[
https://issues.apache.org/jira/browse/HADOOP-1513?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12506989
]
Devaraj Das commented on HADOOP-1513:
-------------------------------------
Dhruba, I was looking at the 'if' clause in conjunction with the exception
that is thrown (if the expression returns true).
The idea behind the if clause,
if (!dir.exists() && !dir.mkdirs()),
is to first check whether the directory exists, and, if not, create it. If the
creation fails, exception is thrown.
I think breaking the if clause into two parts solves the problem in the context
of its usage. If a race condition ever occurs, it will be this way - the first
process will create the dir successfully. The second process will not be able
to do so (inside the OS kernel, things will be atomic). In the
DiskChecker.checkDir method's context, things will still work - we will throw
an exception only when we don't see the directory (we really don't need to care
who created the directory). So, yes, the reason for throwing the exception is
different, but IMO it is consistent overall. There cannot be a race condition
in the exists( ) check since the kernel provided the atomicity in the directory
creation.
BTW, there are some more checks done afterwards in the method
(readable/writable checks). Those will rule out permission issues to do with
processes from different users competing with each other to create the dir (we
will bail out if we discover the dir is not writable/readable).
> A likely race condition between the creation of a directory and checking for
> its existence in the DiskChecker class
> -------------------------------------------------------------------------------------------------------------------
>
> Key: HADOOP-1513
> URL: https://issues.apache.org/jira/browse/HADOOP-1513
> Project: Hadoop
> Issue Type: Bug
> Components: fs
> Affects Versions: 0.14.0
> Reporter: Devaraj Das
> Assignee: Devaraj Das
> Priority: Critical
> Fix For: 0.14.0
>
> Attachments: 1513.patch
>
>
> Got this exception in a job run. It looks like the problem is a race
> condition between the creation of a directory and checking for its existence.
> Specifically, the line:
> if (!dir.exists() && !dir.mkdirs()), doesn't seem safe when invoked by
> multiple processes at the same time.
> 2007-06-21 07:55:33,583 INFO org.apache.hadoop.mapred.MapTask:
> numReduceTasks: 1
> 2007-06-21 07:55:33,818 WARN org.apache.hadoop.fs.AllocatorPerContext:
> org.apache.hadoop.util.DiskChecker$DiskErrorException: can not create
> directory: /export/crawlspace/kryptonite/ddas/dfs/data/tmp
> at org.apache.hadoop.util.DiskChecker.checkDir(DiskChecker.java:26)
> at
> org.apache.hadoop.fs.LocalDirAllocator$AllocatorPerContext.createPath(LocalDirAllocator.java:211)
> at
> org.apache.hadoop.fs.LocalDirAllocator$AllocatorPerContext.getLocalPathForWrite(LocalDirAllocator.java:248)
> at
> org.apache.hadoop.fs.LocalDirAllocator$AllocatorPerContext.createTmpFileForWrite(LocalDirAllocator.java:276)
> at
> org.apache.hadoop.fs.LocalDirAllocator.createTmpFileForWrite(LocalDirAllocator.java:155)
> at
> org.apache.hadoop.dfs.DFSClient$DFSOutputStream.newBackupFile(DFSClient.java:1171)
> at org.apache.hadoop.dfs.DFSClient$DFSOutputStream.(DFSClient.java:1136)
> at org.apache.hadoop.dfs.DFSClient.create(DFSClient.java:342)
> at
> org.apache.hadoop.dfs.DistributedFileSystem$RawDistributedFileSystem.create(DistributedFileSystem.java:145)
> at
> org.apache.hadoop.fs.ChecksumFileSystem$FSOutputSummer.(ChecksumFileSystem.java:368)
> at
> org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:443)
> at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:254)
> at org.apache.hadoop.io.SequenceFile$Writer.(SequenceFile.java:675)
> at org.apache.hadoop.io.SequenceFile.createWriter(SequenceFile.java:165)
> at
> org.apache.hadoop.examples.RandomWriter$Map.map(RandomWriter.java:137)
> at org.apache.hadoop.mapred.MapRunner.run(MapRunner.java:46)
> at org.apache.hadoop.mapred.MapTask.run(MapTask.java:189)
> at
> org.apache.hadoop.mapred.TaskTracker$Child.main(TaskTracker.java:1740)
> 2007-06-21 07:55:33,821 WARN org.apache.hadoop.mapred.TaskTracker: Error
> running child
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
