Re: Removing [Merge]Append nodes which contain a single subpath

On Thu, Oct 26, 2017 at 5:07 PM, David Rowley
<david(dot)rowley(at)2ndquadrant(dot)com> wrote:
> On 26 October 2017 at 23:42, Antonin Houska <ah(at)cybertec(dot)at> wrote:
>> David Rowley <david(dot)rowley(at)2ndquadrant(dot)com> wrote:
>>
>>> Method 1:
>>>
>>> In set_append_rel_size() detect when just a single subpath would be
>>> added to the Append path.
>>
>> I spent some time reviewing the partition-wise join patch during the last CF
>> and have an impression that set_append_rel_size() is called too early to find
>> out how many child paths will eventually exist if Append represents a join of
>> a partitioned table. I think the partition matching takes place at later stage
>> and that determines the actual number of partitions, and therefore the number
>> of Append subpaths.
>
> I understand that we must wait until set_append_rel_size() is being
> called on the RELOPT_BASEREL since partitions may themselves be
> partitioned tables and we'll only know what's left after we've
> recursively checked all partitions of the baserel.
>
> I've not studied the partition-wise join code yet, but if we've
> eliminated all but one partition in set_append_rel_size() then I
> imagine we'd just need to ensure partition-wise join is not attempted
> since we'd be joining directly to the only partition anyway.
>

I think Antonin has a point. The join processing may deem some base
relations dummy (See populate_joinrel_with_paths()). So an Append
relation which had multiple child alive at the end of
set_append_rel_size() might ultimately have only one child after
partition-wise join has worked on it.

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