MIME-Version: 1.0
References: 
 <CAD7nQBD1iy3NAS2YRDUH+zk0g=50px_h_wRsLcFkai7R_RF6VQ@mail.gmail.com>
 <202601281248.qjv5oru7hxlz@alvherre.pgsql>
 <CAD7nQBCfgCeTwO=45OXhGc=wpYBrWXJfzJKRy+T26_jALxatLg@mail.gmail.com>
 <CAHGQGwEaBw9dd+Nzb2ETDJ-40nJTo6GRXrG_FcP0rmw3_-qU_Q@mail.gmail.com>
 <CAJpy0uC3aJzu_1zsaShL0UAy763MuJ3KOBZORb-a842NQ41tpg@mail.gmail.com>
In-Reply-To: 
 <CAJpy0uC3aJzu_1zsaShL0UAy763MuJ3KOBZORb-a842NQ41tpg@mail.gmail.com>
From: Masahiko Sawada <sawada.mshk@gmail.com>
Date: Mon, 23 Mar 2026 17:13:57 -0700
Message-ID: 
 <CAD21AoCB31azOSYNSdGfSi8reUzbPaj+e3C7uMC=rvLzeqMsNA@mail.gmail.com>
Subject: Re: [PATCH] Add max_logical_replication_slots GUC
To: shveta malik <shveta.malik@gmail.com>
Cc: Fujii Masao <masao.fujii@gmail.com>,
 Ahmed Et-tanany <ahmed.ettanany@aiven.io>,
	=?UTF-8?Q?=C3=81lvaro_Herrera?= <alvherre@kurilemu.de>,
	pgsql-hackers@lists.postgresql.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CAD21AoCB31azOSYNSdGfSi8reUzbPaj%2Be3C7uMC%3DrvLzeqMsNA%40mail.gmail.com>
Precedence: bulk

Hi,

On Mon, Mar 9, 2026 at 11:55=E2=80=AFPM shveta malik <shveta.malik@gmail.co=
m> wrote:
>
> On Thu, Jan 29, 2026 at 5:10=E2=80=AFPM Fujii Masao <masao.fujii@gmail.co=
m> wrote:
> >
> > On Wed, Jan 28, 2026 at 10:02=E2=80=AFPM Ahmed Et-tanany
> > <ahmed.ettanany@aiven.io> wrote:
> > >
> > > Yes, that's what I meant.
> >
> > Would something like max_logical_wal_senders also be needed for your pu=
rpose?
> > Otherwise, logical replication connections could exhaust max_wal_sender=
s and
> > prevent physical replication connections from being established.
> >
>
> I could be mistaken, but I haven=E2=80=99t found a way to start a logical
> replication stream without a replication slot. A replication
> connection and walsender can exist without a slot, for example:
> ./psql "host=3Dlocalhost port=3D5432 user=3Duser1 dbname=3Dpostgres
> replication=3Ddatabase"
>
> However, converting that connection to logical replication requires a
> slot from the max_logical_replication_slots pool. If that pool is
> exhausted, the connection cannot be converted =E2=80=94 in which case,
> wouldn=E2=80=99t a single GUC suffice?

In an extreme case like where if all walsenders are used up by logical
walsender who are just connecting and not using logical replication
slots, physical replication cannot start even if there is a free
physical replication slot. But I think it's sufficient to have
something like max_logical_replication_slots in most cases. So a
single GUC seems to suffice unless I'm not missing some cases.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com