[ https://issues.apache.org/jira/browse/PIG-2652?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13256871#comment-13256871 ]
Dmitriy V. Ryaboy commented on PIG-2652: ---------------------------------------- Ok I have a much cleaner version of the patch, but it still suffers from a critical flaw -- the sampler needs to know the number of buckets to get samples for, and if it's not explicitly specified, and the SampleOptimizer stuff doesn't kick in, it still gets a single bucket to "distribute" skewed keys between. I think we have to go ahead with the approach of always sampling for a large number of buckets (10000?), and distributing them evenly in the partitioner. Daniel -- comments? Will attach my work-in-progress patch. > Skew join and order by don't trigger reducer estimation > ------------------------------------------------------- > > Key: PIG-2652 > URL: https://issues.apache.org/jira/browse/PIG-2652 > Project: Pig > Issue Type: Bug > Reporter: Bill Graham > Assignee: Bill Graham > Fix For: 0.10.0, 0.9.3, 0.11 > > Attachments: PIG-2652_1.patch, PIG-2652_2.patch, PIG-2652_3.patch, > PIG-2652_3_10.patch > > > If neither PARALLEL, default parallel or {{mapred.reduce.tasks}} are set, the > number of reducers is not estimated based on input size for skew joins or > order by. Instead, these jobs get only 1 reducer. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira