hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Giovanni Matteo Fumarola (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-16282) Avoid FileStream to improve performance
Date Fri, 03 May 2019 18:18:00 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-16282?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16832717#comment-16832717

Giovanni Matteo Fumarola commented on HADOOP-16282:

Thanks [~daryn] for your feedback.
I think we should fix the potential leaks (they are even present in the original code) instead
of reverting the patch.

InputStream in = null;
OutputStream out =null;
try {
  in = Files.newInputStream(src.toPath());
  out = dstFS.create(dst);
  IOUtils.copyBytes(in, out, conf);
} catch (IOException e) {
  IOUtils.closeStream( out );
  IOUtils.closeStream( in );
  throw e;
The variable in and out can leak resources/handles. I am pretty aware of this problem since
I hit it in production.

We should add the finally clause:
} catch (){
} finally {
or use a Try-with-resources clause:
 try InputStream in = Files.newInputStream(src.toPath()); etc..){{code}

> Avoid FileStream to improve performance
> ---------------------------------------
>                 Key: HADOOP-16282
>                 URL: https://issues.apache.org/jira/browse/HADOOP-16282
>             Project: Hadoop Common
>          Issue Type: Improvement
>            Reporter: Ayush Saxena
>            Assignee: Ayush Saxena
>            Priority: Major
>             Fix For: 3.3.0
>         Attachments: HADOOP-16282-01.patch, HADOOP-16282-02.patch
> The FileInputStream and FileOutputStream classes contains a finalizer method which will
cause garbage collection pauses. See [JDK-8080225|https://bugs.openjdk.java.net/browse/JDK-8080225]
for details.
> The FileReader and FileWriter constructors instantiate FileInputStream and FileOutputStream,
again causing garbage collection issues while finalizer methods are called.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org

View raw message