[HIVE-6979] Hadoop-2 test failures related to quick stats not being populated correctly

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.14.0
• Fix Version/s: 0.14.0
• Assignee: Prasanth Jayachandran
• Reporter: Prasanth Jayachandran
• Created: 26/Apr/14 00:31
• Updated: 13/Nov/14 19:44
• Resolved: 28/Apr/14 15:41

Description

The test failures that are currently reported by Hive QA running on hadoop-2 (https://issues.apache.org/jira/browse/HIVE-6968?focusedCommentId=13980570&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13980570) are related to difference in the way hadoop FileSystem.globStatus() api behaves. For a directory structure like below

dir1/file1
dir1/file2

Two level of path pattern like dir1// will return both files in hadoop 1.x but will return empty result in hadoop 2.x (in fact it will say no such file or directory and return empty file status array). Hadoop 2.x seems to be compliant to linux behaviour (ls dir1//) but hadoop 1.x is not.

As a result of this, the fast statistics (NUM_FILES and TOTAL_SIZE) are populated wrongly causing diffs in qfile tests for hadoop-1 and hadoop-2.

Issue Links

• links to: Review Board