[
https://issues.apache.org/jira/browse/PIG-4438?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
liyunzhang_intel updated PIG-4438:
----------------------------------
Attachment: PIG-4438_2.patch
following 5 unit tests failures are deleted after using this patch:
org.apache.pig.test.TestEvalPipeline2.testLimitAfterSort
org.apache.pig.test.TestEvalPipeline2.testLimitAutoReducer
org.apache.pig.test.TestEvalPipeline2.testLimitedSortWithDump
org.apache.pig.test.TestEvalPipeline2.testLimitAfterSortDesc
org.apache.pig.test.TestLimitAdjuster.simpleTest
Currently, we use RDD# sortByKey to implement this sort feature in spark:
see SortConverter#convert
in MR:
the implementation is:
1. Sample the input data get a sample data
2. Generate quantile file with the sample data. Quantile file contains the
instance of WeightedRangePartitioner.
3. MR uses WeightedRangePartitioner to guarantee the global sort.(Like
TotalOrderPatitioner:
https://books.google.com/books?id=H3mvcxPeUfwC&pg=PA151&lpg=PA151&dq=TotalOrderPartitioner+mapreduce&source=bl&ots=pYrOwd-Fse&sig=Kx3DZ8UBLFNveqepjZefSjyk65s&hl=zh-CN&sa=X&ei=q_MZVbfpIcmLyAS4_IHYCQ&ved=0CFwQ6AEwCTgK#v=onepage&q=TotalOrderPartitioner%20mapreduce&f=false)
Do we need implement sort like what it does in MR?
> Can not work when in "limit after sort" situation in spark mode
> ---------------------------------------------------------------
>
> Key: PIG-4438
> URL: https://issues.apache.org/jira/browse/PIG-4438
> Project: Pig
> Issue Type: Sub-task
> Components: spark
> Reporter: liyunzhang_intel
> Assignee: liyunzhang_intel
> Fix For: spark-branch
>
> Attachments: PIG-4438_1.patch, PIG-4438_2.patch
>
>
> when pig script executes "order" before "limit" in spark mode, the results
> will be wrong.
> cat testlimit.txt
> 1 orange
> 3 coconut
> 5 grape
> 6 pear
> 2 apple
> 4 mango
> testlimit.pig:
> a = load './testlimit.txt' as (x:int, y:chararray);
> b = order a by x;
> c = limit b 1;
> store c into './testlimit.out';
> the result:
> 1 orange
> 2 apple
> 3 coconut
> 4 mango
> 5 grape
> 6 pear
> the correct result should be:
> 1 orange
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
