X4BNet platform actions

k8s-deploy

X4BNet/platform-actions/k8s-deploy@v1 is compatible with the inputs accepted by Azure/k8s-deploy@v1. For deploy operations it first creates a temporary DaemonSet that mounts every deployment image through Kubernetes image volumes, waits for those images to be available on all selected workers, and then invokes the Azure action with the same image references.

- uses: X4BNet/platform-actions/k8s-deploy@v1
  with:
    namespace: development
    manifests: combined.yaml
    images: x4bnet/x4b-site:git-${{ github.sha }}
    imagepullsecrets: regcred
    kubectl-version: latest
    prepull-node-selector: '{"k8s.x4b.net/ready":"1"}'

Each non-empty line in images must contain a repository/image path and an explicit tag or sha256 digest. prepull-node-selector must be a non-empty JSON object so an accidental invocation cannot pull large images onto every cluster node. Optional prepull-tolerations accepts a JSON array of Kubernetes toleration objects.

Before applying manifests, the action detects and gracefully deletes existing pods whose init, regular, or ephemeral containers are in CrashLoopBackOff. If it deletes any, it starts the deployment and performs a second check five seconds later in parallel with the deployment, deleting any pods still crash looping. By default this checks the whole target namespace; set crashloop-pod-selector to a Kubernetes label selector to narrow the pods in scope. Set delete-crashloopbackoff-pods: "false" to disable this behavior.

The default pre-pull rollout timeout is five minutes. A restricted in-cluster watchdog deletes the DaemonSet after ten minutes if the runner disappears, while an always() cleanup step removes the DaemonSet and its namespaced RBAC resources after success or failure. The watchdog ServiceAccount can delete only its own uniquely named DaemonSet.

The target cluster must support the Kubernetes ImageVolume feature and the workflow identity must be able to create, patch, read, and delete DaemonSets, ServiceAccounts, Roles, and RoleBindings, and to list and delete Pods in the target namespace. Cluster context must be configured before invoking the action.