On Tue, Sep 9, 2014 at 4:30 PM, Luyuanfei <ka...@hu...> wrote:
> /*
> * cost_remotequery
> * As of now the function just sets the costs to 0 to make this path the
> * cheapest.
> * PGXC_TODO: Ideally, we should estimate the costs of network transfer from
> * datanodes and any datanode costs involved.
> */
> It's very useful to some statement such as join or subquery.
In short, no. And I am not aware of anybody working in this area for
this project. Any patch would need a paper or research made in this
area to start on good basis IMO. Even without researches in this area,
I imagine that we'd finish for sure with a set of cost parameters that
can be set per-node (you don't want to have the same network cost for
a node on local network and another one thousands of miles away) and
that the planner could use those parameters when evaluating a plan
using an equal qual on an index used for hash. Another problem is that
we need a good idea of how they compare to other cost parameters to
avoid unbalanced plans. The issue remaining is then what are those
parameters, and what they symbolize.
> When you want reliability, choose Perforce.
Not much sure about that. Git is more appealing with its decentralized system.
--
Michael
|
