On Wed, Mar 16, 2016 at 2:25 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Wed, Mar 16, 2016 at 2:02 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Looks pretty close. One point is that if we do end up using a Result
>>> node, then the parent GatherPath does not get charged for the Result
>>> node's cpu_per_tuple overhead. I'm not sure that that's worth changing
>>> though. It's probably better to bet that the subpath is projectable and
>>> so no cost will ensue, than to bet the other way.
>
>> I'm almost sure this way is the better bet.
>
> Actually, we do know what will happen ... so maybe
>
> /*
> * We always use create_projection_path here, even if the subpath is
> * projection-capable, so as to avoid modifying the subpath in place.
> * It seems unlikely at present that there could be any other
> * references to the subpath anyway, but better safe than sorry.
> */
> + if (!is_projection_capable_path(gpath->subpath))
> + gpath->path.total_cost += cpu_tuple_cost * gpath->subpath->rows;
> gpath->subpath = (Path *)
> create_projection_path(root,
> gpath->subpath->parent,
> gpath->subpath,
> target);
>
> The comment could use adjustment if you adopt that, to reference the fact
> that we know create_projection_plan will get rid of the Result if not
> needed.
OK, I've committed something along those lines. Thanks for the
advice, and feel free to whack it around if you have an idea for
improving it further - though IMHO this is good enough for 9.6.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company