[HADOOP-3778] seek(long) in DFSInputStream should catch socket exception for retry later

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.16.0
• Fix Version/s: 0.19.0
• Hadoop Flags: Reviewed
• Assignee: Luo Ning
• Reporter: Luo Ning
• Created: 17/Jul/08 09:17
• Updated: 08/Jul/09 16:43
• Resolved: 22/Jul/08 21:36

Description

HADOOP-2346 introduced data read/write timeout. when data stream borken, DFSClient will retry in read/write methods, but no such mechanism found when seek(long) calling blockReader.skip(diff) (DFSClient.java #1582), will let IOException throw to application. i met NPE when using MapFile in hbase.

i'm supposing in the seek(long) method, let done be 'false' will causing retry (via 'blockEnd = -1'), a patch will attached later for review.