udpate cadvisor dependency to v0.28.2

vendor/github.com/google/cadvisor/manager/BUILD

@@ -15,6 +15,7 @@ go_library(
         "//vendor/github.com/google/cadvisor/cache/memory:go_default_library",
         "//vendor/github.com/google/cadvisor/collector:go_default_library",
         "//vendor/github.com/google/cadvisor/container:go_default_library",
+        "//vendor/github.com/google/cadvisor/container/containerd:go_default_library",
         "//vendor/github.com/google/cadvisor/container/crio:go_default_library",
         "//vendor/github.com/google/cadvisor/container/docker:go_default_library",
         "//vendor/github.com/google/cadvisor/container/raw:go_default_library",
@@ -34,6 +35,7 @@ go_library(
         "//vendor/github.com/google/cadvisor/utils/sysfs:go_default_library",
         "//vendor/github.com/google/cadvisor/version:go_default_library",
         "//vendor/github.com/opencontainers/runc/libcontainer/cgroups:go_default_library",
+        "//vendor/k8s.io/utils/clock:go_default_library",
     ],
 )
 

vendor/github.com/google/cadvisor/manager/container.go

@@ -40,6 +40,7 @@ import (
 
 	units "github.com/docker/go-units"
 	"github.com/golang/glog"
+	"k8s.io/utils/clock"
 )
 
 // Housekeeping interval.
@@ -65,8 +66,11 @@ type containerData struct {
 	housekeepingInterval     time.Duration
 	maxHousekeepingInterval  time.Duration
 	allowDynamicHousekeeping bool
-	lastUpdatedTime          time.Time
+	infoLastUpdatedTime      time.Time
+	statsLastUpdatedTime     time.Time
 	lastErrorTime            time.Time
+	//  used to track time
+	clock clock.Clock
 
 	// Decay value used for load average smoothing. Interval length of 10 seconds is used.
 	loadDecay float64
@@ -77,6 +81,9 @@ type containerData struct {
 	// Tells the container to stop.
 	stop chan bool
 
+	// Tells the container to immediately collect stats
+	onDemandChan chan chan struct{}
+
 	// Runs custom metric collectors.
 	collectorManager collector.CollectorManager
 
@@ -110,16 +117,43 @@ func (c *containerData) Stop() error {
 }
 
 func (c *containerData) allowErrorLogging() bool {
-	if time.Since(c.lastErrorTime) > time.Minute {
-		c.lastErrorTime = time.Now()
+	if c.clock.Since(c.lastErrorTime) > time.Minute {
+		c.lastErrorTime = c.clock.Now()
 		return true
 	}
 	return false
 }
 
+// OnDemandHousekeeping performs housekeeping on the container and blocks until it has completed.
+// It is designed to be used in conjunction with periodic housekeeping, and will cause the timer for
+// periodic housekeeping to reset.  This should be used sparingly, as calling OnDemandHousekeeping frequently
+// can have serious performance costs.
+func (c *containerData) OnDemandHousekeeping(maxAge time.Duration) {
+	if c.clock.Since(c.statsLastUpdatedTime) > maxAge {
+		housekeepingFinishedChan := make(chan struct{})
+		c.onDemandChan <- housekeepingFinishedChan
+		select {
+		case <-c.stop:
+		case <-housekeepingFinishedChan:
+		}
+	}
+}
+
+// notifyOnDemand notifies all calls to OnDemandHousekeeping that housekeeping is finished
+func (c *containerData) notifyOnDemand() {
+	for {
+		select {
+		case finishedChan := <-c.onDemandChan:
+			close(finishedChan)
+		default:
+			return
+		}
+	}
+}
+
 func (c *containerData) GetInfo(shouldUpdateSubcontainers bool) (*containerInfo, error) {
 	// Get spec and subcontainers.
-	if time.Since(c.lastUpdatedTime) > 5*time.Second {
+	if c.clock.Since(c.infoLastUpdatedTime) > 5*time.Second {
 		err := c.updateSpec()
 		if err != nil {
 			return nil, err
@@ -130,7 +164,7 @@ func (c *containerData) GetInfo(shouldUpdateSubcontainers bool) (*containerInfo,
 				return nil, err
 			}
 		}
-		c.lastUpdatedTime = time.Now()
+		c.infoLastUpdatedTime = c.clock.Now()
 	}
 	// Make a copy of the info for the user.
 	c.lock.Lock()
@@ -310,7 +344,7 @@ func (c *containerData) GetProcessList(cadvisorContainer string, inHostNamespace
 	return processes, nil
 }
 
