[ https://issues.apache.org/jira/browse/HIVE-20076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16534007#comment-16534007 ]
Gopal V commented on HIVE-20076: -------------------------------- The order by in the queries make it very hard to tell if the bug is happening or not, because this is about sequential numbering. {code} +{"writeid":### Masked writeid ###,"bucketid":536870912,"rowid":513} jessica garcia 59 3.52 20110926 +{"writeid":### Masked writeid ###,"bucketid":536870912,"rowid":6633} jessica garcia 59 3.69 20110925 {code} I think the fix to getRowNumber() is necessary, which is {code} if (rowInBatch >= batch.size) { + baseRow = super.getRowNumber(); + rowInBatch = 0; return super.nextBatch(theirBatch); } {code} This has a side-effect of reading batch.size again the next time around (if batch.size !=0, then the first batch will be repeated between every fast-path batch). Ideally, at that point it should reset the batch, if the batch.size is > 0 (the invariant is that it has already been consumed by rowInBatch). > Delete on a partitioned table removes more rows than expected > ------------------------------------------------------------- > > Key: HIVE-20076 > URL: https://issues.apache.org/jira/browse/HIVE-20076 > Project: Hive > Issue Type: Bug > Components: Transactions > Reporter: Teddy Choi > Assignee: Teddy Choi > Priority: Major > Attachments: HIVE-20076.2.patch, HIVE-20076.patch > > > Delete on a partitioned table removes more rows than expected -- This message was sent by Atlassian JIRA (v7.6.3#76005)