[zebra] Data sanity check should be performed at the end of writing instead of later at query time ---------------------------------------------------------------------------------------------------
Key: PIG-1207 URL: https://issues.apache.org/jira/browse/PIG-1207 Project: Pig Issue Type: Improvement Reporter: Yan Zhou Currently the equity check of number of rows across different column groups are performed by the query. And the error info is sketchy and only emits a "Column groups are not evenly distributed", or worse, throws an IndexOufOfBound exception from CGScanner.getCGValue since BasicTable.atEnd and BasicTable.getKey, which are called just before BasicTable.getValue, only checks the first column group in projection and any discrepancy of the number of rows per file cross multiple column groups in projection could have BasicTable.atEnd return false and BasicTable.getKey return a key normally but another column group already exaust its current file and the call to its CGScanner.getCGValue throw the exception. This check should also be performed at the end of writing and the error info should be more informational. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.