[HDFS-87] NameNode startup fails if edit log terminates prematurely

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Not A Problem

Environment

~50 node cluster

• Assignee: Unassigned
• Reporter: Bryan Pendleton
• Created: 12/Dec/06 23:47
• Updated: 19/Feb/10 21:37
• Resolved: 19/Feb/10 20:45

Description

I ran out of space on the device that stores the edit log, resulting in an edit log that is truncated mid transaction.

Ideally, the NameNode should start up, in SafeMode or the like, whenever this happens. Right now, you get this stack trace:

2006-12-12 15:33:57,212 ERROR org.apache.hadoop.dfs.NameNode: java.io.EOFExcepti
on
at java.io.DataInputStream.readUnsignedShort(DataInputStream.java:310)
at org.apache.hadoop.io.UTF8.readFields(UTF8.java:104)
at org.apache.hadoop.dfs.FSEditLog.loadFSEdits(FSEditLog.java:227)
at org.apache.hadoop.dfs.FSImage.loadFSImage(FSImage.java:191)
at org.apache.hadoop.dfs.FSDirectory.loadFSImage(FSDirectory.java:320)
at org.apache.hadoop.dfs.FSNamesystem.<init>(FSNamesystem.java:226)
at org.apache.hadoop.dfs.NameNode.<init>(NameNode.java:146)
at org.apache.hadoop.dfs.NameNode.<init>(NameNode.java:138)
at org.apache.hadoop.dfs.NameNode.main(NameNode.java:589)

Issue Links

• is related to: HDFS-955 FSImage.saveFSImage can lose edits
• relates to: HADOOP-1188 processIOError() should update fstime file
• relates to: HDFS-209 Provide tool to view/change edits file