[
https://issues.apache.org/jira/browse/HADOOP-12831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15159532#comment-15159532
]
Mingliang Liu commented on HADOOP-12831:
----------------------------------------
It looks like in {{DataChecksum#newDataChecksum}}, returning null is by desgin:
{code}
public static DataChecksum newDataChecksum(Type type, int bytesPerChecksum ) {
if ( bytesPerChecksum <= 0 ) {
return null;
}
...
{code}
The reason may be that, the bytesPerChecksum was used as divisor for
calculating number of chunks and checksum buffer size. We can document this, or
throw exception with a user-friendly message for this.
I think this is not feasible to disable checksumming and buffering in this way.
Will {{RawLocalFileSystem}} work in your case?
> LocalFS/FSOutputSummer NPEs in constructor if bytes per checksum set to 0
> --------------------------------------------------------------------------
>
> Key: HADOOP-12831
> URL: https://issues.apache.org/jira/browse/HADOOP-12831
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs
> Affects Versions: 2.8.0
> Reporter: Steve Loughran
> Priority: Minor
>
> If you set the number of bytes per checksum to zero,
> {code}
> conf.setInt(LocalFileSystemConfigKeys.LOCAL_FS_BYTES_PER_CHECKSUM_KEY, 0)
> {code}
> then create a "file://" instance, you get to see a stack trace
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
