Operating bare-metal Kubernetes and production delivery pipelines

Context

Java, Node.js, PHP, Python and Go services needed production, staging and development environments; mobile Unity applications needed reliable iOS/Android build pipelines.

Implementation

Independently deployed a roughly 10-node bare-metal Kubernetes cluster with Kubespray, Calico, ingress-nginx and Ceph-backed persistent storage, and operated and upgraded it node by node with Ansible. Administered existing GKE clusters. Worked with developers on incorrect readiness/liveness probes, HPA behavior and making workloads viable in Kubernetes. Used Kafka operators and wrote custom Helm charts. Built and operated pipelines on GitLab CI/Runners, Jenkins and TeamCity, including Unity builds for mobile platforms.

Outcome

A maintained Kubernetes and delivery foundation for heterogeneous workloads, with documented maintenance and troubleshooting responsibilities.

Trade-offs and limitations

The GKE clusters already existed — the work there was administration, not design. Release frequency and availability were not measured.

Technologies

• Kubespray
• Calico
• Ceph
• GitLab CI
• ingress-nginx
• Helm
• Ansible
• Jenkins
• TeamCity
• GKE
• Unity