diff --git a/pkg/kubelet/container/cache.go b/pkg/kubelet/container/cache.go
new file mode 100644
index 00000000000..55722ec5799
--- /dev/null
+++ b/pkg/kubelet/container/cache.go
@@ -0,0 +1,199 @@
+/*
+Copyright 2015 The Kubernetes Authors All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package container
+
+import (
+	"sync"
+	"time"
+
+	"k8s.io/kubernetes/pkg/types"
+)
+
+// Cache stores the PodStatus for the pods. It represents *all* the visible
+// pods/containers in the container runtime. All cache entries are at least as
+// new or newer than the global timestamp (set by UpdateTime()), while
+// individual entries may be slightly newer than the global timestamp. If a pod
+// has no states known by the runtime, Cache returns an empty PodStatus object
+// with ID populated.
+//
+// Cache provides two methods to retrive the PodStatus: the non-blocking Get()
+// and the blocking GetNewerThan() method. The component responsible for
+// populating the cache is expected to call Delete() to explicitly free the
+// cache entries.
+type Cache interface {
+	Get(types.UID) (*PodStatus, error)
+	Set(types.UID, *PodStatus, error, time.Time)
+	// GetNewerThan is a blocking call that only returns the status
+	// when it is newer than the given time.
+	GetNewerThan(types.UID, time.Time) (*PodStatus, error)
+	Delete(types.UID)
+	UpdateTime(time.Time)
+}
+
+type data struct {
+	// Status of the pod.
+	status *PodStatus
+	// Error got when trying to inspect the pod.
+	err error
+	// Time when the data was last modfied.
+	modified time.Time
+}
+
+type subRecord struct {
+	time time.Time
+	ch   chan *data
+}
+
+// cache implements Cache.
+type cache struct {
+	// Lock which guards all internal data structures.
+	lock sync.RWMutex
+	// Map that stores the pod statuses.
+	pods map[types.UID]*data
+	// A global timestamp represents how fresh the cached data is. All
+	// cache content is at the least newer than this timestamp. Note that the
+	// timestamp is nil after initialization, and will only become non-nil when
+	// it is ready to serve the cached statuses.
+	timestamp *time.Time
+	// Map that stores the subscriber records.
+	subscribers map[types.UID][]*subRecord
+}
+
+// NewCache creates a pod cache.
+func NewCache() Cache {
+	return &cache{pods: map[types.UID]*data{}, subscribers: map[types.UID][]*subRecord{}}
+}
+
+// Get returns the PodStatus for the pod; callers are expected not to
+// modify the objects returned.
+func (c *cache) Get(id types.UID) (*PodStatus, error) {
+	c.lock.RLock()
+	defer c.lock.RUnlock()
+	d := c.get(id)
+	return d.status, d.err
+}
+
+func (c *cache) GetNewerThan(id types.UID, minTime time.Time) (*PodStatus, error) {
+	ch := c.subscribe(id, minTime)
+	d := <-ch
+	return d.status, d.err
+}
+
+// Set sets the PodStatus for the pod.
+func (c *cache) Set(id types.UID, status *PodStatus, err error, timestamp time.Time) {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	defer c.notify(id, timestamp)
+	c.pods[id] = &data{status: status, err: err, modified: timestamp}
+}
+
+// Delete removes the entry of the pod.
+func (c *cache) Delete(id types.UID) {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	delete(c.pods, id)
+}
+
+//  UpdateTime modifies the global timestamp of the cache and notify
+//  subscribers if needed.
+func (c *cache) UpdateTime(timestamp time.Time) {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	c.timestamp = &timestamp
+	// Notify all the subscribers if the condition is met.
+	for id := range c.subscribers {
+		c.notify(id, *c.timestamp)
+	}
+}
+
+func makeDefaultData(id types.UID) *data {
+	return &data{status: &PodStatus{ID: id}, err: nil}
+}
+
+func (c *cache) get(id types.UID) *data {
+	d, ok := c.pods[id]
+	if !ok {
+		// Cache should store *all* pod/container information known by the
+		// container runtime. A cache miss indicates that there are no states
+		// regarding the pod last time we queried the container runtime.
+		// What this *really* means is that there are no visible pod/containers
+		// associated with this pod. Simply return an default (mostly empty)
+		// PodStatus to reflect this.
+		return makeDefaultData(id)
+	}
+	return d
+}
+
+// getIfNewerThan returns the data it is newer than the given time.
+// Otherwise, it returns nil. The caller should acquire the lock.
+func (c *cache) getIfNewerThan(id types.UID, minTime time.Time) *data {
+	d, ok := c.pods[id]
+	globalTimestampIsNewer := (c.timestamp != nil && c.timestamp.After(minTime))
+	if !ok && globalTimestampIsNewer {
+		// Status is not cached, but the global timestamp is newer than
+		// minTime, return the default status.
+		return makeDefaultData(id)
+	}
+	if ok && (d.modified.After(minTime) || globalTimestampIsNewer) {
+		// Status is cached, return status if either of the following is true.
+		//   * status was modified after minTime
+		//   * the global timestamp of the cache is newer than minTime.
+		return d
+	}
+	// The pod status is not ready.
+	return nil
+}
+
+// notify sends notifications for pod with the given id, if the requirements
+// are met. Note that the caller should acquire the lock.
+func (c *cache) notify(id types.UID, timestamp time.Time) {
+	list, ok := c.subscribers[id]
+	if !ok {
+		// No one to notify.
+		return
+	}
+	newList := []*subRecord{}
+	for i, r := range list {
+		if timestamp.Before(r.time) {
+			// Doesn't meet the time requirement; keep the record.
+			newList = append(newList, list[i])
+			continue
+		}
+		r.ch <- c.get(id)
+		close(r.ch)
+	}
+	if len(newList) == 0 {
+		delete(c.subscribers, id)
+	} else {
+		c.subscribers[id] = newList
+	}
+}
+
+func (c *cache) subscribe(id types.UID, timestamp time.Time) chan *data {
+	ch := make(chan *data, 1)
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	d := c.getIfNewerThan(id, timestamp)
+	if d != nil {
+		// If the cache entry is ready, send the data and return immediatly.
+		ch <- d
+		return ch
+	}
+	// Add the subscription record.
+	c.subscribers[id] = append(c.subscribers[id], &subRecord{time: timestamp, ch: ch})
+	return ch
+}
diff --git a/pkg/kubelet/container/cache_test.go b/pkg/kubelet/container/cache_test.go
new file mode 100644
index 00000000000..5755005d841
--- /dev/null
