[
https://issues.apache.org/jira/browse/HDFS-7632?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Nauroth updated HDFS-7632:
--------------------------------
Attachment: HDFS-7632.001.patch
The root cause is in {{MiniDFSCluster#getStorageDirPath}}, which had been
hard-coded to assume 2 storage locations per DataNode. The attached patch
fixes it to use the number of storage locations per DataNode requested by the
caller. This required changing a few methods from static methods to instance
methods, and then the corresponding call sites in the test code needed to be
updated. {{MiniDFSCluster}} is annotated as {{Unstable}}, so we're free to
make this change. It's also listed as {{LimitedPrivate}} for a few projects.
I grep'd those projects and didn't see any calls to these methods.
> MiniDFSCluster configures DataNode data directories incorrectly if using more
> than 1 DataNode and more than 2 storage locations per DataNode.
> ---------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-7632
> URL: https://issues.apache.org/jira/browse/HDFS-7632
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: test
> Reporter: Chris Nauroth
> Assignee: Chris Nauroth
> Attachments: HDFS-7632.001.patch
>
>
> {{MiniDFSCluster}} allows the caller to set the number of storage locations
> per DataNode. If this number is set higher than 2, and if the cluster is
> configured with more than 1 DataNode, then the calculation of the storage
> directory paths will be incorrect. Multiple DataNodes will attempt to use
> the same storage directory, and one of them will fail while trying to acquire
> the file lock.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
