CloudNativePG is an open source operator designed to manage PostgreSQL workloads on any supported Kubernetes cluster running in private, public, hybrid, or multi-cloud environments.
CloudNativePG was originally built and sponsored by EDB.
The best way to get started is with the "Quickstart" section in the documentation.
The goal of CloudNativePG is to increase the adoption of PostgreSQL, one of the most loved DBMS in traditional VM and bare metal environments, inside Kubernetes, thus making the database an integral part of the development process and GitOps CI/CD automated pipelines.
CloudNativePG has been designed by Postgres experts with Kubernetes administrators in mind. Put simply, it leverages Kubernetes by extending its controller and by defining, in a programmatic way, all the actions that a good DBA would normally do when managing a highly available PostgreSQL database cluster.
Since the inception, our philosophy has been to adopt a Kubernetes native approach to PostgreSQL cluster management, making incremental decisions that would answer the fundamental question: "What would a Kubernetes user expect from a Postgres operator?".
The most important decision we made is to have the status of a PostgreSQL
cluster directly available in the Cluster resource, so to inspect it through
the Kubernetes API. We've fully embraced the operator pattern and eventual
consistency, two of the core principles upon which Kubernetes is built for
managing complex applications.
As a result, the operator is responsible for managing the status of the
Cluster resource, keeping it up to date with the information that each
PostgreSQL instance manager regularly reports back through the API server. Such
changes might trigger, for example, actions like:
-
a PostgreSQL failover where, after an unexpected failure of a cluster's primary instance, the operator itself elects the new primary, updates the status, and directly coordinates the operation through the reconciliation loop, by relying on the instance managers
-
scaling up or down the number of read-only replicas, based on a positive or negative variation in the number of desired instances in the cluster, so that the operator creates or removes the required resources to run PostgreSQL, such as persistent volumes, persistent volume claims, pods, secrets, config maps, and then coordinates cloning and streaming replication tasks
-
updates of the endpoints of the PostgreSQL services that applications rely on to interact with the database, as Kubernetes represents the single source of truth and authority
-
updates of container images in a rolling fashion, following a change in the image name, by first updating the pods where replicas are running, and then the primary, issuing a switchover first
The latter example is based on another pillar of CloudNativePG: immutable application containers - as explained in the blog article "Why EDB Chose Immutable Application Containers".
The above list can be extended. However, the gist is that CloudNativePG exclusively relies on the Kubernetes API server and the instance manager to coordinate the complex operations that need to take place in a business continuity PostgreSQL cluster, without requiring any assistance from an intermediate management tool responsible for high availability and failover management like similar open source operators.
CloudNativePG also manages additional resources to help the Cluster resource
manage PostgreSQL - currently ScheduledBackup, Pooler, and Backup.
Fully embracing Kubernetes also means that, in case of failure of the whole Kubernetes cluster, the operator won’t do anything, postponing any decision to when the cluster is back up again. In the meantime, Postgres instances should continue running according to the last known state of the cluster.
CloudNativePG is exclusively focused on the PostgreSQL database management system maintained by the PostgreSQL Global Development Group (PGDG). We are not currently considering adding to CloudNativePG extensions or capabilities that are included in forks of the PostgreSQL database management system, unless in the form of extensible or pluggable frameworks.
CloudNativePG doesn't intend to pursue database independence (e.g. control a MariaDB cluster).
- Data on Kubernetes (DoK) Community
- "How CloudNativePG manages physical replication slots for PostgreSQL in Kubernetes" by Gabriele Bartolini (September 2023)
- "The Current State of Major PostgreSQL Upgrades with CloudNativePG" by Gabriele Bartolini (August 2023)
- "The Rise of the Kubernetes Native Database" by Jeff Carpenter (December 2022)
- "Why Run Postgres in Kubernetes?" by Gabriele Bartolini (May 2022)
- "Shift-Left Security: The Path To PostgreSQL On Kubernetes" by Gabriele Bartolini (April 2021)
- "Local Persistent Volumes and PostgreSQL usage in Kubernetes" by Gabriele Bartolini (June 2020)
The current maintainers of the CloudNativePG project are:
- Gabriele Bartolini (EDB)
- Francesco Canovai (EDB)
- Leonardo Cecchi (EDB)
- Jonathan Gonzalez (EDB)
- Marco Nenciarini (EDB)
- Armando Ruocco (EDB)
- Philippe Scorsolini (upbound)
They are listed in the CODEOWNERS file.
A list of publicly known users of the CloudNativePG operator is in ADOPTERS.md. Help us grow our community and CloudNativePG by adding yourself and your organization to this list!
Postgres, PostgreSQL and the Slonik Logo are trademarks or registered trademarks of the PostgreSQL Community Association of Canada, and used with their permission.