+++ b/pkg/kubelet/container/cache_test.go
@@ -0,0 +1,210 @@
+/*
+Copyright 2015 The Kubernetes Authors All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package container
+
+import (
+	"fmt"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"k8s.io/kubernetes/pkg/types"
+)
+
+func newTestCache() *cache {
+	c := NewCache()
+	return c.(*cache)
+}
+
+func TestCacheNotInitialized(t *testing.T) {
+	cache := newTestCache()
+	// If the global timestamp is not set, always return nil.
+	d := cache.getIfNewerThan(types.UID("1234"), time.Time{})
+	assert.True(t, d == nil, "should return nil since cache is not initialized")
+}
+
+func getTestPodIDAndStatus(numContainers int) (types.UID, *PodStatus) {
+	id := types.UID(strconv.FormatInt(time.Now().UnixNano(), 10))
+	name := fmt.Sprintf("cache-foo-%s", string(id))
+	namespace := "ns"
+	var status *PodStatus
+	if numContainers > 0 {
+		status = &PodStatus{ID: id, Name: name, Namespace: namespace}
+	} else {
+		status = &PodStatus{ID: id}
+	}
+	for i := 0; i < numContainers; i++ {
+		status.ContainerStatuses = append(status.ContainerStatuses, &ContainerStatus{Name: string(i)})
+	}
+	return id, status
+}
+
+func TestGetIfNewerThanWhenPodExists(t *testing.T) {
+	cache := newTestCache()
+	timestamp := time.Now()
+
+	cases := []struct {
+		cacheTime time.Time
+		modified  time.Time
+		expected  bool
+	}{
+		{
+			// Both the global cache timestamp and the modified time are newer
+			// than the timestamp.
+			cacheTime: timestamp.Add(time.Second),
+			modified:  timestamp,
+			expected:  true,
+		},
+		{
+			// Global cache timestamp is newer, but the pod entry modified
+			// time is older than the given timestamp. This means that the
+			// entry is up-to-date even though it hasn't changed for a while.
+			cacheTime: timestamp.Add(time.Second),
+			modified:  timestamp.Add(-time.Second * 10),
+			expected:  true,
+		},
+		{
+			// Global cache timestamp is older, but the pod entry modified
+			// time is newer than the given timestamp. This means that the
+			// entry is up-to-date but the rest of the cache are still being
+			// updated.
+			cacheTime: timestamp.Add(-time.Second),
+			modified:  timestamp.Add(time.Second * 3),
+			expected:  true,
+		},
+		{
+			// Both the global cache timestamp and the modified time are older
+			// than the given timestamp.
+			cacheTime: timestamp.Add(-time.Second),
+			modified:  timestamp.Add(-time.Second),
+			expected:  false,
+		},
+	}
+	for i, c := range cases {
+		podID, status := getTestPodIDAndStatus(2)
+		cache.UpdateTime(c.cacheTime)
+		cache.Set(podID, status, nil, c.modified)
+		d := cache.getIfNewerThan(podID, timestamp)
+		assert.Equal(t, c.expected, d != nil, "test[%d]", i)
+	}
+}
+
+func TestGetPodNewerThanWhenPodDoesNotExist(t *testing.T) {
+	cache := newTestCache()
+	cacheTime := time.Now()
+	cache.UpdateTime(cacheTime)
+	podID := types.UID("1234")
+
+	cases := []struct {
+		timestamp time.Time
+		expected  bool
+	}{
+		{
+			timestamp: cacheTime.Add(-time.Second),
+			expected:  true,
+		},
+		{
+			timestamp: cacheTime.Add(time.Second),
+			expected:  false,
+		},
+	}
+	for i, c := range cases {
+		d := cache.getIfNewerThan(podID, c.timestamp)
+		assert.Equal(t, c.expected, d != nil, "test[%d]", i)
+	}
+}
+
+func TestCacheSetAndGet(t *testing.T) {
+	cache := NewCache()
+	cases := []struct {
+		numContainers int
+		error         error
+	}{
+		{numContainers: 3, error: nil},
+		{numContainers: 2, error: fmt.Errorf("unable to get status")},
+		{numContainers: 0, error: nil},
+	}
+	for i, c := range cases {
+		podID, status := getTestPodIDAndStatus(c.numContainers)
+		cache.Set(podID, status, c.error, time.Time{})
+		// Read back the status and error stored in cache and make sure they
+		// match the original ones.
+		actualStatus, actualErr := cache.Get(podID)
+		assert.Equal(t, status, actualStatus, "test[%d]", i)
+		assert.Equal(t, c.error, actualErr, "test[%d]", i)
+	}
+}
+
+func TestCacheGetPodDoesNotExist(t *testing.T) {
+	cache := NewCache()
+	podID, status := getTestPodIDAndStatus(0)
+	// If the pod does not exist in cache, cache should return an status
+	// object with id filled.
+	actualStatus, actualErr := cache.Get(podID)
+	assert.Equal(t, status, actualStatus)
+	assert.Equal(t, nil, actualErr)
+}
+
+func TestDelete(t *testing.T) {
+	cache := &cache{pods: map[types.UID]*data{}}
+	// Write a new pod status into the cache.
+	podID, status := getTestPodIDAndStatus(3)
+	cache.Set(podID, status, nil, time.Time{})
+	actualStatus, actualErr := cache.Get(podID)
+	assert.Equal(t, status, actualStatus)
+	assert.Equal(t, nil, actualErr)
+	// Delete the pod from cache, and verify that we get an empty status.
+	cache.Delete(podID)
+	expectedStatus := &PodStatus{ID: podID}
+	actualStatus, actualErr = cache.Get(podID)
+	assert.Equal(t, expectedStatus, actualStatus)
+	assert.Equal(t, nil, actualErr)
+}
+
+func verifyNotification(t *testing.T, ch chan *data, expectNotification bool) {
+	if expectNotification {
+		assert.True(t, len(ch) > 0, "Did not receive notification")
+	} else {
+		assert.True(t, len(ch) < 1, "Should not have triggered the notification")
+	}
+	// Drain the channel.
+	for i := 0; i < len(ch); i++ {
+		<-ch
+	}
+}
+
+func TestRegisterNotification(t *testing.T) {
+	cache := newTestCache()
+	cacheTime := time.Now()
+	cache.UpdateTime(cacheTime)
+
+	podID, status := getTestPodIDAndStatus(1)
+	ch := cache.subscribe(podID, cacheTime.Add(time.Second))
+	verifyNotification(t, ch, false)
+	cache.Set(podID, status, nil, cacheTime.Add(time.Second))
+	// The Set operation should've triggered the notification.
+	verifyNotification(t, ch, true)
+
+	podID, _ = getTestPodIDAndStatus(1)
+
+	ch = cache.subscribe(podID, cacheTime.Add(time.Second))
+	verifyNotification(t, ch, false)
+	cache.UpdateTime(cacheTime.Add(time.Second * 2))
+	// The advance of cache timestamp should've triggered the notification.
+	verifyNotification(t, ch, true)
+}
diff --git a/pkg/kubelet/container/runtime.go b/pkg/kubelet/container/runtime.go
index 2173a6c55ea..ec215654d19 100644
--- a/pkg/kubelet/container/runtime.go
+++ b/pkg/kubelet/container/runtime.go
@@ -442,6 +442,15 @@ func (p *Pod) FindContainerByName(containerName string) *Container {
 	return nil
 }
 
