Horton

Re: [HACKERS] advanced partition matching algorithm for partition-wise join

Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>

View thread
From: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
To: amul sul <sulamul@gmail.com>
Cc: Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>, Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp>, Dmitry Dolgov <9erthalion6@gmail.com>, Thomas Munro <thomas.munro@enterprisedb.com>, Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>, Antonin Houska <ah@cybertec.at>, Robert Haas <robertmhaas@gmail.com>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2019-03-11T16:43:51Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Suppress unused-variable warning.

  2. Allow partitionwise joins in more cases.

  3. Avoid crash in partitionwise join planning under GEQO.

  4. Disable support for partitionwise joins in problematic cases.

  5. Add plan_cache_mode setting

  6. Add test for partitionwise join involving default partition.

  7. Revise API for partition_rbound_cmp/partition_rbound_datum_cmp.

On Mon, Mar 11, 2019 at 10:40 AM amul sul <sulamul@gmail.com> wrote:

>
> All the places from where this handle_missing_partition() get called
> have the following code to decide the value for missing_side_outer/_inner
> which
> I yet to understand. Do you think this has some flaw?
>
>         /*
>          * For a FULL join, inner relation acts as both OUTER and INNER
>          * relation.  For LEFT and ANTI join the inner relation acts as
>          * INNER relation. For INNER and SEMI join OUTER and INNER
>          * differentiation is immaterial.
>          */
>         missing_side_inner = (jointype == JOIN_FULL ||
>                               jointype == JOIN_LEFT ||
>                               jointype == JOIN_ANTI);
>         missing_side_outer = (jointype == JOIN_FULL);
>

I was wrong, sorry. The comment says it all.


>
>
>
>> argument value which fails to set merged_index.
>>>
>>> In the attached patch, I tried to fix this case by setting merged_index
>>> explicitly which fixes the reported crash.
>>>
>>
>> I expect handle_missing_partition() to set the merged_index always. In
>> your patches, I don't see that function in your patches is setting it
>> explicitly. If we are setting merged_index explicitly somewhere else, other
>> places may miss that explicit assignment. So it's better to move it inside
>> this function.
>>
>>
>
> Ok, that can be fixed.
>
> Similarly, I think merge_null_partitions should set null_index instead of
> asserting when null partitions missing from both the side, make sense?
>

I think not. null_index, once set shouldn't change and hence does not
change with each pair of partitions being matched. So, it makes sense to
make sure that null_index remains invalid if none of the tables have null
partition.

--
Best Wishes,
Ashutosh Bapat