Insights
Notes from building and running our own platform: Kubernetes, Ceph and PostgreSQL on our own hardware in a data centre in Munich. Each article says what we did, when we did it and what the result does not cover, and gives you the checks to repeat it on your own systems. Written by Frybad, from running our own platform.
Where the data behind these articles lives, and our other dated measurements: Where your data lives.
Field report
Ceph OSD encryption, retrofitted in production
All OSDs of our Ceph cluster encrypted since 30 Sep 2026, one OSD at a time while in service. The procedure, plaintext leftovers, mistakes, checklist.
Testing database restores: our restore drill
Restore drill on 30 Aug 2026: two production databases back in service after 3 min 22 s. How we measure, what the number leaves out, and a checklist.
CloudNativePG failover drill: what we measure
Planned CloudNativePG failover on a production database, 4 Oct 2026: back in service after 19 s. How we measured it, and what went wrong.
Guide
Reducing p95 latency: HPA, resources and indexes
Why timeouts appear as a platform grows and in which order to look: gateway logs, database indexes, requests and limits, HPA. With a measuring guide.
Release environments: Crossplane & Argo Workflows
How release environments on Kubernetes are created with Crossplane, how Argo Workflows runs the end-to-end tests on them, and how it all disappears again.
Self-hosted GitLab and CI for hardware teams
GitLab, GitLab CI, OpenProject and paperless-ngx on your own VMs, connected via LDAP/SAML and SMTP: how a hardware team maps its process onto tools.
VMware to Proxmox migration: how to prepare
Inventory, dependencies, storage, network, licences, pilot and way back: what to settle before moving from VMware to Proxmox VE. With a checklist.
On-premises LLM inference with vLLM and LiteLLM
What a self-hosted inference stack needs: GPU servers, Kubernetes GPU nodes, vLLM as the engine, LiteLLM as the company proxy. With a planning checklist.
Questions about your own platform?
Tell us what you run and what worries you about it. We answer.