+func (p *Pod) FindContainerByID(id ContainerID) *Container {
+	for _, c := range p.Containers {
+		if c.ID == id {
+			return c
+		}
+	}
+	return nil
+}
+
 // ToAPIPod converts Pod to api.Pod. Note that if a field in api.Pod has no
 // corresponding field in Pod, the field would not be populated.
 func (p *Pod) ToAPIPod() *api.Pod {
diff --git a/pkg/kubelet/kubelet.go b/pkg/kubelet/kubelet.go
index 713748a91ad..5a7fc0b1817 100644
--- a/pkg/kubelet/kubelet.go
+++ b/pkg/kubelet/kubelet.go
@@ -367,7 +367,7 @@ func NewMainKubelet(
 			serializeImagePulls,
 		)
 
-		klet.pleg = pleg.NewGenericPLEG(klet.containerRuntime, plegChannelCapacity, plegRelistPeriod)
+		klet.pleg = pleg.NewGenericPLEG(klet.containerRuntime, plegChannelCapacity, plegRelistPeriod, nil)
 	case "rkt":
 		conf := &rkt.Config{
 			Path:               rktPath,
@@ -388,7 +388,7 @@ func NewMainKubelet(
 			return nil, err
 		}
 		klet.containerRuntime = rktRuntime
-		klet.pleg = pleg.NewGenericPLEG(klet.containerRuntime, plegChannelCapacity, plegRelistPeriod)
+		klet.pleg = pleg.NewGenericPLEG(klet.containerRuntime, plegChannelCapacity, plegRelistPeriod, nil)
 
