MIME-Version: 1.0
From: Bob Ross <bob.ross.19821@gmail.com>
Date: Wed, 8 Oct 2025 08:30:50 +0200
Message-ID: 
 <CAHtZvrdvvk5YyyQMYoaE-2SY6J0SMQgLSgU1FvztGnAKLPRMLg@mail.gmail.com>
Subject: SQL query latency when using pgPool (v4.6.2)
To: pgpool-general@lists.postgresql.org
Content-Type: multipart/alternative; boundary="000000000000fbffdd06409fd3d8"
Archived-At: 
 <https://www.postgresql.org/message-id/CAHtZvrdvvk5YyyQMYoaE-2SY6J0SMQgLSgU1FvztGnAKLPRMLg%40mail.gmail.com>
Precedence: bulk

--000000000000fbffdd06409fd3d8
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hello,

I am currently testing SQL query performance using PgPool (v4.6.2) as a
load balancer, and I am observing approximately 18=E2=80=9320 ms latency be=
tween
SQL statements within the same transaction. Execution time on the
PostgreSQL backend servers is less than 0.5 ms.

It=E2=80=99s important to note that both the application and database serve=
rs are
located in the same cloud region, with network latency below 1 ms. PgPool
is colocated with the PostgreSQL backend servers.

For example, when executing a transaction containing 10 SQL statements, the
total execution time through pgPool is about 200 ms, as confirmed by both
application and pgPool logs. The delay consistently occurs between the
=E2=80=9CExecute=E2=80=9D and =E2=80=9CParse=E2=80=9D phases. When connecti=
ng directly to the PostgreSQL
backends, the same transaction completes in about 10 ms, with roughly 1 ms
between SQL statements.

We have tested the following configurations:
- Switched from the extended query protocol to the simple query protocol =
=E2=80=94
for some queries this reduced inter-query latency to 2=E2=80=933 ms, but fo=
r others
it increased to 30=E2=80=9340 ms due to additional parsing overhead.
- Adjusted various PgPool parameters related to load balancing and
connection pooling (e.g., load_balance_mode, statement_level_load_balance,
serialize_accept, disable_load_balance_on_write) and enabled debug logging,
but none of these changes brought a significant improvement.

Could you please advise whether there are additional pgPool parameters or
tuning approaches that could help improve throughput and reduce latency in
extended query protocol?

Thank you,
Bob

--000000000000fbffdd06409fd3d8
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><p>Hello,</p>
<p>I am currently testing SQL query performance using PgPool (v4.6.2) as a =
load balancer, and I am observing approximately 18=E2=80=9320 ms latency be=
tween SQL statements within the same transaction. Execution time on the Pos=
tgreSQL backend servers is less than<b> </b>0.5 ms.</p>
<p>It=E2=80=99s important to note that both the application and database se=
rvers are located in the same cloud region, with network latency below 1 ms=
. PgPool is colocated with the PostgreSQL backend servers.</p>
<p>For example, when executing a transaction containing 10 SQL statements, =
the total execution time through pgPool is about 200 ms, as confirmed by bo=
th application and pgPool logs. The delay consistently occurs between the =
=E2=80=9CExecute=E2=80=9D and =E2=80=9CParse=E2=80=9D phases. When connecti=
ng directly to the PostgreSQL backends, the same transaction completes in a=
bout 10 ms, with roughly 1 ms between SQL statements.</p>
<p>We have tested the following configurations:<br>- Switched from the<b> <=
/b>extended query protocol<b> </b>to the simple query protocol =E2=80=94 fo=
r some queries this reduced inter-query latency to 2=E2=80=933 ms, but for =
others it increased to 30=E2=80=9340 ms due to additional parsing overhead.=
<br>- Adjusted various PgPool parameters related to load balancing and conn=
ection pooling (e.g., load_balance_mode, statement_level_load_balance, seri=
alize_accept, disable_load_balance_on_write) and enabled debug logging, but=
 none of these changes brought a significant improvement.</p>
<p>Could you please advise whether there are additional pgPool parameters o=
r tuning approaches that could help improve throughput and reduce latency i=
n extended query protocol?=C2=A0</p>
<p>Thank you,<br>Bob</p></div>

--000000000000fbffdd06409fd3d8--