Kubernetes Test and Production Deployment

Purpose

This runbook defines the Phase 4.5 GitOps deployment model for the dedicated k3s server.

It separates the continuously updated test instance from the manually promoted production instance while keeping both slices on the current single-node server.

Dedicated Server

Field	Value
Public IP	`91.99.127.138`
Restricted SSH user	`deploy-test`
Kubernetes	k3s
Ingress class	`traefik`
Storage class	`local-path`
ACME email	`sens.ops@sens.at`
Argo CD hostname	`argo.iot-sens.schlossers.at`
Test docs hostname	`docs.dev.iot-sens.schlossers.at`
Production docs hostname	`docs.iot-sens.schlossers.at`
Test API hostname	`api.dev.iot-sens.schlossers.at`
Production API hostname	`api.iot-sens.schlossers.at`

The SSH private key must stay outside Git history. The repository ignores ssh-private, ssh-private.*, *.pem, and *.key.

Desired State Repositories

The product repository contains:

service source code,
Dockerfiles,
the Helm chart under infra/helm/sens-platform,
documentation and CI checks.

The private SENS-GmbH/sens-platform-infra repository contains:

cluster bootstrap resources,
Argo CD Applications and Projects,
Argo CD local-user RBAC,
environment-specific Helm values,
release metadata for immutable deployments.

No secret values may be committed to either repository.

Namespaces

Namespace	Purpose
`argocd`	Argo CD control plane
`cert-manager`	Certificate management
`sens-test`	SENS Platform test release
`sens-production`	SENS Platform production

Test and production use separate Kubernetes Secrets, PVCs, TLS Secrets, Argo CD Applications, and Helm release names.

Release Selection

Deployments use immutable product revisions and immutable SemVer image tags. Runtime deployments must not use latest, branch names, git-<short-sha>, full-SHA tags, or mutable channel tags such as test.

Deployable images are published by the publish-images GitHub Actions workflow. After ci succeeds on main, it automatically publishes all deployable images with a generated alpha SemVer tag and updates test GitOps desired state.

The automatic test tag is derived from the root package.json version:

<package.json version>-alpha.<publish workflow run number>.<run attempt>

For example, package version 0.5.0 can produce 0.5.0-alpha.123.1. The package version must be a stable SemVer version because the workflow appends the alpha pre-release suffix for test builds.

Manual publish-images runs require a release_version input and publish that same SemVer tag for all deployable images.

Environment release channels:

Environment	Required version shape	Example
Test	`alpha` pre-release	`0.5.0-alpha.123.1`
Future staging	`rc` pre-release	`0.5.0-rc.1`
Production	stable SemVer	`0.5.0`

The workflow accepts an optional leading v, which is stripped before writing the Docker image tag. SemVer build metadata with + is intentionally rejected because Docker image tags cannot use +. The Git SHA is still recorded as product_commit in the GitOps update and as the OCI image revision label.

Version Progression Check

CI runs pnpm version:check before build and image publication. The check compares the root package.json version with the previous version from Git.

Allowed progression:

unchanged version for repeated test builds on the same release line,
patch increase within the same major and minor version, for example 1.5.2 to 1.5.3,
next minor version only with patch reset to 0, for example 1.5.2 to 1.6.0,
next major version only with minor and patch reset to 0, for example 1.5.2 to 2.0.0.

Rejected progression:

downgrades, for example 0.4.0 to 0.3.1,
skipped minor versions, for example 1.5.2 to 1.7.0,
minor versions with non-zero patch, for example 1.5.2 to 1.6.1,
skipped major versions or major versions with non-zero minor/patch.

The root package.json version must be stable SemVer without pre-release or build metadata. Test, staging, and production channel suffixes are applied by the release workflow, not by changing the root package version.

This check only blocks new product repository builds. Argo CD rollbacks to an already published older image tag remain possible because they are GitOps desired-state changes in the infra repository and do not run this product version progression check.

Normal test flow:

Update the root package.json version when starting a new release line.
Merge to main.
Wait for ci to succeed.
publish-images automatically publishes the generated alpha tag and triggers deploy-test-main.yml in the infra repository.

Manual alpha publish and test deploy for exceptional cases:

gh workflow run publish-images.yml \
  --repo SENS-GmbH/sens-platform \
  -f ref=main \
  -f release_version=0.5.0-alpha.1 \
  -f update_test_environment=true

This publishes 0.5.0-alpha.1 for all deployable images, then triggers the infra repository test deployment with image_tag=0.5.0-alpha.1.

