[
https://issues.apache.org/jira/browse/HADOOP-5618?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12699349#action_12699349
]
Jakob Homan commented on HADOOP-5618:
-------------------------------------
This actually isn't a good idea at this time. The problem is that the
implementation detail that storageDirs is an ordered list leaked out into many
other places in the code and is now relied on. Many places expect the iterator
over the list to traverse it in the order the storage dirs were added to it,
not by the path name associated with the storage (the most reasonable key to a
map). For instance, in FSEditLog.java, divertFileStreams is expecting to find
corresponding storage dirs and edit streams to work over. Of the 27 or so
times that the DirIterator is used (which is based on this ArrayList), only one
is actually searching for something like a pathname.
There's certainly room to correct this but it should be done as a larger clean
up of the fsimage code. Thoughts?
> Convert Storage.storageDirs into a map.
> ---------------------------------------
>
> Key: HADOOP-5618
> URL: https://issues.apache.org/jira/browse/HADOOP-5618
> Project: Hadoop Core
> Issue Type: Improvement
> Components: dfs
> Affects Versions: 0.21.0
> Reporter: Konstantin Shvachko
> Assignee: Jakob Homan
> Fix For: 0.21.0
>
>
> {{Storage.storageDirs}} is currently declared as an {{ArrayList}}. Recent
> changes made {{storageDirs}} a searchable collection. In order to reflect
> this changes the {{storageDirs}} type should be changed to a searchable
> collection. This will simplify and optimize current code.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
