Postgres connection pool on Kubernetes in 1 minute

JM Robles

2 min read

·

Apr 25, 2021

--

2

Share

TL;DR: How to deploy pgbouncer on Kubernetes to allow use more database connections

Zoom image will be displayed

Motivation

If you develop apps using frameworks such as Django o RoR, you have probably faced the following problem/error.

FATAL: sorry, too many clients already

As you know, these kind of frameworks uses a database connection pool with the objective of to reduce the database connection latency.

That’s is great always which your database is optimized to have a lot of connections.

As you can infer, it’s not the case of Postgres.

Each Postgres connection uses about 10 MB, altough most of the time they are idle.

With the microservices boom or gRPC streams, things get worse. We have a lot of idle connections but Postgres requests a lot of resources for nothing, to put in a idle state.

PgBouncer to the rescue

There are serveral options to solve the multiple connections problem but all of them use the same pattern: a proxy in the middle.

The idea is that you connect your consumer software to a proxy which allows a lot of “cheap” connections, and this proxy connects with the Postgres database only when your…