[jira] Updated: (HADOOP-1657) NNBench benchmark hangs with trunk

["Raghu Angadi (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HADOOP-1657?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Raghu Angadi updated HADOOP-1657:
---------------------------------

    Attachment: HADOOP-1657.patch

Attached patch does the following:

 - User can optionally specify -bytePerChecksum.
 - If bytesPerChecksum checksum is not set, then default from config is used.
 - When default is used, and bytesPerBlock is not multiple of bytesPerChecksum, then bytesPerChecksum is set to bytesPerBlock
 - All the exceptions are logged.
 - When number of exceptions is larger than {{numFiles*100}}, a runtime exception is thrown to stop the test.

Above list long but patch is simple :)

> NNBench benchmark hangs with trunk
> ----------------------------------
>
>                 Key: HADOOP-1657
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1657
>             Project: Hadoop
>          Issue Type: Bug
>          Components: dfs
>    Affects Versions: 0.14.0
>            Reporter: Raghu Angadi
>            Assignee: Raghu Angadi
>            Priority: Blocker
>             Fix For: 0.14.0
>
>         Attachments: HADOOP-1657.patch, HADOOP-1657.patch
>
>
> NNBench runs with a small block size (say 20). But uses default value of 512 for io.bytes.per.checksum. Since HADOOP-1134, block size should be a multiple of of bytes.per.checksum. Fix is to set bytes.per.checksum to same as blocksize.
> I think following changes to NNBench would help in general (at least the first one) :
>  - NNBench does not log these exceptions. I think it should. 
>  - It calls create() in an infinite loop as long as create does not succeed. May be we should have an upper limit. Say, max 10000.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.