Notes from Kubecon London 2025

Kubecon

What is it?

Kubernetes Convention is the premier expo for everything k8s, that is put on several times per year by the Cloud native Computing Foundation

My time there

I went to Kubecon Europe in London this year and had a blast absorbing all of the content. It’s easy to live in a bubble within your day job and miss out on some of the cool developments in technology.

I would definitely recommend Kubecon to anyone that works with Kubernetes. Not only do you get awesome talks and panels, but you can browse the showroom floor for k8s related projects and services.

Below I’ll link to some cool talks and their slideshows, but before that I’ll hit a few key points that I thought were most interesting.

Cool Notes

• The paradigm for k8s clusters is to favor multiple smaller clusters versus fewer larger clusters. Tools like Cluster API make it easy to manage numerous clusters. Small clusters are easier to upgrade as there is less chances for upgrades to have detrimental effects on running applications.
• Companies are utilizing components of k8s to do cool stuff
  • Metal3 and kubenet to provision physical hardware and devices
    • Use the reconciliation and scheduling components of k8s to configure and deploy hardware
  • Edge devices with kops to test algae in the ocean
• Kubernetes API Aggregation Layer can be used as an alternative/supplement to CRDs when you need to bypass the limitations of ETCD and utilize another database.
  • Kueue bypassed k8s object size limitation in ETCD and instead stores objects in Redis
• eBPF is the future, made easy by bpftime
• LLM/AI workflows work well in a k8s native environment
  • Kubeflow
  • Ray

Takeaways

Below is a some quick points I gathered from most of the talks.

• General

  • COTS vs Build

  • Commercial-off-the-shelf (COTS) comes up a lot in talks when it comes to DevOps and SRE teams.
    • Small organizations can get away with COTS since their limited footprint makes it easier to migrate to a home grown solution

  • COTS

    • In some cases, allows you to quickly get started. Provides mostly complete turnkey solution
    • Avoid COTS where possible, some solutions lack OSS solutions (Finance, HR, etc)
    • COTS creates vendor and data lock, perpetual costs
    • Still requires development

  • Build

    • Requires investment in hiring talent
    • Allows the most flexibility and control

• AI Agents

  • Incorporating AI for Operations is harder than regular operations

  • Potential use cases

    • Code review
      • On Git pushes can check for common mistakes
    • Run unit tests
    • IaC integration
      • Can determine which use 3rd party tools to test deploy infra and verify changes
    • Create PR’s, write documentation on changes

  • Challenges

    • LLMs are notoriosly unpredictable
    • While deploying, difficult to monitor
      • Monitoring which prompts provided bad output, determining which workflow

  • Solutions

  • LLM Guardrails
    • Slideshow
    • YouTube Talk