Horton

Re: pg_stat_statements and "IN" conditions

Dmitry Dolgov <9erthalion6@gmail.com>

View thread
From: Dmitry Dolgov <9erthalion6@gmail.com>
To: PostgreSQL-development <pgsql-hackers@postgresql.org>, Tom Lane <tgl@sss.pgh.pa.us>, Greg Stark <stark@mit.edu>, Pavel Trukhanov <pavel.trukhanov@gmail.com>
Date: 2020-11-18T16:04:32Z
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. Introduce squashing of constant lists in query jumbling

  2. Make documentation builds reproducible

  3. Include values of A_Const nodes in query jumbling

  4. Teach planner about more monotonic window functions

  5. Split up guc.c for better build speed and ease of maintenance.

Attachments

> On Wed, Aug 12, 2020 at 06:19:02PM +0200, Dmitry Dolgov wrote:
>
> I would like to start another thread to follow up on [1], mostly to bump up the
> topic. Just to remind, it's about how pg_stat_statements jumbling ArrayExpr in
> queries like:
>
>     SELECT something FROM table WHERE col IN (1, 2, 3, ...)
>
> The current implementation produces different jumble hash for every different
> number of arguments for essentially the same query. Unfortunately a lot of ORMs
> like to generate these types of queries, which in turn leads to
> pg_stat_statements pollution. Ideally we want to prevent this and have only one
> record for such a query.
>
> As the result of [1] I've identified two highlighted approaches to improve this
> situation:
>
> * Reduce the generated ArrayExpr to an array Const immediately, in cases where
>   all the inputs are Consts.
>
> * Make repeating Const to contribute nothing to the resulting hash.
>
> I've tried to prototype both approaches to find out pros/cons and be more
> specific. Attached patches could not be considered a completed piece of work,
> but they seem to work, mostly pass the tests and demonstrate the point. I would
> like to get some high level input about them and ideally make it clear what is
> the preferred solution to continue with.

I've implemented the second approach mentioned above, this version was
tested on our test clusters for some time without visible issues. Will
create a CF item and would appreciate any feedback.