Georg Schwarz's Kubernetes production strategy: Lessons from self-hosting a European Google Docs alternative

This review examines Georg Schwarz's detailed account of moving a Kubernetes setup from demo to production, focusing on the specific tools and architectural decisions made for a self-hosted application.

TL;DR

Best for: Teams self-hosting stateful applications on bare-metal or European cloud providers, seeking a comprehensive, opinionated Kubernetes stack. Skip if: You rely exclusively on managed Kubernetes services like GKE/EKS/AKS, or prefer a minimalist, unopinionated approach to cluster tooling. Bottom line: Georg Schwarz provides a robust, battle-tested Kubernetes deployment blueprint, emphasizing operators for stateful workloads and a strong GitOps foundation.

METHODOLOGY

This v0 review draws on the founder's published claims at https://georg-schwarz.com/blog/from-kubernetes-demo-to-production-platform/, accessed on 2026-05-19. The review covers the specific Kubernetes tools, configurations, and architectural decisions detailed by Georg Schwarz, based in Berlin, Germany, for transitioning a self-hosted application from a demo environment to a production platform. It focuses on his lessons learned regarding stateful workloads, storage, monitoring, logging, CI/CD, and security within a Kubernetes context. What's not covered includes independent performance benchmarks, long-term workflow impacts beyond the scope of the article, or edge-case behaviors not explicitly addressed. Update cadence: re-tested when claims diverge from observed behavior.

WHAT IT DOES

Georg Schwarz's blog post outlines a production-grade Kubernetes deployment strategy for self-hosting a stateful application. The core of this strategy involves a curated set of open-source tools and specific configurations to address common challenges in Kubernetes operations.

Managing Stateful Workloads with Operators

The strategy emphasizes using Kubernetes Operators for stateful applications, specifically highlighting the Zalando Postgres Operator for PostgreSQL databases. This approach abstracts away complex database management tasks like provisioning, scaling, and backups, treating databases as first-class Kubernetes citizens. Initially, the bitnami/postgresql Helm chart was used, but the move to an operator was driven by the need for more robust, production-ready state management.

Robust Storage Solutions

For persistent storage, the strategy evolved from NFS for shared file storage to Longhorn for block storage. Longhorn, a Cloud Native Computing Foundation (CNCF) sandbox project, provides distributed block storage for Kubernetes, offering features like snapshots, backups, and disaster recovery. This shift addresses the critical need for reliable and scalable storage for stateful applications within the cluster.

Comprehensive Observability

The deployment incorporates a full observability stack using kube-prometheus-stack for monitoring and Loki for logging. kube-prometheus-stack bundles Prometheus, Grafana, and Alertmanager, providing metrics collection, visualization, and alerting capabilities. Loki, combined with Promtail, offers a cost-effective, log aggregation system optimized for Kubernetes, allowing for efficient querying and analysis of application and cluster logs.

GitOps-driven CI/CD

Argo CD forms the backbone of the continuous integration and continuous deployment (CI/CD) pipeline, implementing a GitOps workflow. Application deployments are managed declaratively through Git repositories, with Argo CD continuously synchronizing the cluster state with the desired state defined in Git. GitHub Actions are used for building container images, which are then pushed to a container registry, ready for Argo CD to deploy. This ensures consistent, auditable, and automated deployments.

WHAT'S INTERESTING / WHAT'S NOT

What's interesting about Georg Schwarz's account is its pragmatic, opinionated selection of tools for a self-hosted, bare-metal-adjacent Kubernetes environment. Unlike many high-level "Kubernetes best practices" articles, Schwarz provides concrete tool names and specific reasons for their adoption or rejection, for example, moving from Bitnami's PostgreSQL chart to Zalando's operator for better production readiness. The explicit mention of Hetzner as the cloud provider and the focus on European data sovereignty (implied by "European Google Docs alternative") adds valuable context for teams with similar regulatory or infrastructure constraints. The detailed breakdown of challenges—from stateful workloads to secrets management with Sealed Secrets and backups with Velero—offers a realistic roadmap for others. His emphasis on MetalLB for bare-metal load balancing and ExternalDNS for DNS management fills crucial gaps often overlooked in cloud-agnostic guides.

What's not interesting, or rather, what needs further exploration, is the performance and cost implications of this specific stack. While Schwarz mentions cost optimization with spot instances, there are no specific metrics on the operational overhead or resource consumption of running Longhorn, Zalando Postgres Operator, or the full observability stack. The article also doesn't delve into the team size or skill set required to maintain such a complex setup. A single founder, Georg Schwarz, working with one full-time engineer and two contributors, as implied by the source, managing this stack is impressive but might not be replicable for all teams. The article is a retrospective, so it lacks real-time debugging or incident response scenarios, which are critical for a "production platform."

PRICING

The blog post details a self-hosted Kubernetes setup, so there are no direct SaaS pricing tiers for the overall "strategy." The costs are primarily for underlying infrastructure and specific open-source tools, which are generally free to use but incur operational overhead.

• Infrastructure: The author mentions using Hetzner, a European cloud provider. Specific costs for VMs, storage, and networking would depend on the chosen instance types and usage.
• Software: All mentioned tools (Kubernetes, Zalando Postgres Operator, Longhorn, kube-prometheus-stack, Loki, Argo CD, Sealed Secrets, Velero, ExternalDNS, MetalLB, cert-manager, ingress-nginx) are open-source and free to use. Pricing snapshot date: 2026-05-19

VERDICT

Georg Schwarz's Kubernetes production strategy is an excellent blueprint for teams building and operating stateful applications on self-managed or bare-metal Kubernetes clusters, particularly within a European context. It is best for those who need a comprehensive, opinionated stack that prioritizes robust state management, reliable storage, and GitOps-driven deployments. The detailed tool choices, like the Zalando Postgres Operator and Longhorn, provide concrete solutions to common pain points. Teams relying heavily on managed Kubernetes services, which abstract away much of this complexity, or those seeking a more minimalist approach, might find this strategy overly prescriptive. Ultimately, this approach offers a battle-tested path to production readiness, grounded in practical lessons learned from real-world self-hosting.

WHAT WE'D TEST NEXT

In a v2 review, we would focus on benchmarking the performance and resource footprint of the recommended stack, particularly Longhorn and the Zalando Postgres Operator, under various load conditions. We would also investigate the operational overhead and maintenance requirements for a small team, quantifying the time spent on upgrades, troubleshooting, and security patching. Specific tests would include disaster recovery scenarios using Velero, measuring recovery time objectives (RTO) and recovery point objectives (RPO). We would also explore the cost-effectiveness of this stack on different bare-metal or European cloud providers beyond Hetzner, comparing resource utilization and infrastructure spend. Finally, we'd examine the security posture of the entire stack, including a deeper dive into network policies and Pod Security Standards implementation, to validate the claims of a secure production environment.