A FLAME Backend for Kubernetes. Manages pods as runners in the cluster the app is running in.

The current implementation is very basic and more like a proof of concept. More configuration options (resources, etc.) will follow.

def deps do [ {:flame_k8s_backend, "~> 0.5.0"} ] end

Configure the flame backend in our configuration or application setup:

 # application.ex children = [ {FLAME.Pool, name: MyApp.SamplePool, backend: FLAMEK8sBackend, min: 0, max: 10, max_concurrency: 5, idle_shutdown_after: 30_000, log: :debug} ]

In order for the runners to be able to join the cluster, you need to configure a few environment variables on your pod/deployment:

The POD_NAME and POD_NAMESPACE are used by the backend to get informations from your pod and use them for the runner pods (e.g. env variables).

apiVersion: apps/v1 kind: Deployment spec: selector: matchLabels: app: myapp template: spec: containers: - env: - name: POD_NAME valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.name - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace

Your application needs run as a service account with permissions to manage pods:

--- apiVersion: v1 kind: ServiceAccount metadata: name: myapp namespace: app-namespace --- apiVersion: rbac.authorization.k8s.io/v1 kind: Role metadata: namespace: app-namespace name: pod-mgr rules: - apiGroups: [""] resources: ["pods"] verbs: ["create", "get", "list", "delete", "patch"] --- apiVersion: rbac.authorization.k8s.io/v1 kind: RoleBinding metadata: name: myapp-pod-mgr namespace: app-namespace subjects: - kind: ServiceAccount name: myapp namespace: app-namespace roleRef: kind: Role name: pod-mgr apiGroup: rbac.authorization.k8s.io --- apiVersion: apps/v1 kind: Deployment spec: template: spec: serviceAccountName: my-app

Your application needs to be able to form a cluster with your runners. Define POD_IP, RELEASE_DISTRIBUTION and RELEASE_NODE environment variables on your pods as follows:

apiVersion: apps/v1 kind: Deployment spec: template: spec: containers: - env: - name: POD_IP valueFrom: fieldRef: apiVersion: v1 fieldPath: status.podIP - name: RELEASE_DISTRIBUTION value: name - name: RELEASE_NODE value: my_app@$(POD_IP)

The FLAME Kubernetes backend first queries the Kubernetes API server to extract information from the running Pod like the container image, resource requests and limits, environment variables etc. This information is then used to build the manifest for the runner pod. The backend then sends the resulting manifest to the API server in order to spin up a runner pod.

If your parent is part of a Deployment, make sure your FLAME runner Pod doesn't contain the labels you used as selectors (i.e. .spec.selector on your Deployment). Otherwise the replica controller sees your FLAME runner as an additional Pod to the Deployment's ReplicaSet and "downscales" the deployment to the desired replica count (i.e. snipes your runner Pod).