Use a thread-local rather than static ENCODER/DECODER variables in Text for synchronization

• Key: HADOOP-3863
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Critical
• Assignee: Arun C Murthy
• Reporter: Arun C Murthy
• Affects Version/s: 0.18.0
• Fix Version/s: 0.19.0

Description

Text.{ENCODER|DECODER} are static variables which need to be synchronized in Text.{read|write}String; given that lots of RPC code calls Text.{read|write}String the contention for this lock shows up very significantly on large clusters.

Comments

[Arun C Murthy]

Fixed Text to use ThreadLocal encoders/decoders.

[Hadoop QA]

-1 overall. Here are the results of testing the latest attachment
http://issues.apache.org/jira/secure/attachment/12387218/HADOOP-3863_0_20080730.patch
against trunk revision 680823.

+1 @author. The patch does not contain any @author tags.

+1 tests included. The patch appears to include 3 new or modified tests.

+1 javadoc. The javadoc tool did not generate any warning messages.

+1 javac. The applied patch does not increase the total number of javac compiler warnings.

+1 findbugs. The patch does not introduce any new Findbugs warnings.

+1 release audit. The applied patch does not increase the total number of release audit warnings.

-1 core tests. The patch failed core unit tests.

+1 contrib tests. The patch passed contrib unit tests.

Test results: http://hudson.zones.apache.org/hudson/job/Hadoop-Patch/2985/testReport/
Findbugs warnings: http://hudson.zones.apache.org/hudson/job/Hadoop-Patch/2985/artifact/trunk/build/test/findbugs/newPatchFindbugsWarnings.html
Checkstyle results: http://hudson.zones.apache.org/hudson/job/Hadoop-Patch/2985/artifact/trunk/build/test/checkstyle-errors.html
Console output: http://hudson.zones.apache.org/hudson/job/Hadoop-Patch/2985/console

This message is automatically generated.

[Owen O'Malley]

This looks good, but I'd suggest renaming to be {DE,EN}CODER_FACTORY to make it clear that each thread will get a different one.

[Arun C Murthy]

Renamed the variable, same patch.

The test failure (org.apache.hadoop.hdfs.TestFileAppend2.testComplexAppend) is unrelated.

[Doug Cutting]

I filed HADOOP-3874 to address this. (We should file bugs whenever unrelated tests fail, no?)

[Tsz Wo (Nicholas), SZE]

> The test failure (org.apache.hadoop.hdfs.TestFileAppend2.testComplexAppend) is unrelated.

I am not sure whether it is unrelated since this patch changes a core class and this is the first time that TestFileAppend2.testComplexAppend fails. TestFileAppend2.testComplexAppend create a number of threads appending files concurrently. It seems to me that it might be related. Could we try Hudson one more time?

[Owen O'Malley]

I just committed this. Thanks, Arun!

[Tsz Wo (Nicholas), SZE]

According to CharsetDecoder javadoc http://java.sun.com/javase/6/docs/api/java/nio/charset/CharsetDecoder.html, there is a specific sequence of decoding operations. In particular, we should call reset() before using the decoder. It seems we are not following it.

[Arun C Murthy]

Fair point, Doug - taken; although it seems TestFileAppend2 works consistently on my machine.

Nicholas - we are are using http://java.sun.com/javase/6/docs/api/java/nio/charset/CharsetDecoder.html#decode(java.nio.ByteBuffer) which does the whole 'decoding' operation, ditto for encoding.

[Tsz Wo (Nicholas), SZE]

> Nicholas - we are are using http://java.sun.com/javase/6/docs/api/java/nio/charset/CharsetDecoder.html#decode(java.nio.ByteBuffer) which does the whole 'decoding' operation, ditto for encoding.

Thanks Arun, I missed this.

BTW, TestFileAppend2 also works consistently in my Windows and Linux machines.

[Hudson]

Integrated in Hadoop-trunk #581 (See http://hudson.zones.apache.org/hudson/job/Hadoop-trunk/581/)