 		// No Docker daemon to put in a container.
 		dockerDaemonContainer = ""
diff --git a/pkg/kubelet/kubelet_test.go b/pkg/kubelet/kubelet_test.go
index 4698371ebc1..00e8d09b4b7 100644
--- a/pkg/kubelet/kubelet_test.go
+++ b/pkg/kubelet/kubelet_test.go
@@ -183,7 +183,7 @@ func newTestKubelet(t *testing.T) *TestKubelet {
 	}
 	kubelet.workQueue = queue.NewBasicWorkQueue()
 	// Relist period does not affect the tests.
-	kubelet.pleg = pleg.NewGenericPLEG(fakeRuntime, 100, time.Hour)
+	kubelet.pleg = pleg.NewGenericPLEG(fakeRuntime, 100, time.Hour, nil)
 	kubelet.clock = fakeClock
 	return &TestKubelet{kubelet, fakeRuntime, mockCadvisor, fakeKubeClient, fakeMirrorClient, fakeClock}
 }
diff --git a/pkg/kubelet/pleg/generic.go b/pkg/kubelet/pleg/generic.go
index 7816062cf89..07ecf71ed1b 100644
--- a/pkg/kubelet/pleg/generic.go
+++ b/pkg/kubelet/pleg/generic.go
@@ -25,6 +25,7 @@ import (
 	"k8s.io/kubernetes/pkg/kubelet/metrics"
 	"k8s.io/kubernetes/pkg/types"
 	"k8s.io/kubernetes/pkg/util"
+	"k8s.io/kubernetes/pkg/util/sets"
 )
 
 // GenericPLEG is an extremely simple generic PLEG that relies solely on
@@ -49,24 +50,54 @@ type GenericPLEG struct {
 	runtime kubecontainer.Runtime
 	// The channel from which the subscriber listens events.
 	eventChannel chan *PodLifecycleEvent
-	// The internal cache for container information.
-	containers map[string]containerInfo
+	// The internal cache for pod/container information.
+	podRecords podRecords
 	// Time of the last relisting.
 	lastRelistTime time.Time
+	// Cache for storing the runtime states required for syncing pods.
+	cache kubecontainer.Cache
 }
 
-type containerInfo struct {
-	podID types.UID
-	state kubecontainer.ContainerState
+// plegContainerState has a one-to-one mapping to the
+// kubecontainer.ContainerState except for the non-existent state. This state
+// is introduced here to complete the state transition scenarios.
+type plegContainerState string
+
+const (
+	plegContainerRunning     plegContainerState = "running"
+	plegContainerExited      plegContainerState = "exited"
+	plegContainerUnknown     plegContainerState = "unknown"
+	plegContainerNonExistent plegContainerState = "non-existent"
+)
+
+func convertState(state kubecontainer.ContainerState) plegContainerState {
+	switch state {
+	case kubecontainer.ContainerStateRunning:
+		return plegContainerRunning
+	case kubecontainer.ContainerStateExited:
+		return plegContainerExited
+	case kubecontainer.ContainerStateUnknown:
+		return plegContainerUnknown
+	default:
+		panic(fmt.Sprintf("unrecognized container state: %v", state))
+	}
 }
 
+type podRecord struct {
+	old     *kubecontainer.Pod
+	current *kubecontainer.Pod
+}
+
+type podRecords map[types.UID]*podRecord
+
 func NewGenericPLEG(runtime kubecontainer.Runtime, channelCapacity int,
-	relistPeriod time.Duration) PodLifecycleEventGenerator {
+	relistPeriod time.Duration, cache kubecontainer.Cache) PodLifecycleEventGenerator {
 	return &GenericPLEG{
 		relistPeriod: relistPeriod,
 		runtime:      runtime,
 		eventChannel: make(chan *PodLifecycleEvent, channelCapacity),
-		containers:   make(map[string]containerInfo),
+		podRecords:   make(podRecords),
+		cache:        cache,
 	}
 }
 
@@ -82,18 +113,30 @@ func (g *GenericPLEG) Start() {
 	go util.Until(g.relist, g.relistPeriod, util.NeverStop)
 }
 
