Engineering Blog

With the new Cluster API infrastructure provider for cloudscale (CAPCS), Kubernetes clusters become declarative resources you can provision, scale, upgrade, and delete through reconciliation. We cover the architecture and a hands-on walkthrough, from setup to Day 2 operations.

Self-hosting coding LLMs is possible today, but getting them to work well requires significant tuning. We share our experience running and evaluating models on cloudscale.

I believe close-to-bare-metal freedom with AI coding agents is possible and important. But I'd be doing you a disservice if I pretended it was easy, or that sovereignty alone makes you safe.

Snapshots make Kubernetes operations safer, but they are only one part of a Disaster Recovery strategy. In this Engineering Blog post, we show how Kubernetes workloads on cloudscale infrastructure can be restored even after losing an entire cluster, and what is required to turn snapshots into real recovery capabilities.

Since the price reduction of our Object Storage in February, usage grew significantly, including total storage consumption. This rapid growth surfaced some issues during load peaks and in our metrics pipeline.

Postgres provides sequence generators to assign unique values to primary key columns. They have very low overhead and are ideal for this purpose. But surprisingly, they do not actually guarantee that values are assigned in increasing order, which was required for us to make audit logs available in our public API. In this post, I'll explain under what circumstances generated IDs can appear in the database out of order and what solution we came up with to prevent this while keeping the impact on performance low.

At cloudscale, we like to exert as much control over our infrastructure as possible. In this blog post, I share some insights about our internal tooling and infrastructure, the risks of relying on U.S. vendors, and what all of that means regarding digital sovereignty.

Learn how to effectively use uv package manager within IntelliJ IDEA for Python projects, with practical setup steps and workarounds for current limitations.

Want to learn how our development team tackled the common struggle of prioritizing tech debt and what it has to do with beers and our next vacations?

At cloudscale, we take security very seriously. Our approach includes rigorous code and architecture reviews, extensive automated testing, and leveraging battle-tested open-source frameworks like Django. For a recent external penetration test in the context of our ISO certification, we engaged a specialized provider to examine our Control Panel and API for potential security vulnerabilities.

I used the launch of the dedicated GPUs as an opportunity to show off some of the new possibilities it adds to our platform. In this Engineering Blog Post I'm going through the setup of a self-hosted AI chatbot using Ollama and Open WebUI, powered by the DeepSeek-R1 70B model running on one of our brand-new NVIDIA L40S GPUs.

How we collaborated with Puzzle ITC to publish a Fleeting plugin for GitLab, enabling our customers to autoscale their GitLab CI workload.

How do we know how much we should charge for your Object Storage usage? A journey into rgw-metrics.

Set up a highly available Kubernetes control plane using cloudscale's load balancer. This guide walks you through provisioning with Terraform, configuring Kubernetes, and testing failover step by step.

Why we use chmod +x on our playbooks and why that is a good idea.

If a new fridge arrives, the urge to just plug it, fill it with beverages and enjoy a cold one is big. But the setup actually requires quite some steps to ensure that the device runs reliably for a long time with low maintenance, so does a proper onboarding process. In this blog post I will use this odd analogy to describe my onboarding process as a Software Engineer in the Dev Team at cloudscale.

How we restart OSDs in our Ceph clusters to avoid customer impact.