Date: Tue, 24 Mar 2026 14:06:23 +0000 (GMT)
From: PALAYRET Jacques <jacques.palayret@meteo.fr>
To: pgsql-general@lists.postgresql.org
Message-ID: <531047943.370846639.1774361183685.JavaMail.zimbra@meteo.fr>
Subject: Logical replication in PostgreSQL Amount of subscriber vs publisher
 WALs
MIME-Version: 1.0
Content-Type: multipart/alternative; 
	boundary="=_68fd56d3-467e-4e7c-b806-e356833cbe7c"
Thread-Index: t5LlkICwqsRtXc/OMOLXOapVC6QOVA==
Thread-Topic: Logical replication in PostgreSQL Amount of subscriber vs publisher WALs
Archived-At: <https://www.postgresql.org/message-id/531047943.370846639.1774361183685.JavaMail.zimbra%40meteo.fr>
Precedence: bulk

--=_68fd56d3-467e-4e7c-b806-e356833cbe7c
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable

Hello,=20

The amount of WAL generated by an SQL INSERT or UPDATE statement applied to=
 a table with multiple indexes can be much greater than the size of the tab=
le (table + index).=20
For example, an INSERT statement in an empty table (with 3 indexes) can gen=
erate WALs twice the size of the table (table + index).=20
This difference (even for an INSERT) may seem surprising, but it's understa=
ndable.=20

What's less intuitive is that, according to my tests, with logical replicat=
ion in PostgreSQL, the amount of WAL generated by an SQL statement can be v=
ery different between the subscriber server (the replica) and the publisher=
 server (the provider).=20
Is this accurate? Sometimes 1.5 or 2 times greater?=20

Regards=20
----- M=C3=A9t=C3=A9o-France -----=20
PALAYRET Jacques=20
DCSC/GDC=20
jacques.palayret@meteo.fr=20
Fixe : +33 561078319=20

--=_68fd56d3-467e-4e7c-b806-e356833cbe7c
Content-Type: text/html; charset=utf-8
Content-Transfer-Encoding: quoted-printable

<html><body><div style=3D"font-family: garamond, new york, times, serif; fo=
nt-size: 12pt; color: #000000"><div>Hello,</div><div><br data-mce-bogus=3D"=
1"></div><div>The amount of WAL generated by an SQL INSERT or UPDATE statem=
ent applied to a table with multiple indexes can be much greater than the s=
ize of the table (table + index).<br>For example, an INSERT statement in an=
 empty table (with 3 indexes) can generate WALs twice the size of the table=
 (table + index).<br>This difference (even for an INSERT) may seem surprisi=
ng, but it's understandable.</div><div><br>What's less intuitive is that, a=
ccording to my tests, with logical replication in PostgreSQL, the amount of=
 WAL generated by an SQL statement can be very different between the subscr=
iber server (the replica) and the publisher server (the provider).<br>Is th=
is accurate? Sometimes 1.5 or 2 times greater?</div><div><br></div><div>Reg=
ards<br data-mce-bogus=3D"1"></div><div data-marker=3D"__SIG_PRE__">----- M=
=C3=A9t=C3=A9o-France -----<br>PALAYRET Jacques<br>DCSC/GDC<br>jacques.pala=
yret@meteo.fr<br>Fixe&nbsp;:&nbsp;+33 561078319</div></div></body></html>
--=_68fd56d3-467e-4e7c-b806-e356833cbe7c--