-func generateEvent(podID types.UID, cid string, oldState, newState kubecontainer.ContainerState) *PodLifecycleEvent {
+func generateEvent(podID types.UID, cid string, oldState, newState plegContainerState) *PodLifecycleEvent {
+	glog.V(7).Infof("GenericPLEG: %v/%v: %v -> %v", podID, cid, oldState, newState)
 	if newState == oldState {
 		return nil
 	}
 	switch newState {
-	case kubecontainer.ContainerStateRunning:
+	case plegContainerRunning:
 		return &PodLifecycleEvent{ID: podID, Type: ContainerStarted, Data: cid}
-	case kubecontainer.ContainerStateExited:
+	case plegContainerExited:
 		return &PodLifecycleEvent{ID: podID, Type: ContainerDied, Data: cid}
-	case kubecontainer.ContainerStateUnknown:
+	case plegContainerUnknown:
 		// Don't generate any event if the status is unknown.
 		return nil
+	case plegContainerNonExistent:
+		// We report "ContainerDied" when container was stopped OR removed. We
+		// may want to distinguish the two cases in the future.
+		switch oldState {
+		case plegContainerExited:
+			// We already reported that the container died before. There is no
+			// need to do it again.
+			return nil
+		default:
+			return &PodLifecycleEvent{ID: podID, Type: ContainerDied, Data: cid}
+		}
 	default:
 		panic(fmt.Sprintf("unrecognized container state: %v", newState))
 	}
@@ -116,40 +159,173 @@ func (g *GenericPLEG) relist() {
 	}()
 
 	// Get all the pods.
-	pods, err := g.runtime.GetPods(true)
+	podList, err := g.runtime.GetPods(true)
 	if err != nil {
 		glog.Errorf("GenericPLEG: Unable to retrieve pods: %v", err)
 		return
 	}
+	pods := kubecontainer.Pods(podList)
+	for _, pod := range pods {
+		g.podRecords.setCurrent(pod)
+	}
 
-	events := []*PodLifecycleEvent{}
-	containers := make(map[string]containerInfo, len(g.containers))
-	// Create a new containers map, compares container statuses, and generates
-	// correspoinding events.
-	for _, p := range pods {
-		for _, c := range p.Containers {
-			cid := c.ID.ID
-			// Get the of existing container info. Defaults to state unknown.
-			oldState := kubecontainer.ContainerStateUnknown
-			if info, ok := g.containers[cid]; ok {
-				oldState = info.state
-			}
-			// Generate an event if required.
-			glog.V(7).Infof("GenericPLEG: %v/%v: %v -> %v", p.ID, cid, oldState, c.State)
-			if e := generateEvent(p.ID, cid, oldState, c.State); e != nil {
-				events = append(events, e)
-			}
-			// Write to the new cache.
-			containers[cid] = containerInfo{podID: p.ID, state: c.State}
+	// Compare the old and the current pods, and generate events.
+	eventsByPodID := map[types.UID][]*PodLifecycleEvent{}
+	for pid := range g.podRecords {
+		oldPod := g.podRecords.getOld(pid)
+		pod := g.podRecords.getCurrent(pid)
+		// Get all containers in the old and the new pod.
+		allContainers := getContainersFromPods(oldPod, pod)
+		for _, container := range allContainers {
+			e := computeEvent(oldPod, pod, &container.ID)
+			updateEvents(eventsByPodID, e)
 		}
 	}
 
-	// Swap the container info cache. This is purely to avoid the need of
-	// garbage collection.
-	g.containers = containers
+	// If there are events associated with a pod, we should update the
+	// podCache.
+	for pid, events := range eventsByPodID {
+		pod := g.podRecords.getCurrent(pid)
+		if g.cacheEnabled() {
+			// updateCache() will inspect the pod and update the cache. If an
+			// error occurs during the inspection, we want PLEG to retry again
+			// in the next relist. To achieve this, we do not update the
+			// associated podRecord of the pod, so that the change will be
+			// detect again in the next relist.
+			// TODO: If many pods changed during the same relist period,
+			// inspecting the pod and getting the PodStatus to update the cache
+			// serially may take a while. We should be aware of this and
+			// parallelize if needed.
+			if err := g.updateCache(pod, pid); err != nil {
+				glog.Errorf("PLEG: Ignoring events for pod %s/%s: %v", pod.Name, pod.Namespace, err)
+				continue
+			}
+		}
+		// Update the internal storage and send out the events.
+		g.podRecords.update(pid)
+		for i := range events {
+			g.eventChannel <- events[i]
+		}
+	}
 
