[
https://issues.apache.org/jira/browse/HADOOP-6566?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12833600#action_12833600
]
Steve Loughran commented on HADOOP-6566:
----------------------------------------
When I start up daemons I do create the directories on demand, code does the
following
# extracts a list of directories from the deployment config info
# goes through trying to create them, logging ones it can't
# fails if there aren't any with a useful error message that even lists the dirs
# returns that list of dirs in a format that can then be handled by normal Conf
parsers
# add that attr to the live graph of system state.
This lets me set up the directories on the fly and publishes that information.
Reviewing my code I see that I am not doing enough checks that the dirs really
are of type "directory", I must be assuming that the stuff later on can handle
that. Which they seem to, but their error messages don't include the directory
listings in " java.io.IOException: All specified directories are not accessible
or do not exist", which is a shame
{code}
/**
* Run through the directories, create all that are there
*
* @param dirs list of directories
* @param createDirs create the directories?
* @return the directories all converted to a list split by commas
*/
public static String createDirectoryList(Vector<String> dirs, boolean
createDirs)
throws FileNotFoundException {
StringBuilder path = new StringBuilder();
int failureCount = 0;
StringBuilder failureText = new StringBuilder();
for (String dir : dirs) {
File directory = new File(dir);
if (createDirs) {
if(!directory.mkdirs() && !directory.exists()) {
String message = "Failed to create directory " + directory;
LogFactory.getLog(HadoopComponentImpl.class.getName())
.warn(message);
failureCount++;
failureText.append(message).append("\n");
}
}
if (path.length() > 0) {
path.append(',');
}
path.append(directory.getAbsolutePath());
}
if (failureCount == dirs.size()) {
throw new FileNotFoundException("Failed to create any of " +
path.toString());
}
return path.toString();
}
{code}
> Hadoop daemons should not start up if the ownership/permissions on the
> directories used at runtime are misconfigured
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HADOOP-6566
> URL: https://issues.apache.org/jira/browse/HADOOP-6566
> Project: Hadoop Common
> Issue Type: New Feature
> Components: security
> Reporter: Devaraj Das
> Fix For: 0.22.0
>
>
> The Hadoop daemons (like datanode, namenode) should refuse to start up if the
> ownership/permissions on directories they use at runtime are misconfigured or
> they are not as expected. For example, the local directory where the
> filesystem image is stored should be owned by the user running the namenode
> process and should be only readable by that user. We can provide this feature
> in common and HDFS and MapReduce can use the same.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
