Multi-environment Kubernetes platform for a software vendor

Context

A vendor hosted twelve business applications for eighty customers on virtual machines configured by hand over several years. Every new customer environment took two to three weeks of manual preparation, with configuration gaps that were hard to identify.

Problem

Version upgrades required weekend interventions, preparing a new customer environment occupied one person full time, and no restore procedure had been tested since the platform went live.

Objectives

• Cut customer environment preparation time to a few hours
• Standardise deployments with no manual intervention
• Make the platform restorable and documented

Role

Platform architect and owner of the foundation implementation, training the two internal operations engineers across the whole chain.

Solution

The work started with a three-week inventory: resources actually used, dependencies between applications, ports, volumes and scheduled tasks. It showed that half the virtual machines ran below fifteen percent capacity, with wide variation between customers. The foundation was built in Terraform, with a network module, a cluster module and one module per application type. Applications were containerised with multi-stage images and packaged as Helm charts, then deployed through Argo CD. Customer-specific configuration is isolated in versioned values files, which makes differences visible in pull requests. The cutover happened customer by customer over six months, starting with staging environments. Restore procedures were tested on the main database and on file volumes before the first production customer was migrated.

Architecture

One Kubernetes cluster per major geographic region, with a namespace per customer and network policies restricting communication to declared dependencies. Infrastructure is described in Terraform with locked state and plans reviewed in pull requests. Argo CD applies manifests from a Git repository; any configuration drift is detected and reported. Images are built by GitHub Actions, scanned and published to an internal registry. Monitoring combines Prometheus for technical metrics, Grafana for per-application dashboards and OpenTelemetry for application traces.

Results (samples)

• Customer environments prepared from a code-defined template
• Application deployments with no manual server intervention
• Restore procedures tested and documented
• Resource consumption aligned with actual usage

Indicators (samples)

• New customer environment preparation

  2 weeks to 6 hours

  Indicative order of magnitude — demonstration sample

• Applications described as Helm charts

  12

  Illustrative volume on a simulated scope — demonstration sample

• Underused CPU resources identified

  around 45 percent

  Result of a simulated inventory — demonstration sample

Technologies

• PostgreSQL
• Docker
• Kubernetes
• Helm
• Terraform
• GitHub Actions
• Argo CD
• Prometheus
• Grafana
• OpenTelemetry

Related services

• Cloud architecture and infrastructure as code

  I design reproducible cloud platforms, fully described as code and sized to actual need. You gain control over costs as well as the ability to recover from incidents.

• DevOps and CI/CD delivery pipelines

  I build delivery pipelines that genuinely test the code and allow rollback within minutes. Deployment stops being an event and becomes a routine operation.

• Docker containerisation and Kubernetes orchestration

  I containerise your applications with lean images and run them on Kubernetes with controlled manifests. The platform becomes predictable, properly sized and repairable.

• Observability and application performance

  I make your applications diagnosable: correlated traces, useful metrics and alerts that signal a real problem. You move from interpreting symptoms to identifying the cause.