Re: [HACKERS] Cached plans and statement generalization

On 02.08.2018 08:25, Yamaji, Ryo wrote:
>> -----Original Message-----
>> From: Konstantin Knizhnik [mailto:k.knizhnik@postgrespro.ru]
>> Sent: Wednesday, August 1, 2018 4:53 PM
>> To: Yamaji, Ryo/山地 亮 <yamaji.ryo@jp.fujitsu.com>
>> Cc: PostgreSQL mailing lists <pgsql-hackers@postgresql.org>
>> Subject: Re: [HACKERS] Cached plans and statement generalization
>>
>> I failed to reproduce the problem.
>> I used the following non-default configuration parameters:
>>
>> autoprepare_limit=1
>> autoprepare_threshold=1
>>
>> create dummy database:
>>
>> create table foo(x integer primary key, y integer); insert into foo values
>> (generate_series(1,10000), 0);
>>
>> and run different queries,  like:
>>
>> postgres=# select *  from foo where x=1; postgres=# select *  from foo
>> where x+x=1; postgres=# select *  from foo where x+x+x=1; postgres=#
>> select *  from foo where x+x+x+x=1; ...
>>
>> and check size of CacheMemoryContext using gdb - it is not increased.
>>
>> Can you please send me your test?
>
> I checked not CacheMemoryContext but "plan cache context".
> Because I think that the memory context that autoprepare mainly uses is "plan cache context".
>
> Non-default configuration parameter was used as well as Konstantin.
> autoprepare_limit=1
> autoprepare_threshold=1
>
> The procedure of the test that I did is shown below.
>
> create dummy database
> create table test (key1 integer, key2 integer, ... , key100 integer);
> insert into test values (1, 2, ... , 100);
>
> And, different queries are executed.
> select key1 from test where key1=1 and key2=2 and ... and key100=100;
> select key2 from test where key1=1 and key2=2 and ... and key100=100;
> select key3 from test where key1=1 and key2=2 and ... and key100=100;...
>
> And, "plan cache context" was confirmed by using gdb.
>
>


Thank you.
Unfortunately expression_tree_walker is not consistent with copyObject: 
I tried to use this walker to destroy raw parse tree of autoprepared 
statement, but looks
like some nodes are not visited by expression_tree_walker. I have to 
create separate memory context for each autoprepared statement.
New version  of the patch is attached.