Home Set up Kubernetes

Set it up on Kubernetes

A published chart, an external Postgres, and whichever way you already expose things. Migrations run themselves on every rollout.

Install

helm repo add morgenruf https://charts.morgenruf.dev
helm repo update

helm upgrade --install morgenruf morgenruf/morgenruf \
  --namespace morgenruf --create-namespace \
  --set slack.clientId="YOUR_CLIENT_ID" \
  --set slack.clientSecret="YOUR_CLIENT_SECRET" \
  --set slack.signingSecret="YOUR_SIGNING_SECRET" \
  --set externalDatabase.url="postgresql://user:pass@host:5432/morgenruf" \
  --set flaskSecretKey="$(openssl rand -hex 32)" \
  --set app.url="https://standups.your-domain.com"

Use a real database. The chart can start one for you, which is fine for a look, but point externalDatabase.url at managed Postgres for anything you intend to keep. The data outliving the release is the point.

A release runs the backend (with a migrate init container), a small frontend that serves the dashboard and proxies everything else to the backend, and a Redis that holds standup sessions across restarts. Set redis.enabled=false and externalRedis.url to use a managed Redis instead.

Exposing it

Slack has to reach the app over HTTPS. Three ways, in order of how common they are:

  • Ingress. On by default, with the nginx class and a cert-manager annotation. Set ingress.hosts and ingress.tls to your own domain, since the defaults name api.morgenruf.dev.
  • Gateway API. An HTTPRoute is supported for clusters that have moved on from Ingress.
  • Cloudflare tunnel. No ingress controller, no public load balancer, no open port. Useful on a homelab cluster or anywhere the cluster is not internet-facing.

One replica, on purpose

The scheduler that fires standups, coffee chat rounds and the midnight kudos reset runs inside the app process. Each pod runs its own, and every scheduled firing is claimed in the database before it runs, so two pods overlapping during a rollout do not both send the morning message. One replica is still the default and enough for most teams; let Kubernetes restart it, and a restart mid-round resumes rather than repeating, because delivery is recorded per person as it happens.

Upgrades

helm repo update
helm upgrade morgenruf morgenruf/morgenruf --reuse-values

Migrations run in an init container before the new app container starts. If you pin an image tag rather than using the chart's, image.tag (or image.digest) covers both the app and the migrate container; set frontend.image.tag to the same release.

What to watch

  • kubectl logs deploy/morgenruf -c migrate after a rollout, to see which migrations applied.
  • Scheduler started with N jobs in the app log. N should grow when you add a standup.
  • The status page endpoint, if you run your own monitoring.
Questions

Running it on Kubernetes

Does the chart ship a database?

It can, but do not use it for anything you care about. Point externalDatabase.url at a managed Postgres so the data outlives the release.

How do migrations run?

An init container runs them before the app container starts, so a rollout is also a migration. When you pin an image, image.tag or image.digest covers both the app and the init container; pin frontend.image.tag to the same release.

Can it run without an ingress controller?

Yes. Set ingress.enabled=false and run a Cloudflare tunnel in the same namespace, pointed at the morgenruf Service. The chart does not deploy the tunnel for you. Nothing needs a public load balancer.

How many replicas?

One is the default and plenty. Every scheduled firing is claimed in the database first, so pods that overlap during a rollout do not send the morning standup twice. The chart runs Redis by default, which more than one replica needs for standup sessions.

Free for every seat, and that is the whole pricing page

Add to Slack puts Morgenruf on the free hosted instance CloudDrove runs, in about two minutes. Or self-host the same MIT code with Docker Compose or Helm and pay only for the server.