Testkube

Throwback to some of our favorite KubeJam moments 🎸 KubeJam is the cloud native community’s open jam session, and it’s coming back for KubeCon Salt Lake City. Bring your instrument or borrow one. Pick up a guitar, sit in on drums, or come hang out and listen. Some of the best conversations of KubeCon week happen between sets. 📍 Publik Space 📅 Wednesday, November 11 ⏰ 6 to 10 PM MT 🎟️ Save your spot: kubejam.com #KubeCon #CloudNativeCon #KubeJam #Testkube

It's that time of year again - KubeCon NA is just around the corner - as is the Kubejam! (secure your spot at kubejam.com!). If you're on-site in SLC and wanna chat about anything from prompts to chords, crds to scales or tests to frets - hit me up and we'll make it work! 🎸

The run button is the least important part of a test platform once ninety nine percent of the runs are started by automation. A data platform company I spoke with last week has grown from one monorepo to about 350 repositories. They sit at roughly 5,000 workflows and 100,000 executions, and almost nobody clicks anything to start them. That changes what the platform has to be good at. The screen a person sees matters when they read a failure. Whether the platform survives depends on everything upstream of the click. How a run gets triggered, and under what identity. What happens when two hundred of them arrive in the same minute, and who waits. Where the result lands once it exists, ideally on the pull request and nowhere a person has to remember to look. How a new repository gets onto the platform without filing a ticket. Teams tend to buy on the demo, and the demo is a person clicking run. A year later the platform is a queue, a set of triggers and a record of results, and the person clicking is the exception. On your platform, what share of test runs does a human start by hand? My guess is under ten percent, and I would like to hear if I am wrong.

Most failure alerts hand an engineer the start of an investigation instead of the end of one. A platform team running Testkube across more than twenty clusters has a good alert today. The Slack message names the workflow, the execution, the logs and the artifacts. Someone still has to open it and decide which kind of failure it is. There are three kinds. The environment was not ready, the workflow was wrong, or the product broke. The first two rarely need a senior engineer, and they are where most of the reading time goes. So the team is building an agent to do the sorting. It reads the logs and the execution through the MCP server, decides environment or workflow, then adjusts the workflow and reruns it, or posts a summary back to the thread. The engineer gets pulled in for the third kind. My view is that the hard part is the permission layer. An agent that can rerun workflows and open pull requests needs a defined scope, a token that stays where it was issued, and a record of what it did. The team says that part is still under construction, and I trust that more than a polished demo. Scope, token, record. Those three decide whether an agent gets near production workflows.

KubeJam is back, and this time we’re plugging in at KubeCon Salt Lake City 🎸 Step away from the booths and into the jam. Grab a guitar, take over the drums, hop on the keys, or just come for the music and the good people. Shredders, first-timers and air-guitar pros all welcome. 📍 Publik Space, Salt Lake City 🗓️ Wednesday, November 11 · 6–10 PM 🎟️ Save your spot: https://kubejam.com Tag the bandmate you’re bringing. #KubeCon #CloudNativeCon #Kubernetes #DevOps #KubeJam

Did the test fail, or did the infrastructure? Testkube’s October release adds failure categories to the executions list, so you can tell without opening the logs. You can also filter by failure type and trigger different webhooks for each category. Also included: AI Test Creation for on-prem, updated user and permission management, dependency caching between executions, and the Testkube for GitHub Chrome extension. Read the release post: https://lnkd.in/e6KNhqaQ #Testkube #TestAutomation #SoftwareTesting #AI

Tests are where a lot of good pipelines quietly go wrong. Usually it isn't the tests. It's where they end up living: a script at the bottom of a CI job, one file at a time, with the results buried in a log nobody reads. A while back I ran a Testkube proof of concept with a team of test engineers. They wanted speed, parallel runs and one place to see every result. They got all three. What won me over was something they didn't care about at all: Testkube is Kubernetes-native. A test is a custom resource with a schema, reviewed in git and run by a controller, just like everything else on a platform built around an API-driven control plane. So when I built Hangar, my home lab platform, Testkube went into the test stage. I just wrote up why: • Tests as declared resources, with no framework lock-in • Parallel workers, shards and matrices in a few lines of YAML • Any CI system can call it, so the tests outlive the pipeline • How Glidepath runs it today, fenced by a Kyverno policy • What running the free tier taught me, including reading the source when the docs disagreed. And a big thank you to the Testkube team, who were great to work with and took my feature requests seriously. That tells you a lot about a tool you'll live with for years. The full essay, with three new diagrams: https://lnkd.in/giTC7i3s #PlatformEngineering #Kubernetes #Testing #DevOps #Testkube

If your team is running AI-generated code, you need a way to test it in the same infrastructure it actually runs in. In this live session, Ole Lensmar and Lucio Daza will demonstrate how AI Test Creation can work for you. Register now to watch the live webinar. Oct 8, 2026 8AM PDT.

From the Testkube blog: Atulpriya Sharma on testing AI systems across all four layers. Model evals and red-teaming answer different questions. One checks whether the model can be tricked in the abstract. The other checks whether your specific prompts and guardrails hold up. You need both, plus infra and agent checks, with the results side by side. Read the blog; link in comments 👇 #Testkube #AITesting #DevOps #PlatformEngineering

"We tried whenever possible to elevate common patterns into something you could specify declaratively within the language." Alexander Buccino on the principle behind SandMan, the testing sandbox system his team built at Google. His advice for teams testing on Kubernetes today: "Languages that allow you to specify the components of your environment declaratively are very important." Alex joined Testkube CTO Ole Lensmar on the Cloud Native Testing Podcast. Full episode in comments 👇 #Testkube #CloudNative #SoftwareTesting #PlatformEngineering