Re: expanding inheritance in partition bound order

On Fri, Aug 18, 2017 at 10:12 AM, Amit Langote
<Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote
>
>>> 0002: Teach expand_inherited_rtentry to use partition bound order
>>
>> 0004 in [1] expands a multi-level partition hierarchy into similar
>> inheritance hierarchy. That patch doesn't need all OIDs in one go. It
>> will have to handle the partition hierarchy level by level, so most of
>> the code added by this patch will need to be changed by that patch. Is
>> there a way we can somehow change this patch so that the delta in 0004
>> is reduced? That may need rethinking about using
>> RelationGetPartitionDispatchInfo().
>
> Regarding that, I have a question:
>
> Does the multi-level partition-wise join planning logic depend on the
> inheritance itself to be expanded in a multi-level aware manner. That is,
> expanding the partitioned table inheritance in multi-level aware manner in
> expan_inherited_rtentry()?

Yes, it needs AppendRelInfos to retain the parent-child relationship.
Please refer [1], [2], [3] for details.

>
> Wouldn't it suffice to just have the resulting Append paths be nested per
> multi-level partitioning hierarchy?

We are joining RelOptInfos, so those need to be nested. For those to
be nested, we need AppendRelInfos to preserve parent-child
relationship. Nesting paths doesn't help. Append paths actually should
be flattened out to avoid any extra time consumed in nested Append
node.

[1] https://www.postgresql.org/message-id/CAFjFpRceMmx26653XFAYvc5KVQcrzcKScVFqZdbXV%3DkB8Akkqg@mail.gmail.com
[2] https://www.postgresql.org/message-id/CAFjFpRefs5ZMnxQ2vP9v5zOtWtNPuiMYc01sb1SWjCOB1CT=uQ@mail.gmail.com
[3] https://www.postgresql.org/message-id/CAFjFpRd6Kzx6Xn%3D7vdwwnh6rEw2VEgo--iPdhV%2BFb7bHfPzsbw%40mail.gmail.com

--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company