Horton

Re: extended stats on partitioned tables

Tomas Vondra <tomas.vondra@enterprisedb.com>

View thread View on postgr.es
From: Tomas Vondra <tomas.vondra@enterprisedb.com>
To: Justin Pryzby <pryzby@telsasoft.com>
Cc: Zhihong Yu <zyu@yugabyte.com>, Jaime Casanova <jcasanov@systemguards.com.ec>, pgsql-hackers@postgresql.org
Date: 2022-01-15T19:05:27Z
Lists: pgsql-hackers 

On 1/15/22 19:35, Tomas Vondra wrote:
> On 1/15/22 06:11, Justin Pryzby wrote:
>> On Mon, Dec 13, 2021 at 09:40:09PM +0100, Tomas Vondra wrote:
>>> 1) If the table is a separate relation (not part of an inheritance
>>> tree), this should make no difference. -> OK
>>>
>>> 2) If the table is using "old" inheritance, this reverts back to
>>> pre-regression behavior. So people will keep using the old statistics
>>> until the ANALYZE, and we need to tell them to ANALYZE or something.
>>>
>>> 3) If the table is using partitioning, it's guaranteed to be empty and
>>> there are no stats at all. Again, we should tell people to run ANALYZE.
>>
>> I think these can be mentioned in the commit message, which can end up 
>> in the
>> minor release notes as a recommendation to rerun ANALYZE.
>>
> 
> Good point. I pushed the 0002 part and added a short paragraph 
> suggesting ANALYZE might be necessary. I did not go into details about 
> the individual cases, because that'd be too much for a commit message.
> 
>> Thanks for pushing 0001.
>>
> 
> Thanks for posting the patches!
> 
> I've pushed the second part - attached are the two remaining parts. I'll 
> wait a bit before pushing the rebased 0001 (which goes into master 
> branch only). Not sure about 0002 - I'm not convinced the refactored ACL 
> checks are an improvement, but I'll think about it.
> 

BTW when backpatching the first part, I had to decide what to do about 
tests. The 10 & 11 branches did not have the check_estimated_rows() 
function. I considered removing the tests, reworking the tests not to 
need the function, or adding the function. Ultimately I added the 
function, which seemed like the best option.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Commits

  1. Add stxdinherit flag to pg_statistic_ext_data

  2. Build inherited extended stats on partitioned tables

  3. Ignore extended statistics for inheritance trees

  4. Don't build extended statistics on inheritance trees

  5. Tighten up relation kind checks for extended statistics

  6. Avoid assuming that statistics for a parent relation reflect the properties of