[ 
https://issues.apache.org/jira/browse/CRUNCH-246?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13732031#comment-13732031
 ] 

Ryan Brush commented on CRUNCH-246:
-----------------------------------

I actually wrote an HFileInputFormat for a specific internal need some time ago 
that I'd be happy to contribute to this (it's actually quite simple).  However, 
I think HBase intentionally doesn't include an HFileInputFormat since it would 
be prone to error, some of which is discussed at [1].  In short, the input 
wouldn't include data sitting in HBase's write ahead log -- and therefore 
logically written to HBase -- but not yet flushed to an HFile. There are ways 
you can work around this if you really know what you're doing...but it might be 
worth debating whether something like that makes sense in a public API. FWIW, 
the HBase community seems to have decided this isn't a good thing to expose 
broadly.

I recall some discussion of HBase snapshotting opening the door to running 
MapReduce jobs directly against HFiles in a predictable way, but unfortunately 
I can't find a reference to that. If and when that becomes available it would 
make complete sense to expose in Crunch (and I'm sure some folks here at Cerner 
would be happy to contribute to that as well).

[1]
http://mail-archives.apache.org/mod_mbox/hbase-user/201202.mbox/%3C-684097822435171056@unknownmsgid%3E
                
> HFileSource
> -----------
>
>                 Key: CRUNCH-246
>                 URL: https://issues.apache.org/jira/browse/CRUNCH-246
>             Project: Crunch
>          Issue Type: Improvement
>          Components: IO
>            Reporter: Chao Shi
>            Assignee: Chao Shi
>
> I found this useful when directly perform MR on HFiles. I used it yesterday 
> when copying a bunch of HFiles to another cluster (where the region layout is 
> different).
> There is no HFileInputFormat provided by HBase, but I found the following 
> from google:
> https://gist.github.com/leifwickland/1120311
> http://blog.csdn.net/kirayuan/article/details/7794402 (Java version of the 
> above. The webpage is in chinese, but you can see the code)
> I'm not sure if we copy their code directly (copyright issue?). Anyone knows?

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to