hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ayappan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-9627) TestSocketIOTimeout should be rewritten without platform-specific assumptions
Date Wed, 02 Jul 2014 12:31:26 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-9627?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14049867#comment-14049867
] 

Ayappan commented on HADOOP-9627:
---------------------------------

TestSocketIOWithTimeout uses a block size of 4192 bytes to simulate a partial write. This
seems to be a valid in x86 architecture where the default minimum blocksize is 4096. 
This testcase fails in PowerPC where the default minimum block size is 65536 bytes (64KB).
So for PowerPC, using a blocksize little more than 64K , say 65632(65536 + 96) holds good
for this scenario.
I attached a patch here where i made it very general by introducing NativeIO.POSIX.getCacheManipulator().getOperatingSystemPageSize()
to get the page size.
I tested my patch in both ppc64 and x86 linux machines.

> TestSocketIOTimeout should be rewritten without platform-specific assumptions
> -----------------------------------------------------------------------------
>
>                 Key: HADOOP-9627
>                 URL: https://issues.apache.org/jira/browse/HADOOP-9627
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: test
>    Affects Versions: 3.0.0, 2.3.0
>            Reporter: Arpit Agarwal
>         Attachments: HADOOP-9627.patch
>
>
> TestSocketIOTimeout makes some assumptions about the behavior of file channels wrt partial
writes that do not appear to hold true on Windows [details in HADOOP-8982].
> Currently part of the test is skipped on Windows.
> This bug is to track fixing the test.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message