[jira] [Issue Comment Edited] (CASSANDRA-2855) Skip rows with empty columns when slicing entire row

["Jeremy Hanna (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-2855?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13072557#comment-13072557 ] 

Jeremy Hanna edited comment on CASSANDRA-2855 at 7/28/11 10:16 PM:
-------------------------------------------------------------------

Added a configuration property cassandra.skip.empty.results which defaults to false.  We can't skip just complete empty rows because there is no way to tell if the complete row is empty based on a result from a slice predicate.

      was (Author: jeromatron):
    Added a configuration property cassandra.skip.empty.results which defaults to false.  We can't skip just complete empty rows because there is no way to tell if the complete row is empty based on a result that is a slice predicate.
  
> Skip rows with empty columns when slicing entire row
> ----------------------------------------------------
>
>                 Key: CASSANDRA-2855
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-2855
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: API
>            Reporter: Jeremy Hanna
>            Assignee: Jeremy Hanna
>            Priority: Minor
>              Labels: hadoop
>             Fix For: 0.7.9, 0.8.3
>
>         Attachments: 2855-v2.txt, 2855-v3.txt
>
>
> We have been finding that range ghosts appear in results from Hadoop via Pig.  This could also happen if rows don't have data for the slice predicate that is given.  This leads to having to do a painful amount of defensive checking on the Pig side, especially in the case of range ghosts.
> We would like to add an option to skip rows that have no column values in it.  That functionality existed before in core Cassandra but was removed because of the performance penalty of that checking.  However with Hadoop support in the RecordReader, that is batch oriented anyway, so individual row reading performance isn't as much of an issue.  Also we would make it an optional config parameter for each job anyway, so people wouldn't have to incur that penalty if they are confident that there won't be those empty rows or they don't care.
> It could be parameter cassandra.skip.empty.rows and be true/false.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira