06 · A production platform in one command¶
Outcome: the basek8s group on your cluster: GitOps (ArgoCD), metrics and dashboards (kube-prometheus-stack), logs (Loki 3 + Alloy), OpenTelemetry, Gateway API with Envoy Gateway behind a private load balancer, cert-manager with a cluster CA, external-secrets and sealed-secrets. Every web UI reachable over TLS, a CI pipeline template, and an uninstall you can undo.
Verified live on VMware Fusion 13.6
tests/scenarios/06-platform-in-one-command.sh installs basek8s on the scenario 05 cluster with CLOUDSEED_LIVE=1. Without it, it renders the cluster, runs the catalog commands that work offline (list, info, plan, template) and checks the rest say the cluster is not created yet.
| Time | Cost | Level | Needs |
|---|---|---|---|
| ~30 min (the install ~15 min) | $0 on the local cluster | Intermediate | The cluster from 05 (or EKS / GKE / AKS) |
What you'll build¶
%%{init: {"flowchart": {"nodeSpacing": 30, "rankSpacing": 38, "padding": 12}}}%%
flowchart TB
you["Your browser<br/>https://argocd.vmware-lab.local ..."] --> gw
subgraph cluster["vmware-lab cluster"]
gw["Shared Gateway<br/>cloudseed/cloudseed<br/>Envoy Gateway +<br/>MetalLB IP<br/>wildcard TLS from<br/>the cluster CA"]
subgraph gitops["GitOps"]
argo["ArgoCD"]
end
subgraph obs["Observability"]
prom["Prometheus +<br/>Alertmanager"]
graf["Grafana"]
loki["Loki 3"]
alloy["Alloy<br/>(pod logs)"]
otel["OpenTelemetry<br/>operator"]
end
subgraph sec["Certificates and secrets"]
cm["cert-manager +<br/>cluster CA issuer"]
es["external-secrets"]
ss["sealed-secrets"]
end
gw --> argo & graf
alloy --> loki
graf --> prom & loki
endOn a cloud cluster the same group gets an internal load balancer instead of MetalLB, and items that need cloud identities (external-secrets, external-dns, the AWS load balancer controller) get them from the environment's Terraform stack automatically.
Before you start¶
- A cluster managed by cloudseed and selected as current: 05 · Kubernetes on your laptop ends with
cs env use vmware-lab. kubectlandhelm(cs install kubernetes).
Step 1: Browse the catalog¶
cs platform list
cs platform list --charts
cs platform info basek8s
cs platform info kube-prometheus-stack
cs explain platform
list shows every group and item with its install state; --charts adds where each one comes from (Helm repo, OCI registry, manifest URL) and the version pinned for this cloudseed release. Nothing is vendored.
Step 2: Plan before touching the cluster¶
Expected output (abbreviated)
╭─ Plan for vmware-lab (vmware/rke2) ────────────────────────────────────────────╮
│ ✔ metrics-server built into the distro - skip Resource metrics API │
│ ↳ already provided by rke2 out of the box │
│ + gateway-api install Kubernetes Gateway API CRDs (experimental… │
│ + cert-manager install X.509 certificates for in-cluster services… │
│ + cert-manager-issuer install Cluster CA issuer (cert-manager) for TLS on… │
│ + metallb install LoadBalancer IPs on the private network… │
│ + envoy-gateway install Envoy Gateway: the cluster's Gateway API… │
│ + argocd install GitOps continuous delivery │
│ + kube-prometheus-stack install Prometheus, Alertmanager, Grafana, node/kube…│
│ + alloy install Grafana Alloy: collects every pod's logs… │
│ + local-path-provisioner install Default StorageClass for local clusters… │
│ + loki install Log aggregation (Loki 3, single binary… │
│ + opentelemetry-operator install OpenTelemetry operator (collectors… │
│ + external-secrets install Sync secrets from AWS/GCP/Azure secret… │
│ + sealed-secrets install Encrypt secrets so they can live in git │
╰─────────────────────────────────────────────────────────────────────────────────╯
Dependencies come first (cert-manager before the issuer, MetalLB before the Gateway), components the distro already ships are skipped (RKE2 has metrics-server), and nothing is ever installed twice.
Step 3: Install it¶
Expected output (last lines)
✔ sealed-secrets installed
╭─ Done ────────────────────────────────────────────────────────────────╮
│ installed gateway-api, cert-manager, cert-manager-issuer, metallb, │
│ envoy-gateway, argocd, kube-prometheus-stack, alloy, ... │
│ status cs platform status │
│ pods cs kubectl get pods -A │
│ web UIs cs platform ui │
│ k9s cs k9s │
╰───────────────────────────────────────────────────────────────────────╯
Values are adapted to the target and distro, chart versions are pinned, and the install is idempotent: run it again and installed items are skipped (--upgrade re-applies them at the pinned version).
