pgBouncer

pgBouncer is a lightweight connection poooler and state manager that provides an efficient gateway to metering connections to PostgreSQL. The PostgreSQL Operator provides an integration with pgBouncer that allows you to deploy it alongside your PostgreSQL cluster.

This tutorial covers how you can set up pgBouncer, functionality that the PostgreSQL Operator provides to manage it, and more.

Setup pgBouncer

pgBouncer lives as an independent Deployment next to your PostgreSQL cluster but, thanks to the PostgreSQL Operator, is synchronized with various aspects of your environment.

There are two ways you can set up pgBouncer for your cluster. You can add pgBouncer when you create your cluster, e.g.:

pgo create cluster hippo --pgbouncer

or after your PostgreSQL cluster has been provisioned with the pgo create pgbouncer:

pgo create pgbouncer hippo

There are several managed objects that are created alongside the pgBouncer Deployment, these include:

  • The pgBouncer Deployment itself
    • One or more pgBouncer Pods
  • A pgBouncer ConfigMap, e.g. hippo-pgbouncer-cm which has two entries:
    • pgbouncer.ini, which is the configuration for the pgBouncer instances
    • pg_hba.conf, which controls how clients can connect to pgBouncer
  • A pgBouncer Secret e.g. hippo-pgbouncer-secret, that contains the following values:
    • password: the password for the pgbouncer user. The pgbouncer user is described in more detail further down.
    • users.txt: the description for how the pgbouncer user and only the pgbouncer user can explicitly connect to a pgBouncer instance.

The pgbouncer user

The pgbouncer user is a special type of PostgreSQL user that is solely for the administration of pgBouncer. It performs several roles, including:

  • Securely load PostgreSQL user credentials into pgBouncer so pgBouncer can perform authentication and connection forwarding
  • The ability to log into pgBouncer itself for administration, introspection, and looking at statistics

The pgBouncer user is not meant to be used to log into PostgreSQL directly: the account is given permissions for ad hoc tasks. More information on how to connect to pgBouncer is provided in the next section.

Connect to a Postgres Cluster Through pgBouncer

Connecting to a PostgreSQL cluster through pgBouncer is similar to how you connect to PostgreSQL directly, but you are connecting through a different service. First, note the types of users that can connect to PostgreSQL through pgBouncer:

  • Any regular user that’s created through pgo create user or a user that is not a system account.
  • The postgres superuser

The following example will follow similar steps for how you would connect to a Postgres Cluster via psql, but applies to all other connection methods.

First, get a list of Services that are available in your namespace:

kubectl -n pgo get svc

You should see a list similar to:

NAME                         TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)                      AGE
hippo                        ClusterIP   10.96.104.207   <none>        2022/TCP,5432/TCP            12m
hippo-backrest-shared-repo   ClusterIP   10.96.134.253   <none>        2022/TCP                     12m
hippo-pgbouncer              ClusterIP   10.96.85.35     <none>        5432/TCP                     11m

We are going to want to create a port forward to the hippo-pgbouncer service. In a separate terminal window, run the following command:

kubectl -n pgo port-forward svc/hippo-pgbouncer 5432:5432

Recall in the earlier part of the tutorial that we created a user called testuser with a password of securerandomlygeneratedpassword. We can the connect to PostgreSQL via pgBouncer by executing the following command:

PGPASSWORD=securerandomlygeneratedpassword psql -h localhost -p 5432 -U testuser hippo

You should then be greeted with the PostgreSQL prompt:

psql (12.10)
Type "help" for help.

hippo=>

Validation: Did this actually work?

This looks just like how we connected to PostgreSQL before, so how do we know that we are connected to PostgreSQL via pgBouncer? Let’s log into pgBoucner as the pgbouncer user and demonstrate this.

In another terminal window, get the credential for the pgBouncer user. This can be done with the pgo show pgbouncer command:

pgo show pgbouncer hippo

which yields something that looks like:

CLUSTER SERVICE         USERNAME  PASSWORD                 CLUSTER IP  EXTERNAL IP
------- --------------- --------- ------------------------ ----------- -----------
hippo   hippo-pgbouncer pgbouncer randompassword           10.96.85.35             

Copy the actual password and log into pgbouncer with the following command:

PGPASSWORD=randompassword psql -h localhost -p 5432 -U pgbouncer pgbouncer

You should see something similar to this:

psql (12.10, server 1.14.0/bouncer)
Type "help" for help.

pgbouncer=#

In the pgboucner terminal, run the following command. This will show you the overall connection statistics for pgBouncer:

SHOW stats;

Success, you have connected to pgBouncer!

Customize CPU / Memory for pgBouncer

Provisioning

The PostgreSQL Operator provides several flags for pgo create cluster to help manage resources for pgBouncer:

  • --pgbouncer-cpu: Specify the CPU Request for pgBouncer
  • --pgbouncer-cpu-limit: Specify the CPU Limit for pgBouncer
  • --pgbouncer-memory: Specify the Memory Request for pgBouncer
  • --pgbouncer-memory-limit: Specify the Memory Limit for pgBouncer

Additional, the PostgreSQL Operator provides several flags for pgo create pgbouncer to help manage resources for pgBouncer:

  • --cpu: Specify the CPU Request for pgBouncer
  • --cpu-limit: Specify the CPU Limit for pgBouncer
  • --memory: Specify the Memory Request for pgBouncer
  • --memory-limit: Specify the Memory Limit for pgBouncer

To create a pgBouncer Deployment that makes a CPU Request of 1.0 with a CPU Limit of 2.0 and a Memory Request of 64Mi with a Memory Limit of 256Mi:

pgo create pgbouncer hippo \
  --cpu=1.0 --cpu-limit=2.0 \
  --memory=64Mi --memory-limit=256Mi

Updating

You can also add more memory and CPU resources to pgBouncer with the pgo update pgbouncer command, including:

  • --cpu: Specify the CPU Request for pgBouncer
  • --cpu-limit: Specify the CPU Limit for pgBouncer
  • --memory: Specify the Memory Request for pgBouncer
  • --memory-limit: Specify the Memory Limit for pgBouncer

For example, to update a pgBouncer to a CPU Request of 2.0 with a CPU Limit of 3.0 and a Memory Request of 128Mi with a Memory Limit of 512Mi:

pgo update pgbouncer hippo \
  --cpu=2.0 --cpu-limit=3.0 \
  --memory=128Mi --memory-limit=512Mi

Scaling pgBouncer

You can add more pgBouncer instances when provisioning pgBouncer and to an existing pgBouncer Deployment.

Provisioning

To add pgBouncer instances when creating a PostgreSQL cluster, use the --pgbouncer-replicas flag on pgo create cluster. For example, to add 2 replicas:

pgo create cluster hippo --pgbouncer --pgbouncer-replicas=2

If adding a pgBouncer to an already provisioned PostgreSQL cluster, use the --replicas flag on pgo create pgbouncer. For example, to add a pgBouncer instance with 2 replicas:

pgo create pgbouncer hippo --replicas=2

Updating

To update pgBouncer instances to scale the replicas, use the pgo update pgbouncer command with the --replicas flag. This flag can scale pgBouncer up and down. For example, to run 3 pgBouncer replicas:

pgo update pgbouncer hippo --replicas=3

Rotate pgBouncer Password

If you wish to rotate the pgBouncer password, you can use the --rotate-password flag on pgo update pgbouncer:

pgo update pgbouncer hippo --rotate-password

This will change the pgBouncer password and synchronize the change across all pgBouncer instances.

Next Steps

Now that you have connection pooling set up, let’s create a high availability PostgreSQL cluster!