On Thu, Feb 18, 2021 at 4:35 PM Amit Langote <amitlangote09@gmail.com> wrote:
>
> Looking at this again, I am a bit concerned about going over the whole
> partition tree *twice* when making a parallel plan for insert into
> partitioned tables. Maybe we should do what you did in your first
> attempt a slightly differently -- add partition OIDs during the
> max_parallel_hazard() initiated scan of the partition tree as you did.
> Instead of adding them directly to PlannerGlobal.relationOids, add to,
> say, PlannerInfo.targetPartitionOids and have set_plan_references() do
> list_concat(glob->relationOids, list_copy(root->targetPartitionOids)
> in the same place as setrefs-v3 does
> add_target_partition_oids_recurse(). Thoughts?
>
Agreed, that might be a better approach, and that way we're also only
recording the partition OIDs that the parallel-safety checks are
relying on.
I'll give it a go and see if I can detect any issues with this method.
Regards,
Greg Nancarrow
Fujitsu Australia