Where does PGBouncer sit on a Kubernetes Cluster

Question

I'm wanting to deploy some sort of connection pooling within my applications across multiple nodes. Python naturally provides some pooling with the SQLAlchemy implementation; but obviously this probably does not scale as well as 1 central pool.

As far as I see it PGBouncer can Sit within each pod, it can be running as its own pod, or it can be running as a daemonset which ties a PGBouncer instance to be running for each kube node.

How would you / how have you deployed PGBouncer in a similar environment?

This question is equally relevant for twemproxy.

Answer 1

PGBouncer is a connection pooler. So, one instance of PGBouncer should be enough in between the apps and postgres.