Re: pg15 inherited stats expressions: cache lookup failed for statistics object

Richard Guo <guofenglinux@gmail.com>

From: Richard Guo <guofenglinux@gmail.com>
To: Justin Pryzby <pryzby@telsasoft.com>
Cc: Tomas Vondra <tomas.vondra@enterprisedb.com>, pgsql-hackers@lists.postgresql.org
Date: 2022-10-31T05:12:09Z
Lists: pgsql-hackers

Attachments

On Mon, Oct 31, 2022 at 12:26 PM Richard Guo <guofenglinux@gmail.com> wrote:

> On Mon, Oct 31, 2022 at 1:05 AM Justin Pryzby <pryzby@telsasoft.com>
> wrote:
>
>> I think this is what's needed.
>>
>> diff --git a/src/backend/utils/adt/selfuncs.c
>> b/src/backend/utils/adt/selfuncs.c
>> index 14e0885f19f..4450f0d682f 100644
>> --- a/src/backend/utils/adt/selfuncs.c
>> +++ b/src/backend/utils/adt/selfuncs.c
>> @@ -5240,6 +5240,8 @@ examine_variable(PlannerInfo *root, Node *node, int
>> varRelid,
>>                         /* skip stats without per-expression stats */
>>                         if (info->kind != STATS_EXT_EXPRESSIONS)
>>                                 continue;
>> +                       if (info->inherit != rte->inh)
>> +                               continue;
>>
>>                         pos = 0;
>>                         foreach(expr_item, info->exprs)
>>
>
> I think we also need to do this when loading the ndistinct value, to
> skip statistics with mismatching stxdinherit in
> estimate_multivariate_ndistinct().
>

To be concrete, I mean something like attached.

BTW, I noticed a micro-optimization opportunity in examine_variable that
we can fetch the RangeTblEntry for 'onerel' outside the foreach loop
when iterating the extended stats so that we can do it only once rather
than for each stat.

Thanks
Richard

Commits

  1. Fix planner failure with extended statistics on partitioned tables.