-func newContainerData(containerName string, memoryCache *memory.InMemoryCache, handler container.ContainerHandler, logUsage bool, collectorManager collector.CollectorManager, maxHousekeepingInterval time.Duration, allowDynamicHousekeeping bool) (*containerData, error) {
+func newContainerData(containerName string, memoryCache *memory.InMemoryCache, handler container.ContainerHandler, logUsage bool, collectorManager collector.CollectorManager, maxHousekeepingInterval time.Duration, allowDynamicHousekeeping bool, clock clock.Clock) (*containerData, error) {
 	if memoryCache == nil {
 		return nil, fmt.Errorf("nil memory storage")
 	}
@@ -332,6 +366,8 @@ func newContainerData(containerName string, memoryCache *memory.InMemoryCache, h
 		loadAvg:                  -1.0, // negative value indicates uninitialized.
 		stop:                     make(chan bool, 1),
 		collectorManager:         collectorManager,
+		onDemandChan:             make(chan chan struct{}, 100),
+		clock:                    clock,
 	}
 	cont.info.ContainerReference = ref
 
@@ -362,7 +398,7 @@ func newContainerData(containerName string, memoryCache *memory.InMemoryCache, h
 }
 
 // Determine when the next housekeeping should occur.
-func (self *containerData) nextHousekeeping(lastHousekeeping time.Time) time.Time {
+func (self *containerData) nextHousekeepingInterval() time.Duration {
 	if self.allowDynamicHousekeeping {
 		var empty time.Time
 		stats, err := self.memoryCache.RecentStats(self.info.Name, empty, empty, 2)
@@ -385,7 +421,7 @@ func (self *containerData) nextHousekeeping(lastHousekeeping time.Time) time.Tim
 		}
 	}
 
-	return lastHousekeeping.Add(jitter(self.housekeepingInterval, 1.0))
+	return jitter(self.housekeepingInterval, 1.0)
 }
 
 // TODO(vmarmol): Implement stats collecting as a custom collector.
@@ -411,24 +447,19 @@ func (c *containerData) housekeeping() {
 
 	// Housekeep every second.
 	glog.V(3).Infof("Start housekeeping for container %q\n", c.info.Name)
-	lastHousekeeping := time.Now()
+	houseKeepingTimer := c.clock.NewTimer(0 * time.Second)
+	defer houseKeepingTimer.Stop()
 	for {
-		select {
-		case <-c.stop:
-			// Stop housekeeping when signaled.
+		if !c.housekeepingTick(houseKeepingTimer.C(), longHousekeeping) {
 			return
-		default:
-			// Perform housekeeping.
-			start := time.Now()
-			c.housekeepingTick()
-
-			// Log if housekeeping took too long.
-			duration := time.Since(start)
-			if duration >= longHousekeeping {
-				glog.V(3).Infof("[%s] Housekeeping took %s", c.info.Name, duration)
+		}
+		// Stop and drain the timer so that it is safe to reset it
+		if !houseKeepingTimer.Stop() {
+			select {
+			case <-houseKeepingTimer.C():
+			default:
 			}
 		}
-
 		// Log usage if asked to do so.
 		if c.logUsage {
 			const numSamples = 60
@@ -455,26 +486,35 @@ func (c *containerData) housekeeping() {
 				glog.Infof("[%s] %.3f cores (average: %.3f cores), %s of memory", c.info.Name, instantUsageInCores, usageInCores, usageInHuman)
 			}
 		}
-
-		next := c.nextHousekeeping(lastHousekeeping)
-
-		// Schedule the next housekeeping. Sleep until that time.
-		if time.Now().Before(next) {
-			time.Sleep(next.Sub(time.Now()))
-		} else {
-			next = time.Now()
-		}
-		lastHousekeeping = next
+		houseKeepingTimer.Reset(c.nextHousekeepingInterval())
 	}
 }
 
