Re: [Postgres-xc-general] modifying values at the datanode

[Ashutosh B. <ash...@en...>]

The changes have to be in pgxc_make_modifytable() to construct a shippable
DML out of the scan plan and the modification plan. Changes will be needed
in ExecUpdate/Delete/Insert for execution of this shippable DML. Also,
there will be changes needed in pgxc_shippability_walker() to improve FQS
technique for the same.


On Mon, Oct 7, 2013 at 9:10 PM, Sandeep Gupta <gup...@gm...>wrote:

> Hi Ashutosh,
>
>  Thanks for the note.  I cannot commit right away. However, whenever you
> have some time can you mention relevant portions in the codebase where
> changes have to be made. I have a general understanding of the execution
> engine. I will take a look see if it feasible for me.
>
> Thanks.
> Sandeep
>
>
>
> On Sun, Oct 6, 2013 at 11:53 PM, Ashutosh Bapat <
> ash...@en...> wrote:
>
>>
>>
>>
>> On Sat, Oct 5, 2013 at 7:44 PM, Sandeep Gupta <gup...@gm...>wrote:
>>
>>> Thanks Michael. I understand. The only issue is that we have an update
>>> query as
>>>
>>> update T set T.a = -1 from A where A.x = T.x
>>>
>>>
>>> Both A and T and distributed by x column. The problem is that
>>> coordinator first does the join and then
>>> calls update several times at each datanode. This is turning out to be
>>> too slow. Would have
>>> been better if the entire query was shipped to the datanodes.
>>>
>>>
>> Right now there is no way to ship a DML with more than one relation
>> involved there. But that's something, I have been thinking about. If you
>> have developer resources and can produce a patch. I can help.
>>
>>
>>> Thanks.
>>> Sandeep
>>>
>>>
>>>
>>> On Sat, Oct 5, 2013 at 6:27 AM, Michael Paquier <
>>> mic...@gm...> wrote:
>>>
>>>> On Sat, Oct 5, 2013 at 2:58 AM, Sandeep Gupta <gup...@gm...>
>>>> wrote:
>>>> >  I understand that the datanodes are read only and that
>>>> updates/insert can
>>>> > happen at coordinator.
>>>> You got it.
>>>>
>>>> > Also, it does not allow modification of column over which the records
>>>> are distributed.
>>>> Hum no, 1.1 allows ALTER TABLE that you can use to change the
>>>> distribution type of a table.
>>>>
>>>> > However, in case I know what I am doing, it there anyway possible to
>>>> modify
>>>> > the values directly at datanodes.
>>>> > The modifications are not over column over which distribution happens.
>>>> If you mean by connecting directly to the Datanodes, no. You would
>>>> break data consistency if table is replicated by the way by doing
>>>> that. Let the Coordinator planner do the job and choose the remote
>>>> nodes for you.
>>>>
>>>> There have been discussion to merge Coordinators and Datanodes
>>>> together though. This would allow what you say, with a simpler cluster
>>>> design.
>>>> --
>>>> Michael
>>>>
>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> October Webinars: Code for Performance
>>> Free Intel webinars can help you accelerate application performance.
>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most
>>> from
>>> the latest Intel processors and coprocessors. See abstracts and register
>>> >
>>>
>>> http://pubads.g.doubleclick.net/gampad/clk?id=60134791&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Postgres-xc-general mailing list
>>> Pos...@li...
>>> https://lists.sourceforge.net/lists/listinfo/postgres-xc-general
>>>
>>>
>>
>>
>> --
>> Best Wishes,
>> Ashutosh Bapat
>> EnterpriseDB Corporation
>> The Postgres Database Company
>>
>
>


-- 
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company