
