hadoop-hdfs-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Todd Lipcon (JIRA)" <j...@apache.org>
Subject [jira] Created: (HDFS-1059) completeFile loops forever if the block's only replica has become corrupt
Date Mon, 22 Mar 2010 01:45:27 GMT
completeFile loops forever if the block's only replica has become corrupt
-------------------------------------------------------------------------

                 Key: HDFS-1059
                 URL: https://issues.apache.org/jira/browse/HDFS-1059
             Project: Hadoop HDFS
          Issue Type: Bug
    Affects Versions: 0.21.0, 0.22.0
            Reporter: Todd Lipcon


If a writer is appending to a block with replication factor 1, and that block has become corrupt,
a reader will report the corruption to the NN. Then when the writer tries to complete the
file, it will loop forever with an error like:

    [junit] 2010-03-21 17:40:08,093 INFO  namenode.FSNamesystem (FSNamesystem.java:checkFileProgress(1613))
- BLOCK* NameSystem.checkFileProgress: block blk_-4256412191814117589_1001{blockUCState=COMMITTED,
primaryNodeIndex=-1, replicas=[ReplicaUnderConstruction[127.0.0.1:56782|RBW]]} has not reached
minimal replication 1
    [junit] 2010-03-21 17:40:08,495 INFO  hdfs.DFSClient (DFSOutputStream.java:completeFile(1435))
- Could not complete file /TestReadWhileWriting/file1 retrying...

Should add tests that cover the case of a writer appending to a block that is corrupt while
a reader accesses it.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message