Workaround for running Hadoop 1.0.x on Windows (HADOOP-7682)
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.

Workaround for HADOOP-7682 on Windows: taskTracker could not start because "Failed to set permissions" to "ttprivate to 0700"

This simple patch for Hadoop 1.0.3 should allow you to avoid the longstanding, dreaded file permission exceptions when running Hadoop on Windows as described in issue HADOOP-7682. It was [suggested by Joshua Caplan] ( as a simpler alternative to shaving a yak.

The patch comprises a single class with two overridden methods that ignore IOExceptions when trying to set file permissions. Since these potentially important exceptions are swallowed, I don't recommend running this patch in production on Windows, or in an environment that supports setting full file permissions, like Linux or Mac OS X.


The patch allows getting past the file permission exceptions when running on Windows, but doesn't change the fact that Hadoop must be run using Cygwin. It also doesn't address any of the other potential issues that may arise when running on Windows, but it seems to work like a charm for single-node job development and testing.

Note, the patch installation described below only appears to work when running Hadoop in standalone mode, which is the norm for job development and testing. If you've configured Hadoop to run in pseudo-distributed or fully distributed mode on Windows (i.e. "production" mode), there is some difference in the way child processes are spawned that won't include the patch in the child VMs' classpaths. As a result, I strongly recommend running only in standalone mode on Windows for testing (with the patch and its unknown side effects), and running Hadoop on a fully-supported platform (without the patch) for production jobs.

If you absolutely must run in pseudo-distributed or fully distributed mode on Windows and the patch installed according to the instructions below doesn't work, you could instead try putting the patch JAR in the JDK's ${JAVA_HOME}/jre/lib/ext directory, which includes it in the classpath for all Java VMs. I haven't tried this myself, so YMMV.

Usage instructions

  1. Download the pre-built JAR, patch-hadoop_7682-1.0.x-win.jar, from the Downloads section (or build it yourself).

  2. Copy patch-hadoop_7682-1.0.x-win.jar to the ${HADOOP_HOME}/lib directory

  3. Modify ${HADOOP_HOME}/conf/core-site.xml to enable the overriden implementation as shown below:

     <?xml version="1.0"?>
     <?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
     <!-- Put site-specific property overrides in this file. -->
     		<description>Enables patch for issue HADOOP-7682 on Windows</description>
  4. Run your job as usual (using Cygwin). You should see some logging to System.err if the patch is enabled and working.

Build instructions

This project is a NetBeans 7.x Ant-based project, which means you can also build it on the command line with a modern version of Ant.

The project was built for Hadoop 1.0.3. At minimum, you will need to specify the ${var.hadoop.home} property in your Ant environment in order to build. If you're running a different version of Hadoop, you can change the Hadoop Core JAR file name in nbbuild/ as the value of the file.reference.hadoop-core-1.0.3.jar key (you shouldn't change the key name itself, as it's merely symbolic and arbitrary).