What we've learned from running a PostgreSQL managed service on Kubernetes (DoK Day EU 2022) // Oleksii Kliukin

https://go.dok.community/slack

https://dok.community/

From the DoK Day EU 2022 (https://youtu.be/Xi-h4XNd5tE)

Kubernetes is an emerging platform of choice for deploying and running PostgresSQL. Deploying 100 Postgres clusters is as easy as deploying one, and there is no need to tinker with tools like Ansible or Puppet. Resource sharing can be applied when it makes sense, allowing to run multiple Postgres databases in isolation on a single instance, each storing the data on a dedicated persistent volume. There are great open-source tools out there to deal with high-availability and backups than support or can be easily integrated into the Kubernetes workflow. Monitoring and alerting is easy to implement. People reported success in running Postgres on Kubernetes before. But there are also rough edges, like memory management or certain Postgres maintenance operations, such as installing extensions, that normally cause unnecessary database downtimes on Kubernetes. They are less of a problem for in-house deployments, but may become a deciding factor when running a managed service, competing with other such services running on bare-metal servers or virtual machines that are free of those issues.

In this talk, I will share some of our learnings from running a managed PostgreSQL/TimescaleDB service on Kubernetes on AWS for a little more than a year: I’ll start with the motivation of running managed PostgreSQL on Kubernetes, the benefits and drawbacks. I’ll describe the architecture of the managed PostgreSQL cloud on Kubernetes I’ll zoom in on how we solved some of the Kubernetes-specific issues within our cloud, such as upgrading extensions without downtimes, taming the dreaded OOM killer, and doing regular maintenance and PostgreSQL major upgrades. I’ll share how open-source tools from the PostgreSQL ecosystem helps us to run the service and explain how we use them in a slightly non-trivial way.

Oleksii has been working with PostgresSQL for almost 20 years, and has been deploying Postgres on Kubernetes since 2016, when his team at Zalando started the internal managed PostgreSQL service based on the in-house and open-source postgresql-operator. Around 2015, with some other team members, he stared working on a PostgreSQL HA project that later became Patroni. Long before that he was hacking PosgreSQL source code to implement binary replication on PostgreSQL 7.x, authoring some PostgreSQL extensions and contributing to the core PostgreSQL itself. He started PosgreSQL meetups in Berlin in 2015 and hopes to get back to meeting in-person somewhere in 2022.

Being Ukrainian, he lives in Berlin for a bit more than 9 years with his wife, two children and numerous plants