hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kan Zhang (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HDFS-637) DataNode sends an Success ack when block write fails
Date Tue, 22 Sep 2009 21:44:16 GMT

    [ https://issues.apache.org/jira/browse/HDFS-637?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12758456#action_12758456

Kan Zhang commented on HDFS-637:

I like Raghu's suggestion, which is to simply set a boolean flag in the catch clause (the
clause has to be moved to the outset) and add a checking of the flag to the if (Thread.interrupted())
{} block. That way the exit logic is easier to understand (with the comments there). Having
the exit logic in one place also facilitates adding further checking on the reason of the
interrupt (i.e., local error or downstream error).

> DataNode sends an Success ack when block write fails
> ----------------------------------------------------
>                 Key: HDFS-637
>                 URL: https://issues.apache.org/jira/browse/HDFS-637
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: data-node
>            Reporter: Hairong Kuang
>            Assignee: Hairong Kuang
>            Priority: Blocker
>             Fix For: 0.21.0
>         Attachments: interrupted.patch
> When I work on HDFS-624, I saw TestFileAppend3#TC7 occasionally fails. After lots of
debug, I saw that the client unexpected received a response of "-2 SUCCESS SUCCESS" in which
-2 is the packet sequence number. This happened in a pipeline of 2 datanodes and one of them
failed. It turned out when block receiver fails, it shuts down itself and interrupts the packet
responder but responder tries to handle interruption with the condition "Thread.isInterrupted()"
but unfortunately a thread's interrupt status is not set in some cases as explained in the
Thread#interrupt javadoc:
>  If this thread is blocked in an invocation of the wait(), wait(long), or wait(long,
int) methods of the Object  class, or of the join(), join(long), join(long, int), sleep(long),
or sleep(long, int), methods of this class, then its interrupt status will be cleared and
it will receive an InterruptedException. 
> So datanode does not detect the interruption and continues as if no error occurs.

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

View raw message