The workflow rejects test deployment requests where release_version is not an alpha pre-release.

Image tags must not be moved after publication. If another test build is needed, merge another change to main or publish the next alpha pre-release manually.

After image publication, the workflow triggers deploy-test-main.yml in SENS-GmbH/sens-platform-infra. That workflow updates only the test Application and test values. It must not update production files.

Production promotion is a manual GitOps change in the infra repository:

Run promote-production-from-test in GitHub Actions or run scripts/promote-production-from-test.sh from the infra repository.
Review and merge the generated promotion pull request.
Sync sens-platform-production manually in Argo CD as sebastian or break-glass admin.

Production must promote a stable SemVer tag that has already been validated through the earlier release channels. Pre-release tags must not be used for customer-facing production.

The infra repository branch protection must require CODEOWNERS review for production paths. Without that GitHub setting, the CODEOWNERS file is only documentation and does not enforce review. The GitHub CODEOWNERS entry uses the repository collaborator @SensSebastianSchlosser; Argo CD uses the local user name sebastian.

GitHub Actions Secret

The product repository requires one Actions secret:

Secret	Purpose
`SENS_PLATFORM_INFRA_WORKFLOW_TOKEN`	Triggers the infra repository test deployment workflow

The token must be restricted to the infra repository and must not have general contents-write access. Prefer a fine-grained token or GitHub App credential with permission to dispatch the target workflow only where possible.

GitHub does not allow a workflow to generate its own fine-grained personal access token. The token must be created by an authorized GitHub user or replaced by a GitHub App installation token. If the secret is absent or empty, image publishing still succeeds but the automatic test GitOps update is skipped.

Argo CD Git Webhook

The infra repository has a GitHub push webhook that notifies Argo CD when the GitOps desired state changes:

Field	Value
Repository	`SENS-GmbH/sens-platform-infra`
Event	`push`
Payload URL	`https://argo.iot-sens.schlossers.at/api/webhook`
Content type	`application/json`

The webhook removes the normal Git polling delay after the deploy-test-main.yml workflow updates the test desired state. Argo CD still owns sync decisions; the webhook only requests a repository refresh.

The current webhook is configured without a shared secret because the restricted deploy-test access boundary does not allow patching argocd-secret. Argo CD supports unauthenticated webhook refreshes, but a shared secret should be added by a cluster administrator to reduce public webhook abuse risk.

Admin hardening command:

WEBHOOK_SECRET="$(openssl rand -hex 32)"
kubectl -n argocd patch secret argocd-secret --type merge \
  -p "{\"stringData\":{\"webhook.github.secret\":\"${WEBHOOK_SECRET}\"}}"

After patching Argo CD, set the same secret value on the GitHub webhook in SENS-GmbH/sens-platform-infra.

Public Exposure

Public HTTPS ingress is limited to:

Argo CD at argo.iot-sens.schlossers.at,
test documentation at docs.dev.iot-sens.schlossers.at,
production documentation at docs.iot-sens.schlossers.at,
test platform-api operation endpoints at api.dev.iot-sens.schlossers.at,
production platform-api operation endpoints at api.iot-sens.schlossers.at.

The following components stay cluster-internal:

mqtt-ingestion-worker,
telemetry-worker,
embedded TimescaleDB,
embedded NATS JetStream,
embedded mock MQTT broker.

The mock MQTT broker is enabled only for test.

Argo CD Access

Local Argo CD users:

User	Purpose
`deploy`	Test and standard non-production operations
`sebastian`	Production owner and manual production deployer
`admin`	Break-glass administrator, not automation identity

Rotate the initial Argo CD admin password immediately after first login. The admin password must not be stored in the repository, SSH user environment, or automation logs.

Required Runtime Secrets

Create secrets directly in the target cluster or through a documented external secret process. Do not commit their values.

Required in sens-test:

Secret	Type	Purpose
`ghcr-pull`	`kubernetes.io/dockerconfigjson`	Pull private GHCR images
`sens-test-timescaledb-auth`	`Opaque`	Embedded TimescaleDB auth

Required in sens-production:

Secret	Type	Purpose
`ghcr-pull`	`kubernetes.io/dockerconfigjson`	Pull private GHCR images
`sens-production-timescaledb-auth`	`Opaque`	Embedded TimescaleDB auth

Required key in each TimescaleDB auth secret:

postgres-password

SSH and Kubernetes Access Boundary

The restricted Linux user deploy-test must not be able to deploy production by SSH.