-func (c *containerData) housekeepingTick() {
+func (c *containerData) housekeepingTick(timer <-chan time.Time, longHousekeeping time.Duration) bool {
+	select {
+	case <-c.stop:
+		// Stop housekeeping when signaled.
+		return false
+	case finishedChan := <-c.onDemandChan:
+		// notify the calling function once housekeeping has completed
+		defer close(finishedChan)
+	case <-timer:
+	}
+	start := c.clock.Now()
 	err := c.updateStats()
 	if err != nil {
 		if c.allowErrorLogging() {
 			glog.Infof("Failed to update stats for container \"%s\": %s", c.info.Name, err)
 		}
 	}
+	// Log if housekeeping took too long.
+	duration := c.clock.Since(start)
+	if duration >= longHousekeeping {
+		glog.V(3).Infof("[%s] Housekeeping took %s", c.info.Name, duration)
+	}
+	c.notifyOnDemand()
+	c.statsLastUpdatedTime = c.clock.Now()
+	return true
 }
 
 func (c *containerData) updateSpec() error {
@@ -550,7 +590,7 @@ func (c *containerData) updateStats() error {
 	var customStatsErr error
 	cm := c.collectorManager.(*collector.GenericCollectorManager)
 	if len(cm.Collectors) > 0 {
-		if cm.NextCollectionTime.Before(time.Now()) {
+		if cm.NextCollectionTime.Before(c.clock.Now()) {
 			customStats, err := c.updateCustomStats()
 			if customStats != nil {
 				stats.CustomMetrics = customStats

vendor/github.com/google/cadvisor/manager/manager.go

@@ -30,6 +30,7 @@ import (
 	"github.com/google/cadvisor/cache/memory"
 	"github.com/google/cadvisor/collector"
 	"github.com/google/cadvisor/container"
+	"github.com/google/cadvisor/container/containerd"
 	"github.com/google/cadvisor/container/crio"
 	"github.com/google/cadvisor/container/docker"
 	"github.com/google/cadvisor/container/raw"
@@ -49,6 +50,7 @@ import (
 
 	"github.com/golang/glog"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"k8s.io/utils/clock"
 )
 
 var globalHousekeepingInterval = flag.Duration("global_housekeeping_interval", 1*time.Minute, "Interval between global housekeepings")
@@ -279,6 +281,11 @@ func (self *manager) Start() error {
 		self.containerWatchers = append(self.containerWatchers, watcher)
 	}
 
+	err = containerd.Register(self, self.fsInfo, self.ignoreMetrics)
+	if err != nil {
+		glog.Warningf("Registration of the containerd container factory failed: %v", err)
+	}
+
 	err = crio.Register(self, self.fsInfo, self.ignoreMetrics)
 	if err != nil {
 		glog.Warningf("Registration of the crio container factory failed: %v", err)
@@ -707,6 +714,18 @@ func (self *manager) getRequestedContainers(containerName string, options v2.Req
 	default:
 		return containersMap, fmt.Errorf("invalid request type %q", options.IdType)
 	}
+	if options.MaxAge != nil {
+		// update stats for all containers in containersMap
+		var waitGroup sync.WaitGroup
+		waitGroup.Add(len(containersMap))
+		for _, container := range containersMap {
+			go func(cont *containerData) {
+				cont.OnDemandHousekeeping(*options.MaxAge)
+				waitGroup.Done()
+			}(container)
+		}
+		waitGroup.Wait()
+	}
 	return containersMap, nil
 }
 