-	// Send out the events.
-	for i := range events {
-		g.eventChannel <- events[i]
+	if g.cacheEnabled() {
+		// Update the cache timestamp.  This needs to happen *after*
+		// all pods have been properly updated in the cache.
+		g.cache.UpdateTime(timestamp)
 	}
 }
+
+func getContainersFromPods(pods ...*kubecontainer.Pod) []*kubecontainer.Container {
+	cidSet := sets.NewString()
+	var containers []*kubecontainer.Container
+	for _, p := range pods {
+		if p == nil {
+			continue
+		}
+		for _, c := range p.Containers {
+			cid := string(c.ID.ID)
+			if cidSet.Has(cid) {
+				continue
+			}
+			cidSet.Insert(cid)
+			containers = append(containers, c)
+		}
+	}
+	return containers
+}
+
+func computeEvent(oldPod, newPod *kubecontainer.Pod, cid *kubecontainer.ContainerID) *PodLifecycleEvent {
+	var pid types.UID
+	if oldPod != nil {
+		pid = oldPod.ID
+	} else if newPod != nil {
+		pid = newPod.ID
+	}
+	oldState := getContainerState(oldPod, cid)
+	newState := getContainerState(newPod, cid)
+	return generateEvent(pid, cid.ID, oldState, newState)
+}
+
+func (g *GenericPLEG) cacheEnabled() bool {
+	return g.cache != nil
+}
+
+func (g *GenericPLEG) updateCache(pod *kubecontainer.Pod, pid types.UID) error {
+	if pod == nil {
+		// The pod is missing in the current relist. This means that
+		// the pod has no visible (active or inactive) containers.
+		glog.V(8).Infof("PLEG: Delete status for pod %q", string(pid))
+		g.cache.Delete(pid)
+		return nil
+	}
+	timestamp := time.Now()
+	// TODO: Consider adding a new runtime method
+	// GetPodStatus(pod *kubecontainer.Pod) so that Docker can avoid listing
+	// all containers again.
+	status, err := g.runtime.GetPodStatus(pod.ID, pod.Name, pod.Namespace)
+	glog.V(8).Infof("PLEG: Write status for %s/%s: %+v (err: %v)", pod.Name, pod.Namespace, status, err)
+	g.cache.Set(pod.ID, status, err, timestamp)
+	return err
+}
+
+func updateEvents(eventsByPodID map[types.UID][]*PodLifecycleEvent, e *PodLifecycleEvent) {
+	if e == nil {
+		return
+	}
+	eventsByPodID[e.ID] = append(eventsByPodID[e.ID], e)
+}
+
+func getContainerState(pod *kubecontainer.Pod, cid *kubecontainer.ContainerID) plegContainerState {
+	// Default to the non-existent state.
+	state := plegContainerNonExistent
+	if pod == nil {
+		return state
+	}
+	container := pod.FindContainerByID(*cid)
+	if container == nil {
+		return state
+	}
+	return convertState(container.State)
+}
+
+func (pr podRecords) getOld(id types.UID) *kubecontainer.Pod {
+	r, ok := pr[id]
+	if !ok {
+		return nil
+	}
+	return r.old
+}
+
+func (pr podRecords) getCurrent(id types.UID) *kubecontainer.Pod {
+	r, ok := pr[id]
+	if !ok {
+		return nil
+	}
+	return r.current
+}
+
+func (pr podRecords) setCurrent(pod *kubecontainer.Pod) {
+	if r, ok := pr[pod.ID]; ok {
+		r.current = pod
+		return
+	}
+	pr[pod.ID] = &podRecord{current: pod}
+}
+
+func (pr podRecords) update(id types.UID) {
+	r, ok := pr[id]
+	if !ok {
+		return
+	}
+	pr.updateInternal(id, r)
+}
+
+func (pr podRecords) updateInternal(id types.UID, r *podRecord) {
+	if r.current == nil {
+		// Pod no longer exists; delete the entry.
+		delete(pr, id)
+		return
+	}
+	r.old = r.current
+	r.current = nil
+}
diff --git a/pkg/kubelet/pleg/generic_test.go b/pkg/kubelet/pleg/generic_test.go
index 7f9562a4f53..c702cf1a0ba 100644
--- a/pkg/kubelet/pleg/generic_test.go
+++ b/pkg/kubelet/pleg/generic_test.go
@@ -17,12 +17,15 @@ limitations under the License.
 package pleg
 
 import (
+	"fmt"
 	"reflect"
 	"sort"
 	"testing"
 	"time"
 
+	"github.com/stretchr/testify/assert"
 	kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
+	"k8s.io/kubernetes/pkg/types"
 	"k8s.io/kubernetes/pkg/util"
 )
 
@@ -43,7 +46,7 @@ func newTestGenericPLEG() *TestGenericPLEG {
 		relistPeriod: time.Hour,
 		runtime:      fakeRuntime,
 		eventChannel: make(chan *PodLifecycleEvent, 100),
-		containers:   make(map[string]containerInfo),
+		podRecords:   make(podRecords),
 	}
 	return &TestGenericPLEG{pleg: pleg, runtime: fakeRuntime}
 }
@@ -79,7 +82,7 @@ func verifyEvents(t *testing.T, expected, actual []*PodLifecycleEvent) {
 	sort.Sort(sortableEvents(expected))
 	sort.Sort(sortableEvents(actual))
 	if !reflect.DeepEqual(expected, actual) {
-		t.Errorf("Actual events differ from the expected; diff: %v", util.ObjectDiff(expected, actual))
+		t.Errorf("Actual events differ from the expected; diff:\n %v", util.ObjectDiff(expected, actual))
 	}
 }
 
@@ -87,7 +90,6 @@ func TestRelisting(t *testing.T) {
 	testPleg := newTestGenericPLEG()
 	pleg, runtime := testPleg.pleg, testPleg.runtime
 	ch := pleg.Watch()
-
 	// The first relist should send a PodSync event to each pod.
 	runtime.AllPodList = []*kubecontainer.Pod{
 		{
@@ -146,3 +148,172 @@ func TestRelisting(t *testing.T) {
 	actual = getEventsFromChannel(ch)
 	verifyEvents(t, expected, actual)
 }
+
+func TestReportMissingContainers(t *testing.T) {
+	testPleg := newTestGenericPLEG()
+	pleg, runtime := testPleg.pleg, testPleg.runtime
+	ch := pleg.Watch()
+	runtime.AllPodList = []*kubecontainer.Pod{
+		{
+			ID: "1234",
+			Containers: []*kubecontainer.Container{
+				createTestContainer("c1", kubecontainer.ContainerStateRunning),
+				createTestContainer("c2", kubecontainer.ContainerStateRunning),
+				createTestContainer("c3", kubecontainer.ContainerStateExited),
+			},
+		},
+	}
+	// Drain the events from the channel
+	pleg.relist()
+	getEventsFromChannel(ch)
+
+	// Container c2 was stopped and removed between relists. We should report
+	// the event. The exited container c3 was garbage collected (i.e., removed)
+	// between relists. We should ignore that event.
+	runtime.AllPodList = []*kubecontainer.Pod{
+		{
+			ID: "1234",
+			Containers: []*kubecontainer.Container{
+				createTestContainer("c1", kubecontainer.ContainerStateRunning),
+			},
+		},
+	}
+	pleg.relist()
+	expected := []*PodLifecycleEvent{
+		{ID: "1234", Type: ContainerDied, Data: "c2"},
+	}
+	actual := getEventsFromChannel(ch)
+	verifyEvents(t, expected, actual)
+}
+
+func TestReportMissingPods(t *testing.T) {
+	testPleg := newTestGenericPLEG()
+	pleg, runtime := testPleg.pleg, testPleg.runtime
+	ch := pleg.Watch()
+	runtime.AllPodList = []*kubecontainer.Pod{
+		{
+			ID: "1234",
+			Containers: []*kubecontainer.Container{
+				createTestContainer("c2", kubecontainer.ContainerStateRunning),
+			},
+		},
+	}
+	// Drain the events from the channel
+	pleg.relist()
+	getEventsFromChannel(ch)
+
+	// Container c2 was stopped and removed between relists. We should report
+	// the event.
+	runtime.AllPodList = []*kubecontainer.Pod{}
+	pleg.relist()
+	expected := []*PodLifecycleEvent{
+		{ID: "1234", Type: ContainerDied, Data: "c2"},
+	}
+	actual := getEventsFromChannel(ch)
+	verifyEvents(t, expected, actual)
+}
+
+func newTestGenericPLEGWithRuntimeMock() (*GenericPLEG, *kubecontainer.Mock) {
+	runtimeMock := &kubecontainer.Mock{}
+	pleg := &GenericPLEG{
+		relistPeriod: time.Hour,
+		runtime:      runtimeMock,
+		eventChannel: make(chan *PodLifecycleEvent, 100),
+		podRecords:   make(podRecords),
+		cache:        kubecontainer.NewCache(),
+	}
+	return pleg, runtimeMock
+}
+
+func createTestPodsStatusesAndEvents(num int) ([]*kubecontainer.Pod, []*kubecontainer.PodStatus, []*PodLifecycleEvent) {
+	var pods []*kubecontainer.Pod
+	var statuses []*kubecontainer.PodStatus
+	var events []*PodLifecycleEvent
+	for i := 0; i < num; i++ {
+		id := types.UID(fmt.Sprintf("test-pod-%d", i))
+		cState := kubecontainer.ContainerStateRunning
+		container := createTestContainer(fmt.Sprintf("c%d", i), cState)
+		pod := &kubecontainer.Pod{
+			ID:         id,
+			Containers: []*kubecontainer.Container{container},
+		}
+		status := &kubecontainer.PodStatus{
+			ID:                id,
+			ContainerStatuses: []*kubecontainer.ContainerStatus{{ID: container.ID, State: cState}},
+		}
+		event := &PodLifecycleEvent{ID: pod.ID, Type: ContainerStarted, Data: container.ID.ID}
+		pods = append(pods, pod)
+		statuses = append(statuses, status)
+		events = append(events, event)
+
+	}
+	return pods, statuses, events
+}
+
+func TestRelistWithCache(t *testing.T) {
+	pleg, runtimeMock := newTestGenericPLEGWithRuntimeMock()
+	ch := pleg.Watch()
+
+	pods, statuses, events := createTestPodsStatusesAndEvents(2)
+	runtimeMock.On("GetPods", true).Return(pods, nil)
+	runtimeMock.On("GetPodStatus", pods[0].ID, "", "").Return(statuses[0], nil).Once()
+	// Inject an error when querying runtime for the pod status for pods[1].
+	statusErr := fmt.Errorf("unable to get status")
+	runtimeMock.On("GetPodStatus", pods[1].ID, "", "").Return(&kubecontainer.PodStatus{}, statusErr).Once()
+
+	pleg.relist()
+	actualEvents := getEventsFromChannel(ch)
+	cases := []struct {
+		pod    *kubecontainer.Pod
+		status *kubecontainer.PodStatus
+		error  error
+	}{
+		{pod: pods[0], status: statuses[0], error: nil},
+		{pod: pods[1], status: &kubecontainer.PodStatus{}, error: statusErr},
+	}
+	for i, c := range cases {
+		testStr := fmt.Sprintf("test[%d]", i)
+		actualStatus, actualErr := pleg.cache.Get(c.pod.ID)
+		assert.Equal(t, c.status, actualStatus, testStr)
+		assert.Equal(t, c.error, actualErr, testStr)
+	}
+	// pleg should not generate any event for pods[1] because of the error.
+	assert.Exactly(t, []*PodLifecycleEvent{events[0]}, actualEvents)
+
+	// Return normal status for pods[1].
+	runtimeMock.On("GetPodStatus", pods[1].ID, "", "").Return(statuses[1], nil).Once()
+	pleg.relist()
+	actualEvents = getEventsFromChannel(ch)
+	cases = []struct {
+		pod    *kubecontainer.Pod
+		status *kubecontainer.PodStatus
+		error  error
+	}{
+		{pod: pods[0], status: statuses[0], error: nil},
+		{pod: pods[1], status: statuses[1], error: nil},
+	}
+	for i, c := range cases {
+		testStr := fmt.Sprintf("test[%d]", i)
+		actualStatus, actualErr := pleg.cache.Get(c.pod.ID)
+		assert.Equal(t, c.status, actualStatus, testStr)
+		assert.Equal(t, c.error, actualErr, testStr)
+	}
+	// Now that we are able to query status for pods[1], pleg should generate an event.
+	assert.Exactly(t, []*PodLifecycleEvent{events[1]}, actualEvents)
+}
+
+func TestRemoveCacheEntry(t *testing.T) {
+	pleg, runtimeMock := newTestGenericPLEGWithRuntimeMock()
+	pods, statuses, _ := createTestPodsStatusesAndEvents(1)
+	runtimeMock.On("GetPods", true).Return(pods, nil).Once()
+	runtimeMock.On("GetPodStatus", pods[0].ID, "", "").Return(statuses[0], nil).Once()
+	// Does a relist to populate the cache.
+	pleg.relist()
+	// Delete the pod from runtime. Verify that the cache entry has been
+	// removed after relisting.
+	runtimeMock.On("GetPods", true).Return([]*kubecontainer.Pod{}, nil).Once()
+	pleg.relist()
+	actualStatus, actualErr := pleg.cache.Get(pods[0].ID)
+	assert.Equal(t, &kubecontainer.PodStatus{ID: pods[0].ID}, actualStatus)
+	assert.Equal(t, nil, actualErr)
+}
diff --git a/pkg/kubelet/util/queue/work_queue.go b/pkg/kubelet/util/queue/work_queue.go
index 97ab440b79c..6870c18ace2 100644
--- a/pkg/kubelet/util/queue/work_queue.go
+++ b/pkg/kubelet/util/queue/work_queue.go
@@ -29,8 +29,7 @@ import (
 type WorkQueue interface {
 	// GetWork dequeues and returns all ready items.
 	GetWork() []types.UID
-	// Enqueue inserts a new item or overwrites an existing item with the
-	// new timestamp (time.Now() + delay) if it is greater.
+	// Enqueue inserts a new item or overwrites an existing item.
 	Enqueue(item types.UID, delay time.Duration)
 }
 
