[
https://issues.apache.org/jira/browse/HADOOP-6433?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12789078#action_12789078
]
Todd Lipcon commented on HADOOP-6433:
-------------------------------------
Hi Zheng,
- Can you convert the unit test to JUnit 4 style? (ie using annotations instead
of extending TestCase)
- Can you add a test case for a volume that is not in vols?
- Rather than sleeping in the test case, can you call shutdown and then
awaitTermination() with a longer timeout? This should ensure that the test is
allowed to run longer if the machine is super slow, but doesn't waste time if
not necessary. This would also make you add an awaitTermination() call, which I
think is a good thing to expose.
- What's with the changes to FsShell? Was this accidental?
- Missing license header on AsyncDIskService.java
- The comment still says "In the future, we should put it in common". Since
this is a common jira, please remove that
- IMO the thread counts per volume should be configurable. Some systems can
handle plenty of concurrent metadata ops while others, not so much. I don't
feel strongly about this, though - happy to leave it to a later JIRA if people
want this.
- Should the threads be made daemon threads? I'm not sure what the expected
behavior should be when the server gets SIGINTed, but I think if they're not
daemon threads it will block server shutdown.
Some of the above comments probably apply to the existing code, since this is
just a move. If so, it seems fine to file a new JIRA for these improvements so
we can move forward with the main thrust of this issue.
Thanks!
> Add AsyncDiskService that is used in both hdfs and mapreduce
> ------------------------------------------------------------
>
> Key: HADOOP-6433
> URL: https://issues.apache.org/jira/browse/HADOOP-6433
> Project: Hadoop Common
> Issue Type: New Feature
> Reporter: Zheng Shao
> Assignee: Zheng Shao
> Attachments: HADOOP-6433.1.patch
>
>
> Both MAPREDUCE-1213 and HDFS-611 are using a class called AsyncDiskService.
> The idea is to create a thread pool per disk volume, and use that for
> scheduling async disk operations.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
