[HIVE-22936] NPE in SymbolicInputFormat

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.1.2
• Component/s: File Formats
• Labels: pull-request-available
• Flags: Patch
• Assignee: Unassigned
• Reporter: Redis Liu
• Created: 26/Feb/20 23:43
• Updated: 06/Jun/20 00:24

Description

Symptom

I was running Hive over AWS S3 Inventory Report, which uses SymlinkTextInputFormat, and symlink file content is the FQDN S3 URL of each s3 file, like :

s3://inventory-bucket/bucket1/2020-02-04-11-23-00-AEFEDDCE
s3://inventory-bucket/bucket1/2020-02-05-11-23-00-BCEDCCDD

When I have the following setting:

set hive.rework.mapredwork=true;  

The job fails with NullPointException, without stack trace.

Cause

The content of symlink may be arbitrary full qualified FS path, while SymbolicInputFormat uses the default FS instance to get the status of the data files, which fails (and returns null) when the schema of data file differs from Hive's default FS.

Code point:
https://github.com/apache/hive/blob/cfc12f05f0c034f9aad149960e58d40902e0dcfe/ql/src/java/org/apache/hadoop/hive/ql/io/SymbolicInputFormat.java#L78

              // "fileSystem" may not be able to list status for given file uri.
              FileStatus[] matches = fileSystem.globStatus(new Path(line));

Fix

Please check attached npe-symbolic-inputformat.patch

 

Issue Links

• links to: GitHub Pull Request #925