Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> Tom Lane wrote:
>> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
>>> Won't this cause issues to MergeAppend optimizations?
>> Like what?
> Well, as I understand, MergeAppend needs to know the sort order of the
> child node, right? But that's available only on the relation RTE, not
> on the colstore-join RTE.
Uh, what? Sort order is a property of a path, not an RTE. And we
have always understood which join types preserve sort order.
>> And if there are such issues, why do you think you wouldn't be
>> expected to solve them?
> Precisely. If I simply reject having column stores in partitioned
> tables, then I don't *need* to solve them.
You misunderstood the thrust of my comment, which basically is that
I doubt anyone will think that rejecting that combination is an
acceptable implementation restriction. It might be all right if it
doesn't work very well in v0, but not if the implementation is designed
so that it can never be fixed.
regards, tom lane