[
https://issues.apache.org/jira/browse/PIG-1831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12988910#comment-12988910
]
Daniel Dai commented on PIG-1831:
---------------------------------
This issue is caused by race condition in using static variable
PigMapReduce.sJobConf in local mode. In local mode, all mapreduce job share a
single VM. Pig keep on overwriting static variable PigMapReduce.sJobConf each
time we launch a new mapreduce job. When multiple mapreduce jobs launching
simultaneously, one mapreduce job may use config for other mapreduce job, and
cause indeterministic behavior. Options to fix this issue are:
1. force local mode run mapreduce job sequentially, if there is a way
2. Make sJobConf an array keyed by mapreduce jobid. However, some UDFs is using
sJobConf, we could break backward compatibility
> Indeterministic behavior in local mode due to static variable
> PigMapReduce.sJobConf
> -----------------------------------------------------------------------------------
>
> Key: PIG-1831
> URL: https://issues.apache.org/jira/browse/PIG-1831
> Project: Pig
> Issue Type: Bug
> Affects Versions: 0.8.0
> Reporter: Vivek Padmanabhan
> Assignee: Daniel Dai
>
> The below script when run in local mode gives me a different output. It looks
> like in local mode I have to store a relation obtained through streaming in
> order to use it afterwards.
> For example consider the below script :
> DEFINE MySTREAMUDF `test.sh`;
> A = LOAD 'myinput' USING PigStorage() AS (myId:chararray, data2, data3,data4
> );
> B = STREAM A THROUGH MySTREAMUDF AS (wId:chararray, num:int);
> --STORE B into 'output.B';
> C = JOIN B by wId LEFT OUTER, A by myId;
> D = FOREACH C GENERATE B::wId,B::num,data4 ;
> D = STREAM D THROUGH MySTREAMUDF AS (f1:chararray,f2:int);
> --STORE D into 'output.D';
> E = foreach B GENERATE wId,num;
> F = DISTINCT E;
> G = GROUP F ALL;
> H = FOREACH G GENERATE COUNT_STAR(F) as TotalCount;
> I = CROSS D,H;
> STORE I into 'output.I';
> test.sh
> ---------
> #/bin/bash
> cut -f1,3
> And input is
> abcd label1 11 feature1
> acbd label2 22 feature2
> adbc label3 33 feature3
> Here if I store relation B and D then everytime i get the result :
> acbd 3
> abcd 3
> adbc 3
> But if i dont store relations B and D then I get an empty output. Here again
> I have observed that this behaviour is random ie sometimes like 1out of 5
> runs there will be output.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