@@ -64,10 +63,5 @@ func (q *basicWorkQueue) GetWork() []types.UID {
 func (q *basicWorkQueue) Enqueue(item types.UID, delay time.Duration) {
 	q.lock.Lock()
 	defer q.lock.Unlock()
-	now := q.clock.Now()
-	timestamp := now.Add(delay)
-	existing, ok := q.queue[item]
-	if !ok || (ok && existing.Before(timestamp)) {
-		q.queue[item] = timestamp
-	}
+	q.queue[item] = q.clock.Now().Add(delay)
 }
diff --git a/pkg/kubelet/util/queue/work_queue_test.go b/pkg/kubelet/util/queue/work_queue_test.go
index f105b037321..6dd452a9112 100644
--- a/pkg/kubelet/util/queue/work_queue_test.go
+++ b/pkg/kubelet/util/queue/work_queue_test.go
@@ -63,15 +63,3 @@ func TestGetWork(t *testing.T) {
 	compareResults(t, expected, q.GetWork())
 	compareResults(t, []types.UID{}, q.GetWork())
 }
-
-func TestEnqueueKeepGreaterTimestamp(t *testing.T) {
-	q, _ := newTestBasicWorkQueue()
-	item := types.UID("foo")
-	q.Enqueue(item, -7*time.Hour)
-	q.Enqueue(item, 3*time.Hour)
-	compareResults(t, []types.UID{}, q.GetWork())
-
-	q.Enqueue(item, 3*time.Hour)
-	q.Enqueue(item, -7*time.Hour)
-	compareResults(t, []types.UID{}, q.GetWork())
-}