[
https://issues.apache.org/jira/browse/HADOOP-4619?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12648476#action_12648476
]
Brian Bockelman commented on HADOOP-4619:
-----------------------------------------
Hey Pete,
I think Hudson will kill us if we post another patch but ....
There are two issues:
1) Size of the file to create: this should be off_t (look at the struct defined
in stat.h:
http://www.opengroup.org/onlinepubs/000095399/basedefs/sys/stat.h.html)
2) Possible size of the reads returned by hdfsWrite; libhdfs defines this to be
tSize. Currently, it's an unsigned int32 (I think), but really should be
ssize_t.
We can't fix (2) without fixing libhdfs. We can fix (1); however, I can't
think of a system where off_t would still be a 32-bit int...
> hdfs_write infinite loop when dfs fails and cannot write files > 2 GB
> ---------------------------------------------------------------------
>
> Key: HADOOP-4619
> URL: https://issues.apache.org/jira/browse/HADOOP-4619
> Project: Hadoop Core
> Issue Type: Bug
> Components: libhdfs
> Affects Versions: 0.19.0, 0.20.0
> Reporter: Pete Wyckoff
> Assignee: Pete Wyckoff
> Fix For: 0.19.1, 0.20.0
>
> Attachments: HADOOP-4619.txt, HADOOP-4619.txt, HADOOP-4619.txt,
> HADOOP-4619.txt, HADOOP-4619.txt, HADOOP-4619.txt
>
>
> 1. hdfs_write does not check hdfsWrite return code so -1 return code is
> ignored.
> 2. hdfs_write uses int for overall file length
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
