Small service, real platform behavior
FastAPI keeps application logic minimal so the evidence stays focused on packaging, configuration, security, rollout, and runtime metadata.
Kubernetes and cloud delivery
KubePulse
A small service that makes production-minded Kubernetes behavior visible.
- Role
- Solo developer across API, container, manifests, security context, tests, and local cluster verification
- Evidence status
- Verified on Minikube with Docker Desktop
- Verified
- 2026-07-22
01
Problem and scope
KubePulse packages a FastAPI service as a non-root container and deploys it to Kubernetes with two replicas, rolling updates, health probes, runtime metadata, resource controls, and optional autoscaling and ingress.
02
Architecture
Local client
Calls the stable ClusterIP service
Kubernetes service
Distributes new connections across replicas
Two FastAPI pods
Probes, limits, non-root security, downward API
Control plane
Maintains readiness and rolling deployment state
Text equivalent: the stages above flow from left to right, with each stage passing its result to the next.
03
Engineering decisions
Secure container baseline
The workload runs as a non-root user with resource requests and limits instead of relying on permissive defaults.
Optional platform features
Ingress and HPA stay outside the base overlay so the first deployment remains understandable and repeatable.
04
Verification evidence
- Minikube v1.38.1 created a Kubernetes v1.35.1 cluster with the Docker driver.
- Both replicas reached Ready with zero restarts, and both health endpoints returned success.
- Repeated service calls returned metadata from both pod identities.
- A rolling restart replaced both pods while the deployment completed successfully.
- Metrics Server reported pod CPU and memory. The HPA was configured for two to five replicas, but no load-driven scale event was claimed.
$ kubectl -n kubepulse get deployment,pods
deployment/kubepulse · 2/2 ready
health/live: alive · health/ready: ready
service responses: pod ...j5qvk · pod ...lmklh
✓ rolling restart completed05
Limitations
- The verification used a local single-node Minikube cluster, not a managed multi-node production cluster.
- The HPA and Metrics Server were configured and observed, but load generation was not run long enough to prove an autoscale event.
- Ingress remains optional and was not part of the verified port-forward-free service test.
Next step