@@ -804,6 +823,8 @@ func (m *manager) Exists(containerName string) bool {
 func (m *manager) GetProcessList(containerName string, options v2.RequestOptions) ([]v2.ProcessInfo, error) {
 	// override recursive. Only support single container listing.
 	options.Recursive = false
+	// override MaxAge.  ProcessList does not require updated stats.
+	options.MaxAge = nil
 	conts, err := m.getRequestedContainers(containerName, options)
 	if err != nil {
 		return nil, err
@@ -919,7 +940,7 @@ func (m *manager) createContainerLocked(containerName string, watchSource watche
 	}
 
 	logUsage := *logCadvisorUsage && containerName == m.cadvisorContainer
-	cont, err := newContainerData(containerName, m.memoryCache, handler, logUsage, collectorManager, m.maxHousekeepingInterval, m.allowDynamicHousekeeping)
+	cont, err := newContainerData(containerName, m.memoryCache, handler, logUsage, collectorManager, m.maxHousekeepingInterval, m.allowDynamicHousekeeping, clock.RealClock{})
 	if err != nil {
 		return err
 	}

vendor/github.com/google/cadvisor/manager/watcher/raw/BUILD

@@ -6,11 +6,11 @@ go_library(
     importpath = "github.com/google/cadvisor/manager/watcher/raw",
     visibility = ["//visibility:public"],
     deps = [
-        "//vendor/github.com/fsnotify/fsnotify:go_default_library",
         "//vendor/github.com/golang/glog:go_default_library",
         "//vendor/github.com/google/cadvisor/container/common:go_default_library",
         "//vendor/github.com/google/cadvisor/container/libcontainer:go_default_library",
         "//vendor/github.com/google/cadvisor/manager/watcher:go_default_library",
+        "//vendor/golang.org/x/exp/inotify:go_default_library",
     ],
 )
 

vendor/github.com/google/cadvisor/manager/watcher/raw/raw.go

@@ -27,8 +27,8 @@ import (
 	"github.com/google/cadvisor/container/libcontainer"
 	"github.com/google/cadvisor/manager/watcher"
 
-	"github.com/fsnotify/fsnotify"
 	"github.com/golang/glog"
+	"golang.org/x/exp/inotify"
 )
 
 type rawContainerWatcher struct {
@@ -121,7 +121,7 @@ func (self *rawContainerWatcher) watchDirectory(dir string, containerName string
 		if cleanup {
 			_, err := self.watcher.RemoveWatch(containerName, dir)
 			if err != nil {
-				glog.Warningf("Failed to remove fsnotify watch for %q: %v", dir, err)
+				glog.Warningf("Failed to remove inotify watch for %q: %v", dir, err)
 			}
 		}
 	}()
@@ -152,16 +152,18 @@ func (self *rawContainerWatcher) watchDirectory(dir string, containerName string
 	return alreadyWatching, nil
 }
 
-func (self *rawContainerWatcher) processEvent(event fsnotify.Event, events chan watcher.ContainerEvent) error {
-	// Convert the fsnotify event type to a container create or delete.
+func (self *rawContainerWatcher) processEvent(event *inotify.Event, events chan watcher.ContainerEvent) error {
+	// Convert the inotify event type to a container create or delete.
 	var eventType watcher.ContainerEventType
 	switch {
-	case event.Op == fsnotify.Create:
+	case (event.Mask & inotify.IN_CREATE) > 0:
 		eventType = watcher.ContainerAdd
-	case event.Op == fsnotify.Remove:
+	case (event.Mask & inotify.IN_DELETE) > 0:
 		eventType = watcher.ContainerDelete
-	case event.Op == fsnotify.Rename:
+	case (event.Mask & inotify.IN_MOVED_FROM) > 0:
 		eventType = watcher.ContainerDelete
+	case (event.Mask & inotify.IN_MOVED_TO) > 0:
+		eventType = watcher.ContainerAdd
 	default:
 		// Ignore other events.
 		return nil