(2018/06/22 23:58), Robert Haas wrote:
I think this approach is going to run into trouble if the level at which we have to apply the ConvertRowTypeExpr happens not to be a projection-capable node.
Actually, the level we have to do that would be a child rel of a partitioned table or a child join of a partitionwise join, so the plan node would be a scan or join plan unless the child rel or child join is itself partitioned, in which case the plan node would be Append/MergeAppend and the proposed patch recursively would apply that conversion to child plans for the Append/MergeAppend).
And, in general, it seems to me that we want to produce the right outputs at the lowest possible level of the plan tree. For instance, suppose that one of the relations being scanned is not parallel-safe but the others are. Then, we could end up with a plan like this: Append -> Seq Scan on temp_rela -> Gather -> Parallel Seq Scan on thing1 -> Gather -> Parallel Seq Scan on thing2 If a projection is required to convert the row type expression, we certainly want that to get pushed below the Gather, not to happen at the Gather level itself.
IIUC, we currently don't consider such a plan for partition-wise join; we'll only consider gathering partial paths for the parent appendrel. So, I'm not sure we need to take into account that when applying the ConvertRowtypeExpr. Maybe I'm missing something, though.
We certainly don't want it to happen at the Append level, which can't even handle it.
I think so too. Best regards, Etsuro Fujita