Step 4: Open the web UIs¶
platform ui attaches an HTTPRoute to the shared Gateway for every installed UI and prints the URLs and where the credentials are, then how to reach them:
Expected output (abbreviated; the IP comes from MetalLB)
╭─ UIs on vmware-lab · ingress 172.16.56.100 · TLS by cert-manager (cloudseed-ca) ─╮
│ argocd https://argocd.vmware-lab.local admin / kubectl -n argocd ... │
│ kube-prometheus-stack https://kube-prometheus-stack.vmware-lab.local │
│ admin / grafana_password (platform/secrets.json) │
╰────────────────────────────────────────────────────────────────────────────────────────╯
╭─ Reach them ───────────────────────────────────────────────────────────────────────────╮
│ route the ingress IP is on the host-only network: reachable from this machine │
│ dns add to /etc/hosts: 172.16.56.100 argocd.vmware-lab.local ... │
│ trust cs kubectl -n cert-manager get secret cloudseed-root-ca ... > cloudseed-ca.crt │
╰────────────────────────────────────────────────────────────────────────────────────────╯
Import the cluster CA once and the certificates are trusted (they are valid for ten years):
cs kubectl -n cert-manager get secret cloudseed-root-ca -o jsonpath='{.data.ca\.crt}' | base64 -d > cloudseed-ca.crt
Step 5: Add a CI pipeline to your app repository¶
Writes .gitlab-ci.yml into the current directory: build, test, a Trivy image scan, release, and an ArgoCD deploy stage. cs undo removes it again (a copy is kept if you edited it since).
Step 6: Remove an item, then undo it¶
uninstall never removes shared dependencies (cert-manager, the Gateway API, MetalLB) and refuses items another installed item still needs. cs undo re-installs what the uninstall removed. Unattended: add --auto-approve to both.
Use an agent, MCP or the UI¶
Follow the same numbered steps and verification/cleanup conditions through your chosen interface. Start with the interface setup and coverage guide; replace account/project/subscription and SSH placeholders before any live request.
Agent prompt: “On vmware-lab, follow the platform walkthrough. Inspect and plan basek8s first, explain dependencies and private UI access, then install only approved items and verify their releases. Show undo history before a rollback.”
MCP starter: cloudseed_platform with:
Use the matching tool for each remaining step in this page; the command-to-tool map lists the tool family. Keep vmware-lab selected. Preview first; add confirm:true only to the specific change you have authorized. Host bootstrap, provider login and interactive applications retain their documented human steps.
UI: Select vmware-lab → Platform. Choose basek8s, inspect items and Plan before Install. Follow catalog UI links after the install; use All actions → Helm for the verification commands and Undo for the recorded change.
Verify it worked¶
- Every pod in
argocd,monitoring(Prometheus, Grafana, Loki, Alloy),cert-manager,envoy-gateway-system,metallb-system,opentelemetry-operator-systemandexternal-secretsis Running (sealed-secrets runs inkube-system). - The Gateway
cloudseed/cloudseedisProgrammedwith an address from the MetalLB pool. - Grafana answers at
https://kube-prometheus-stack.vmware-lab.localand shows the cluster dashboards; logs from every pod are searchable through the Loki data source.
Clean up¶
This removes every item of the group, cert-manager, the Gateway API, MetalLB and local-path included, unless another installed item still needs one: then it stops and names it (for example local-path-provisioner is still needed by minio). Namespaces left empty are deleted; namespaces that still hold objects (for example the cluster CA secret in cert-manager) are kept. Unattended: add --auto-approve. To remove everything, destroy the cluster (cs destroy vmware --env lab), or keep it for scenarios 07 to 10.
What just happened¶
- The catalog lives in
cloudseed/platform.py: each item records its upstream source, a pinned version, its namespace, its dependencies and per-target values.cs explain platform <item>shows all of it. - Generated passwords (Grafana, MinIO, ...) are in
~/.cloudseed/envs/vmware-lab/platform/secrets.json(0600). - Every install and uninstall is an undo point for
cs undo. - Learn more: Platform guide · Platform catalog · Undo and audit · Explain index
Next steps¶
- 07 · Data and AI stack: MinIO, Postgres, Kafka, Ollama + Open WebUI.
- 08 · Backups you can trust: Velero with a restore drill.
- 15 · FinOps: OpenCost on top of this Prometheus.