Install the platform

ankka-flow on Kubernetes is three things:

Piece	What it is
The AnkkaFlow custom resource definition	group flow.ankka.thinkmorestupidless.com, version v1alpha1, kind AnkkaFlow, short name aflow; one resource per pipeline
The operator	one Deployment, ankka-flow-operator in namespace ankka-flow, watching AnkkaFlow resources in every namespace
Kafka cluster Secrets	kafka-cluster-<name> Secrets in the operator's namespace, saying how to reach each Kafka cluster; default is used by any topic that names no cluster and no brokers

Kafka itself is not part of the platform. A pipeline uses whatever Kafka its cluster Secrets point at.

The manifests

The manifests are Kustomize components in kustomization/:

Path	Holds
components/crd	the custom resource definition
components/operator	the ankka-flow namespace, the operator's ServiceAccount, ClusterRole and ClusterRoleBinding, and its Deployment
components/kafka	a single-node development Kafka in namespace kafka, and a kafka-cluster-default Secret pointing at it; one broker and no persistence beyond the pod, so never for production
overlays/local	all three components, for a kind cluster

Render an overlay to see exactly what it applies:

kubectl kustomize kustomization/overlays/local

On kind

just up

This creates a kind cluster named ankka if there is none, switches to its context, and runs kustomization/deploy-local.sh. The script builds the sidecar, operator and sample images, loads them into the cluster, applies the custom resource definition and waits for it, applies the local overlay, restarts the operator so it runs the image just loaded, and waits for the operator and Kafka. It refuses to touch any kubectl context but kind-<cluster>. ANKKA_FLOW_KIND_CLUSTER changes the cluster name, and just down deletes the cluster.

Run just deploy to rebuild and reinstall on a cluster that already exists.

On any other cluster

Install the custom resource definition first, and wait until it is established:

kubectl apply --server-side -f kustomization/components/crd/ankkaflow.yaml
kubectl wait --for=condition=Established crd/ankkaflows.flow.ankka.thinkmorestupidless.com --timeout=60s

Then install the operator with two changes to components/operator/operator.yaml, made in an overlay of your own:

• the operator container's image, a reference your cluster can pull, such as ghcr.io/thinkmorestupidless/ankka-flow-operator:<version>;
• the FLOW_SIDECAR_IMAGE environment variable, the sidecar image every pipeline's pods run, such as ghcr.io/thinkmorestupidless/ankka-flow-sidecar:<version>.

A release publishes both images to ghcr.io/thinkmorestupidless with the release's version as the tag.

Finally, create a kafka-cluster-default Secret in namespace ankka-flow for your Kafka, and one kafka-cluster-<name> Secret for every other cluster a pipeline names. Configure at deploy time describes the Secret's keys. Keep real credentials out of version control.

The sidecar image is the operator's

No pipeline names the sidecar image. The operator adds the sidecar to every streamlet pod from its own FLOW_SIDECAR_IMAGE setting, so changing that setting upgrades every pipeline's sidecar the next time each streamlet rolls. An operator with no sidecar image refuses every pipeline: the resource's phase is Failed and each problem is a SidecarImageMissing event.

Operator lists the operator's other settings: where it looks for Kafka cluster Secrets, its resync interval, its retry backoff and how many pipelines it reconciles at once.

Check the installation

kubectl -n ankka-flow rollout status deployment/ankka-flow-operator
kubectl -n ankka-flow logs deployment/ankka-flow-operator

The operator logs the sidecar image it will use when it starts.