fix: synchronize bootkube timeouts and various boot timeouts

When bootkube service fails, it can clean up manifests after itself, but
it only happens if we give it a chance to shut down cleanly. If boot
sequence times out, `machined` does emergency reboot and it doesn't let
`bootkube` do the cleanup.

So this fix has two paths:

* synchronize boot/bootstrap sequence timeouts with bootkube asset
timeout;

* cleanup bootkube-generated manifests and bootkube service startup.

Also logs errors on initial phases like `labelNodeAsMaster` to provide
some feedback on why boot is stuck.

Signed-off-by: Andrey Smirnov <smirnov.andrey@gmail.com>

internal/app/bootkube/main.go

@@ -11,6 +11,7 @@ import (
 	"os"
 	"path/filepath"
 
+	"github.com/hashicorp/go-multierror"
 	"github.com/kubernetes-sigs/bootkube/pkg/bootkube"
 	"github.com/kubernetes-sigs/bootkube/pkg/util"
 
@@ -59,6 +60,10 @@ func run() error {
 		return err
 	}
 
+	// cleanup manifests which might have been left from previous bootkube run
+	cleanupManifests("bootkube-*")                     //nolint: errcheck
+	cleanupManifests("kube-system-pod-checkpointer-*") //nolint: errcheck
+
 	defaultRequiredPods := []string{
 		"kube-system/pod-checkpointer",
 		"kube-system/kube-apiserver",
@@ -92,19 +97,8 @@ func run() error {
 			log.Printf("failed to cleanup bootkube assets dir %s", constants.AssetsDirectory)
 		}
 
-		bootstrapWildcard := filepath.Join(constants.ManifestsDirectory, "bootstrap-*")
-
-		var bootstrapFiles []string
-
-		bootstrapFiles, err = filepath.Glob(bootstrapWildcard)
-		if err != nil {
-			log.Printf("error finding bootstrap files in manifests dir %s", constants.ManifestsDirectory)
-		}
-
-		for _, bootstrapFile := range bootstrapFiles {
-			if err = os.Remove(bootstrapFile); err != nil {
-				log.Printf("error deleting bootstrap file in manifests dir : %s", err)
-			}
+		if err = cleanupManifests("bootstrap-*"); err != nil {
+			log.Printf("%s", err)
 		}
 	}()
 
@@ -117,6 +111,25 @@ func run() error {
 	return nil
 }
 
+func cleanupManifests(wildcard string) error {
+	bootstrapWildcard := filepath.Join(constants.ManifestsDirectory, wildcard)
+
+	bootstrapFiles, err := filepath.Glob(bootstrapWildcard)
+	if err != nil {
+		return fmt.Errorf("error finding bootstrap files in manifests dir %s", constants.ManifestsDirectory)
+	}
+
+	var multiErr *multierror.Error
+
+	for _, bootstrapFile := range bootstrapFiles {
+		if err = os.Remove(bootstrapFile); err != nil {
+			multiErr = multierror.Append(multiErr, fmt.Errorf("error deleting bootstrap file in manifests dir: %s", err))
+		}
+	}
+
+	return multiErr.ErrorOrNil()
+}
+
 func main() {
 	if err := run(); err != nil {
 		log.Fatalf("bootkube failed: %s", err)

internal/app/machined/pkg/runtime/v1alpha1/v1alpha1_sequencer_tasks.go

@@ -732,7 +732,7 @@ func StartAllServices(seq runtime.Sequence, data interface{}) (runtime.TaskExecu
 			all = append(all, cond)
 		}
 
-		ctx, cancel := context.WithTimeout(ctx, 10*time.Minute)
+		ctx, cancel := context.WithTimeout(ctx, constants.BootkubeRunTimeout)
 
 		defer cancel()
 
@@ -1456,7 +1456,7 @@ func LabelNodeAsMaster(seq runtime.Sequence, data interface{}) (runtime.TaskExec
 			return err
 		}
 
-		err = retry.Constant(10*time.Minute, retry.WithUnits(3*time.Second)).Retry(func() error {
+		err = retry.Constant(constants.NodeReadyTimeout, retry.WithUnits(3*time.Second), retry.WithErrorLogging(true)).Retry(func() error {
 			if err = h.LabelNodeAsMaster(hostname, !r.Config().Cluster().ScheduleOnMasters()); err != nil {
 				return retry.ExpectedError(err)
 			}
@@ -1778,7 +1778,7 @@ func BootstrapKubernetes(seq runtime.Sequence, data interface{}) (runtime.TaskEx
 
 		system.Services(r).LoadAndStart(svc)
 
-		ctx, cancel := context.WithTimeout(ctx, 10*time.Minute)
+		ctx, cancel := context.WithTimeout(ctx, constants.BootkubeRunTimeout)
 		defer cancel()
 
 		return system.WaitForService(system.StateEventFinished, svc.ID(r)).Wait(ctx)

pkg/kubernetes/kubernetes.go

@@ -251,7 +251,7 @@ func (h *Client) LabelNodeAsMaster(name string, taintNoSchedule bool) (err error
 
 // WaitUntilReady waits for a node to be ready.
 func (h *Client) WaitUntilReady(name string) error {
-	return retry.Exponential(3*time.Minute, retry.WithUnits(250*time.Millisecond), retry.WithJitter(50*time.Millisecond)).Retry(func() error {
+	return retry.Exponential(10*time.Minute, retry.WithUnits(250*time.Millisecond), retry.WithJitter(50*time.Millisecond), retry.WithErrorLogging(true)).Retry(func() error {
 		attemptCtx, attemptCtxCancel := context.WithTimeout(context.TODO(), 30*time.Second)
 		defer attemptCtxCancel()
 

pkg/machinery/constants/constants.go

@@ -377,6 +377,16 @@ const (
 	// InitializedKey is the key used to indicate if the cluster has been
 	// initialized.
 	InitializedKey = "initialized"
+
+	// BootkubeAssetTimeout is the constant in bootkube implementation.
+	BootkubeAssetTimeout = 20 * time.Minute
+
+	// BootkubeRunTimeout is the timeout to run bootkube.
+	BootkubeRunTimeout = BootkubeAssetTimeout + 5*time.Minute
+
+	// NodeReadyTimeout is the timeout to wait for the node to be ready (CNI to be running).
+	// For bootstrap API, this includes time to run bootkube.
+	NodeReadyTimeout = BootkubeRunTimeout
 )
 
 // See https://linux.die.net/man/3/klogctl