From: Tom Lane <tgl@sss.pgh.pa.us>
To: "Benjamin Coutu" <ben.coutu@zeyos.com>
cc: pgsql-performance@postgresql.org
Subject: Re: Unclamped row estimates whith OR-ed subplans
In-reply-to: <20200619151232.7F3285FB26@mx.zeyos.com>
References: <20200619151232.7F3285FB26@mx.zeyos.com>
Comments: In-reply-to "Benjamin Coutu" <ben.coutu@zeyos.com>
	message dated "Fri, 19 Jun 2020 17:12:31 +0200"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1666225.1592582687.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Fri, 19 Jun 2020 12:04:47 -0400
Message-ID: <1666226.1592582687@sss.pgh.pa.us>
Precedence: bulk

"Benjamin Coutu" <ben.coutu@zeyos.com> writes:
> please consider the following SQL query:

> SELECT * FROM "transactions" WHERE =

> 	"account" IN (SELECT "ID" FROM "accounts" WHERE "name" ~~* '%test%') OR
> 	"contract" IN (SELECT "ID" FROM "contracts" WHERE "name" ~~* '%test%')

> This yields the following plan on Postgres 11:

> Seq Scan on transactions  (cost=3D67.21..171458.03 rows=3D1301316 width=3D=
1206)
>   Filter: ((hashed SubPlan 1) OR (hashed SubPlan 2))
>   SubPlan 1
>     ->  Bitmap Heap Scan on accounts  (cost=3D33.36..61.16 rows=3D46 wid=
th=3D4)
>           Recheck Cond: ((name)::text ~~* '%test%'::text)
>           ->  Bitmap Index Scan on s_accounts  (cost=3D0.00..33.35 rows=3D=
46 width=3D0)
>                 Index Cond: ((name)::text ~~* '%test%'::text)
>   SubPlan 2
>     ->  Seq Scan on contracts  (cost=3D0.00..5.93 rows=3D5 width=3D4)
>           Filter: ((name)::text ~~* '%test%'::text)

> So the where clause of this query has just two subplans OR-ed together, =
one is estimated to yield 46 rows and one is estimated to yield 5 rows.
> I'd expect the total rows for the seqscan to be estimated at 46 then, fo=
llowing the logic that rows_seqscan =3D max(rows_subplan1, rows_subplan2).=
 As you can see, the optimizer estimates a whopping 1301316 rows instead.

No.  The subplan estimates are for the number of rows produced by one
execution of the subplan, ie the numbers of "accounts" or "contracts"
rows that match those inner WHERE conditions.  This has very little
a-priori relationship to the number of "transactions" rows that will
satisfy the outer WHERE condition.  If we knew that transactions.account
and transactions.contract were unique columns, then we might be able
to say that there shouldn't be more than one outer match per subplan
result row ... but you didn't say that, and it seems unlikely.

(Having said that, I think that the estimates for these cases very
possibly are quite stupid.  But that doesn't mean that 46+5 would
be the right answer.)

			regards, tom lane