Re: A performance issue with Memoize - Mailing list pgsql-hackers

From Richard Guo
Subject Re: A performance issue with Memoize
Date
Msg-id CAMbWs4_64wc4KggGEdTGv-BA1-EA6pmE1RE26fYdfFQNd79qgA@mail.gmail.com
Whole thread Raw
In response to Re: A performance issue with Memoize  (David Rowley <dgrowleyml@gmail.com>)
Responses Re: A performance issue with Memoize
List pgsql-hackers

On Fri, Jan 26, 2024 at 12:18 PM David Rowley <dgrowleyml@gmail.com> wrote:
On Fri, 26 Jan 2024 at 16:51, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> However ... it seems like we're not out of the woods yet.  Why
> >> is Richard's proposed test case still showing
> >> +         ->  Memoize (actual rows=5000 loops=N)
> >> +               Cache Key: t1.two, t1.two
> >> Seems like there is missing de-duplication logic, or something.
>
> > This seems separate and isn't quite causing the same problems as what
> > Richard wants to fix so I didn't touch this for now.
>
> Fair enough, but I think it might be worth pursuing later.

Here's a patch for that.

At first I wondered if we should assume that the same param expr must
have the same equality operator. If not, we should also check the
operator to tell if the cache key is a duplicate, like

-           if (!list_member(*param_exprs, expr))
+           if (!list_member(*param_exprs, expr) ||
+               !list_member_oid(*operators, hasheqoperator))

But after looking at how rinfo->left_hasheqoperator/right_hasheqoperator
is set, it seems we can assume that: the operator is from the type cache
entry which is fetched according to the expr datatype.

So I think the patch makes sense.  +1.

Thanks
Richard

pgsql-hackers by date:

Previous
From: Alexander Lakhin
Date:
Subject: Race condition in FetchTableStates() breaks synchronization of subscription tables
Next
From: Masahiko Sawada
Date:
Subject: Re: Small fix on COPY ON_ERROR document