Improve update time of readiness state

pkg/kubelet/prober/prober_manager.go

@@ -17,7 +17,9 @@ limitations under the License.
 package prober
 
 import (
+	"k8s.io/apimachinery/pkg/util/clock"
 	"sync"
+	"time"
 
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/types"
@@ -94,6 +96,8 @@ type manager struct {
 
 	// prober executes the probe actions.
 	prober *prober
+
+	start time.Time
 }
 
 // NewManager creates a Manager for pod probing.
@@ -113,6 +117,7 @@ func NewManager(
 		livenessManager:  livenessManager,
 		startupManager:   startupManager,
 		workers:          make(map[probeKey]*worker),
+		start:            clock.RealClock{}.Now(),
 	}
 }
 
@@ -253,8 +258,15 @@ func (m *manager) UpdatePodStatus(podUID types.UID, podStatus *v1.PodStatus) {
 				ready = result == results.Success
 			} else {
 				// The check whether there is a probe which hasn't run yet.
-				_, exists := m.getWorker(podUID, c.Name, readiness)
-				ready = !exists
+				w, exists := m.getWorker(podUID, c.Name, readiness)
+				ready = !exists // no readinessProbe -> always ready
+				if exists {
+					// Trigger an immediate run of the readinessProbe to update ready state
+					select {
+					case w.manualTriggerCh <- struct{}{}:
+					default: // Non-blocking.
+					}
+				}
 			}
 			podStatus.ContainerStatuses[i].Ready = ready
 		}

pkg/kubelet/prober/worker.go

@@ -38,6 +38,9 @@ type worker struct {
 	// Channel for stopping the probe.
 	stopCh chan struct{}
 
+	// Channel for triggering the probe manually.
+	manualTriggerCh chan struct{}
+
 	// The pod containing this probe (read-only)
 	pod *v1.Pod
 
@@ -82,11 +85,12 @@ func newWorker(
 	container v1.Container) *worker {
 
 	w := &worker{
-		stopCh:       make(chan struct{}, 1), // Buffer so stop() can be non-blocking.
-		pod:          pod,
-		container:    container,
-		probeType:    probeType,
-		probeManager: m,
+		stopCh:          make(chan struct{}, 1), // Buffer so stop() can be non-blocking.
+		manualTriggerCh: make(chan struct{}, 1), // Buffer so prober_manager can do non-blocking calls to doProbe.
+		pod:             pod,
+		container:       container,
+		probeType:       probeType,
+		probeManager:    m,
 	}
 
 	switch probeType {
@@ -130,7 +134,10 @@ func (w *worker) run() {
 
 	// If kubelet restarted the probes could be started in rapid succession.
 	// Let the worker wait for a random portion of tickerPeriod before probing.
-	time.Sleep(time.Duration(rand.Float64() * float64(probeTickerPeriod)))
+	// Do it only if the kubelet has started recently.
+	if probeTickerPeriod > time.Since(w.probeManager.start) {
+		time.Sleep(time.Duration(rand.Float64() * float64(probeTickerPeriod)))
+	}
 
 	probeTicker := time.NewTicker(probeTickerPeriod)
 
@@ -154,6 +161,7 @@ probeLoop:
 		case <-w.stopCh:
 			break probeLoop
 		case <-probeTicker.C:
+		case <-w.manualTriggerCh:
 			// continue
 		}
 	}

test/e2e/common/node/container_probe.go

@@ -374,7 +374,8 @@ var _ = SIGDescribe("Probing container", func() {
 		Description: A Pod is created with startup and readiness probes. The Container is started by creating /tmp/startup after 45 seconds, delaying the ready state by this amount of time. This is similar to the "Pod readiness probe, with initial delay" test.
 	*/
 	ginkgo.It("should not be ready until startupProbe succeeds", func() {
-		cmd := []string{"/bin/sh", "-c", "echo ok >/tmp/health; sleep 45; echo ok >/tmp/startup; sleep 600"}
+		sleepBeforeStarted := time.Duration(45)
+		cmd := []string{"/bin/sh", "-c", fmt.Sprintf("echo ok >/tmp/health; sleep %d; echo ok >/tmp/startup; sleep 600", sleepBeforeStarted)}
 		readinessProbe := &v1.Probe{
 			Handler: v1.Handler{
 				Exec: &v1.ExecAction{
@@ -382,6 +383,7 @@ var _ = SIGDescribe("Probing container", func() {
 				},
 			},
 			InitialDelaySeconds: 0,
+			PeriodSeconds:       60,
 		}
 		startupProbe := &v1.Probe{
 			Handler: v1.Handler{
@@ -390,7 +392,8 @@ var _ = SIGDescribe("Probing container", func() {
 				},
 			},
 			InitialDelaySeconds: 0,
-			FailureThreshold:    60,
+			PeriodSeconds:       1,
+			FailureThreshold:    600,
 		}
 		p := podClient.Create(startupPodSpec(startupProbe, readinessProbe, nil, cmd))
 
@@ -414,9 +417,12 @@ var _ = SIGDescribe("Probing container", func() {
 		framework.ExpectNoError(err)
 
 		framework.Logf("Container started at %v, pod became ready at %v", startedTime, readyTime)
-		if readyTime.Sub(startedTime) < 40*time.Second {
+		if readyTime.Sub(startedTime) < sleepBeforeStarted*time.Second {
 			framework.Failf("Pod became ready before startupProbe succeeded")
 		}
+		if readyTime.Sub(startedTime) > (sleepBeforeStarted+20)*time.Second {
+			framework.Failf("Pod became ready more than 20s after startupProbe succeeded")
+		}
 	})
 })