Required controls:

deploy-test has no sudo access.
deploy-test cannot read /etc/rancher/k3s/k3s.yaml.
no production-capable kubeconfig is stored in the deploy-test home directory or workspace.
the deploy-test user’s $HOME/.kube/config uses the restricted sens-test/deploy-test ServiceAccount.
Kubernetes RBAC for that ServiceAccount grants at most sens-test permissions and no mutation rights in sens-production.
Argo CD RBAC grants non-production users no applications/sync permission for sens-production/sens-platform-production.
GitHub automation from the product repository can trigger the test deployment workflow but cannot directly write production desired state.

If an SSH user has root access, sudo, or a cluster-admin kubeconfig, there is no technical guarantee that production cannot be changed. Those credentials must stay outside the AI/deploy context.

DNS and ACME Checks

DNS records must resolve before HTTP-01 certificates can become ready:

dig +short argo.iot-sens.schlossers.at A
dig +short docs.dev.iot-sens.schlossers.at A
dig +short docs.iot-sens.schlossers.at A
dig +short api.dev.iot-sens.schlossers.at A
dig +short api.iot-sens.schlossers.at A

Expected result for each host:

91.99.127.138

Cluster DNS check:

kubectl -n sens-test run dns-check --attach --rm --restart=Never \
  --image=busybox:1.36 -- nslookup api.dev.iot-sens.schlossers.at

If public DNS resolves but cert-manager still reports an HTTP-01 self-check DNS failure, restart CoreDNS and wait for cert-manager to retry:

kubectl -n kube-system rollout restart deployment/coredns
kubectl -n kube-system rollout status deployment/coredns
kubectl get certificate -A

Smoke Checks

After Argo CD sync, check test:

kubectl -n sens-test get pods
kubectl -n sens-test rollout status deployment/sens-platform-test-platform-api
kubectl -n sens-test rollout status deployment/sens-platform-test-docs
curl -fsS https://api.dev.iot-sens.schlossers.at/healthz
curl -fsS https://api.dev.iot-sens.schlossers.at/readyz
curl -fsS https://api.dev.iot-sens.schlossers.at/version
curl -fsS https://api.dev.iot-sens.schlossers.at/test
curl -fsS https://api.dev.iot-sens.schlossers.at/metrics
curl -fsS https://docs.dev.iot-sens.schlossers.at/healthz
curl -fsS https://docs.dev.iot-sens.schlossers.at/readyz
curl -fsS https://docs.dev.iot-sens.schlossers.at/

Check production only after an intentional manual production sync:

kubectl -n sens-production get pods
kubectl -n sens-production rollout status deployment/sens-platform-production-platform-api
kubectl -n sens-production rollout status deployment/sens-platform-production-docs
curl -fsS https://api.iot-sens.schlossers.at/healthz
curl -fsS https://api.iot-sens.schlossers.at/readyz
curl -fsS https://api.iot-sens.schlossers.at/version
curl -fsS https://api.iot-sens.schlossers.at/test
curl -fsS https://api.iot-sens.schlossers.at/metrics
curl -fsS https://docs.iot-sens.schlossers.at/healthz
curl -fsS https://docs.iot-sens.schlossers.at/readyz
curl -fsS https://docs.iot-sens.schlossers.at/

View the platform-api logs through Argo CD’s pod log view or with:

kubectl -n sens-test logs deployment/sens-platform-test-platform-api
kubectl -n sens-production logs deployment/sens-platform-production-platform-api

Calling GET /test writes a structured log entry with the message test endpoint called.

Internal checks:

kubectl -n sens-test exec statefulset/sens-platform-test-timescaledb -- \
  pg_isready -U sens_platform -d sens_platform

kubectl -n sens-test exec statefulset/sens-platform-test-nats -- \
  wget -qO- http://127.0.0.1:8222/healthz

Validation

Local and CI validation commands:

pnpm helm:lint
pnpm helm:template
pnpm helm:template:production

The chart must render without hard-coded cloud provider, DNS provider, certificate issuer, or storage class assumptions. Environment-specific values belong in the GitOps repository.

Limitations

This controlled production slice does not provide:

production high availability,
customer-data backup and restore readiness,
Netmore ingestion,
authentication,
tenant data handling,
telemetry persistence behavior.

Before real customer data is stored, backup, restore, monitoring, retention, and database operating procedures must be completed and tested.