[
https://issues.apache.org/jira/browse/HDFS-4058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13476588#comment-13476588
]
Aaron T. Myers commented on HDFS-4058:
--------------------------------------
+1, the patch looks good to me.
A test would be nice, but given how simple this problem is once it's been
pointed out, I don't think it's strictly necessary in this case. To prevent
this from regressing, maybe just add a comment along the lines of:
{code}
// Use an array here instead of a dynamically-sized data structure
// so there's no issues when we insert the results in the event the
// threads return out of order. See HDFS-4058 for more info.
{code}
> DirectoryScanner may fail with IOOB if the directory scanning threads return
> out of volume order
> ------------------------------------------------------------------------------------------------
>
> Key: HDFS-4058
> URL: https://issues.apache.org/jira/browse/HDFS-4058
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: data-node
> Affects Versions: 2.0.0-alpha
> Reporter: Eli Collins
> Assignee: Eli Collins
> Attachments: hdfs-4058.txt
>
>
> The DirectoryScanner may fail with an IOOB if the directory scanning threads
> return out of volume order (ie volume scanner #3 returns before volume
> scanner #2). This is because it's using an ArrayList and ArrayList#add(index,
> element) throws IOOB if idx >= size, and size is only increased as elements
> are added, therefore adds have to be done in index order. Since we know the
> size when we create the ArrayList let's just use an array (perhaps an
> ArrayList was used originally because someone thought array creation called
> the default constructor?).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
