[
https://issues.apache.org/jira/browse/HADOOP-5859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12713894#action_12713894
]
Kan Zhang commented on HADOOP-5859:
-----------------------------------
> I just checked, the JDK6 implementation of LinkedList didn't declare the
> "size" variable volatile. So Todd is right, a second thread may see a stale
> cached value
I take it back. Actually, in this particular case the writeChunk() thread will
be able to see the fresh value of ackQueue size whenever it's been modified.
The reason is that writeChunk() reads ackQueue size while holding the dataQueue
lock and waits on the dataQueue lock in a loop when condition is not satisfied.
There are 2 threads that may modify ackQueue size. One is DataStreamer, who
always holds both dataQueue and ackQueue locks when modifying ackQueue. So any
changes to ackQueue by DataStreamer will be visible to writeChunk() thread. The
other thread that modifies ackQueue is ResponseProcessor. It may modify
ackQueue while holding only the ackQueue lock. However, immediately after
ackQueue modification, it grabs the dataQueue lock and calls
dataQueue.notifyAll() to wake up the writeChunk() thread. So when the
writeChunk() thread grabs the dataQueue lock afterwards, the fresh value of
ackQueue size will be visible. Make sense?
> FindBugs : fix "wait() or sleep() with locks held" warnings in hdfs
> -------------------------------------------------------------------
>
> Key: HADOOP-5859
> URL: https://issues.apache.org/jira/browse/HADOOP-5859
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Reporter: Kan Zhang
> Assignee: Kan Zhang
> Attachments: 5859-21.patch, 5859-22.patch, 5859-26.patch,
> 5859-33.patch, 5859-4.patch, 5859-5.patch, 5859-8.patch
>
>
> This JIRA fixes the following warnings:
> SWL org.apache.hadoop.hdfs.DFSClient$DFSOutputStream.closeInternal() calls
> Thread.sleep() with a lock held
> TLW wait() with two locks held in
> org.apache.hadoop.hdfs.DFSClient$DFSOutputStream.flushInternal()
> TLW wait() with two locks held in
> org.apache.hadoop.hdfs.DFSClient$DFSOutputStream.flushInternal()
> TLW wait() with two locks held in
> org.apache.hadoop.hdfs.DFSClient$DFSOutputStream.writeChunk(byte[], int, int,
> byte[])
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
