[
https://issues.apache.org/jira/browse/HAWQ-583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15398007#comment-15398007
]
Shivram Mani edited comment on HAWQ-583 at 7/28/16 7:10 PM:
------------------------------------------------------------
Refer to HAWQ-927 for the http header params introduced to support column
projection.
If the table has 3 columns col1, col2, col3 the following table describes the
PXF response corresponding to various types of query patterns
||Query||X-GP-ATTRS-PROJ||X-GP-ATTRS-PROJ-IDX||PXF Action||
| select col1 |1|0|Project col1 data with NULL values padded in col2 and col3|
| select col1,col2 |2|0, 1|Project col1 and col2 data with NULL values in col3|
| select count(col2) |1|1|Project col2 with NULL values in col1 and col3|
| select count ( * ) |0|NULL|Project only one column. We will pick col1 for
consistency purpose|
| select * |NULL|NULL|Project all columns|
| select col1,* |4|0,0,1,2|Project col1,col2 and col3|
was (Author: shivram):
Refer to HAWQ-927 for the http header params introduced to support column
projection.
If the table has 3 columns col1, col2, col3 the following table describes the
PXF response corresponding to various types of query patterns
||Query||X-GP-ATTRS-PROJ||X-GP-ATTRS-PROJ-IDX||PXF Action||
| select col1 |1|0|Project col1 data with NULL values padded in col2 and col3|
| select col1,col2 |2|0, 1|Project col1 and col2 data with NULL values in col3|
| select count(col2) |1|1|Project col2 with NULL values in col1 and col3|
| select count(*) |0|NULL|Project only one column. We will pick col1 for
consistency purpose|
| select * |NULL|NULL|Project all columns|
| select col1,* |4|0,0,1,2|Project col1,col2 and col3|
> Extend PXF to allow plugins to support returning partial content of
> SELECT(column projection)
> ---------------------------------------------------------------------------------------------
>
> Key: HAWQ-583
> URL: https://issues.apache.org/jira/browse/HAWQ-583
> Project: Apache HAWQ
> Issue Type: New Feature
> Components: PXF
> Reporter: Michael Andre Pearce (IG)
> Assignee: Kavinder Dhaliwal
> Fix For: backlog
>
>
> Currently PXF supports being able to push down the predicate WHERE logic to
> the external system to reduce the amount data needed to be retrieved.
> SELECT a, b FROM external_pxf_source WHERE z < 3 AND x > 6
> As such we can filter the rows returned, but currently still would have to
> return all the fields / complete row.
> This proposal is so that we can return only the columns in SELECT part.
> For data sources where it is columnar storage or selectable such as remote
> database that PXF can read or connect to this has advantages in the data that
> needs to be accessed or even transferred.
> As like with the push down Filter it should be optional so that plugins that
> provide support can use it but others that do not, continue to work as they
> do.
> The proposal would be for
> 1) create an interface created for plugins to optionally implement, where the
> columns needed to be returned are given to the plugin.
> 2) update pxf api for hawq to send columns defined in SELECT, for pxf to
> invoke the plugin interface and pass this information onto if provided
> 3) update pxf integration within hawq itself so that hawq passes this
> additonal information to pxf.
> This Ticket is off the back of discussion on HAWQ-492.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
