Re: BUG #16179: is it reasonable to callback pgss_post_parse_analyzeor pg_hint_plan_post_parse_analyze ??? - Mailing list pgsql-bugs

From Julien Rouhaud
Subject Re: BUG #16179: is it reasonable to callback pgss_post_parse_analyzeor pg_hint_plan_post_parse_analyze ???
Date
Msg-id CAOBaU_aNCoKx=y6gRB9Znd3RfNxjpAQLx=O+4j=sYfaA7ydJcA@mail.gmail.com
Whole thread Raw
In response to BUG #16179: is it reasonable to callback pgss_post_parse_analyze or pg_hint_plan_post_parse_analyze ???  (PG Bug reporting form <noreply@postgresql.org>)
Responses Re:Re: BUG #16179: is it reasonable to callbackpgss_post_parse_analyze or pg_hint_plan_post_parse_analyze ???  (Dam <zhq651@126.com>)
List pgsql-bugs
Hi,

On Wed, Dec 25, 2019 at 9:32 AM PG Bug reporting form
<noreply@postgresql.org> wrote:
>
> The following bug has been logged on the website:
>
> Bug reference:      16179
> Logged by:          DamionZ Zhao
> Email address:      zhq651@126.com
> PostgreSQL version: 11.5
> Operating system:   linux
> Description:
>
> When I only configure GUC as shared_preload_libraries = 'pg_hint_plan,
> pg_stat_statment'
> and have not run sql: 'create extension pg_hint_plan;  create extension
> pg_stat_statment',
> I expected that parse_analyze will not run into functions
> :pgss_post_parse_analyze or pg_hint_plan_post_parse_analyze
>
> In fact , it does not .  I print runtime call stack.
> -----------------------------------------------------
> statment 1: select
> #0  pgss_post_parse_analyze (pstate=0x2a45818, query=0x2a45928) at
> pg_stat_statements.c:812
> #1  0x00007fbe60506734 in pg_hint_plan_post_parse_analyze (pstate=0x2a45818,
> query=0x2a45928) at pg_hint_plan.c:2767
> #2  0x000000000058fbe5 in parse_analyze (parseTree=0x2a45798,
> sourceText=0x2a44a90 "select * from abcd;", paramTypes=0x0, numParams=0,
> queryEnv=0x0) at analyze.c:124
> [...]
>
> It will degrade performance and may cause the wrong result if it contains
> modify operation.

This is expected.  The hooks are called as soon as the libraries are
loaded, so configuring shared_preload_libraries (and restarting
postgres) will activate the extensions.  What you should do instead is
disable pg_stat_statements (or other extensions if they have such
possibility) if you want to be able to use it only when required,
without the need to restart postgres.  The CREATE EXTENSION part is
only necessary to access the counters stored in shared memory.

Regarding the performance overhead, it's true that the queryid
generation will still happen even if you disable pg_stat_statements.
However, note that this will be fixed with pg13, see
https://www.postgresql.org/message-id/BN8PR21MB1217B003C4F79DE230AA36B9B1580%40BN8PR21MB1217.namprd21.prod.outlook.com
for more details.



pgsql-bugs by date:

Previous
From: PG Bug reporting form
Date:
Subject: BUG #16179: is it reasonable to callback pgss_post_parse_analyze or pg_hint_plan_post_parse_analyze ???
Next
From: Загороднев Роман Евгеньевич
Date:
Subject: pg_upgrade