From: Michael P. <mic...@gm...> - 2012-06-20 04:55:25
|
On Wed, Jun 20, 2012 at 12:58 PM, Ashutosh Bapat < ash...@en...> wrote: > > > On Wed, Jun 20, 2012 at 9:18 AM, Michael Paquier < > mic...@gm...> wrote: > >> >> >> On Wed, Jun 20, 2012 at 12:46 PM, Ashutosh Bapat < >> ash...@en...> wrote: >> >>> One fix, I can think of is to create volatile functions only on >>> coordinator. Although, I would still take a step back, and find out why we >>> took the decision not to store views on the datanodes. >>> >> View => projection of table data => need distribution type of table => >> distribution data only available on Coordinator for data distribution => no >> sense to define views on Datanodes >> > > In the case, where a view type is used as function argument or return > type, it does make sense to have the view definition on the datanodes. The > implication behind my question is whether there is any correctness problem > by creating view and related definitions at the datanodes. > By taking this question from another angle: Are there any problems to push down clauses using views to Datanodes? Just based on correctness, the answer is no problem. Btw, the function using a view should be volatile as it reads data, so it will not be used on Datanodes at all... -- Michael Paquier http://michael.otacoo.com |