Merge pull request #618 from vdice/fix/k3d-restart

fix(containerd): update restart for k3d scenarios

Author: vdice

.github/workflows/helm-chart-smoketest.yml

@@ -172,8 +172,6 @@ jobs:
         run: kubectl label node --all spin=true
 
       - name: verify only one installer pod with Succeeded status
-        # TODO: provisioning on k3d still leads to the first installer pod finishing with provisioner status Unknown and phase Failed
-        if: matrix.config.type != 'k3d'
         run: |
           timeout 60s bash -c 'until [[ "$(kubectl -n rcm get $(kubectl get pods -n rcm --no-headers -o name | grep install | head -n1) -o jsonpath="{.status.phase}" 2>/dev/null)" == "Succeeded" ]]; do sleep 2; done'
 
@@ -201,8 +199,7 @@ jobs:
           kubectl describe runtimeclass wasmtime-spin-v2
 
           # Get install pod logs
-          # Note: there may be multiple pods pending k3d fix for issue https://github.com/spinframework/runtime-class-manager/issues/393
-          install_pod=$(kubectl get pods -n rcm --no-headers -o name | awk '{if ($1 ~ "-spin-v2-install") print $0}' | tail -n 1)
+          install_pod=$(kubectl get pods -n rcm --no-headers -o name | awk '{if ($1 ~ "-spin-v2-install") print $0}')
           kubectl describe -n rcm $install_pod || true
           kubectl logs -n rcm -c downloader $install_pod || true
           kubectl logs -n rcm -c provisioner $install_pod || true

internal/containerd/restart_unix.go

@@ -96,21 +96,14 @@ func (c K3sRestarter) Restart() error {
 			return fmt.Errorf("unable to restart k3s: %w", err)
 		}
 	} else {
-		// TODO: this approach still leads to the behavior mentioned in https://github.com/spinframework/runtime-class-manager/issues/140:
-		// The first pod's provisioner container exits with code 255, leading to pod status Unknown,
-		// followed by the subsequent pod's provisioner container no-op-ing and finishing with status Completed.
-		pid, err := getPid("k3s")
-		if err != nil {
-			return err
-		}
-		slog.Debug("found k3s process", "pid", pid)
-
-		err = syscall.Kill(pid, syscall.SIGHUP)
+		// Assuming k3d; PID 1 is the main k3d entrypoint which should be restarted
+		// rather than the nested k3s process
+		out, err := nsenterCmd("kill", "1").CombinedOutput()
+		slog.Debug(string(out))
 		if err != nil {
-			return fmt.Errorf("failed to send SIGHUP to k3s: %w", err)
+			return fmt.Errorf("unable to restart k3s via TERM signal to PID 1: %w", err)
 		}
 	}
-
 	return nil
 }