Re: Question about optimising (Postgres_)FDW

(2014/04/16 22:16), Hannu Krosing wrote:
> On 04/16/2014 01:35 PM, Etsuro Fujita wrote:

>> Maybe I'm missing something, but I think that you can do what I think
>> you'd like to do by the following procedure:
> No, what I'd like PostgreSQL to do is to
>
> 1. select the id+set from local table
> 2. select the rows from remote table with WHERE ID IN (<set selected in
> step 1>)
> 3. then join the original set to selected set, with any suitable join
> strategy
>
> The things I do not want are
>
> A. selecting all rows from remote table
>      (this is what your examples below do)
>
> or
>
> B. selecting rows from remote table by single selects using "ID = $"
>      (this is something that I managed to do by some tweaking of costs)
>
> as A will be always slow if there are millions of rows in remote table
> and B is slow(ish) when the idset is over a few hundred ids
>
> I hope this is a bit better explanation than I provided before .

Ah, I understand what you'd like to do.  Thank you for the explanation.

> P.S. I am not sure if this is a limitation of postgres_fdw or postgres
> itself

If I understand correctly, neither the current postgres_fdw planning 
function nor the current postgres planner itself support such a plan. 
For that I think we would probably need to implement a distributed query 
processing technique such as semijoin or bloomjoin in those modules.

Thanks,

P.S.
>> or, that as Tom mentioned, by disabling the use_remote_estimate function:

I misunderstood the meaning of what Tom pointed out.  Sorry for that.

Best regards,
Etsuro Fujita