On 2018/07/05 23:00, Robert Haas wrote:
> With respect to partitioning specifically, it seems like we might be
> able to come up with some way of planning that doesn't need a full
> relcache entry for every partition, particularly if there are no
> partition-local objects (indexes, triggers, etc.).
We won't know that there are no partition-local objects until we open them
though, right? As you said, there might be a way to refactor things such
that just knowing that there are no partition-local objects becomes
cheaper than doing a full-fledged RelationBuildDesc.
Thanks,
Amit