Update cAdvisor.

Also update golang.org/x/sys because of google/cadvisor#1786

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

@@ -11,6 +11,7 @@ go_library(
     deps = [
         "//vendor/github.com/docker/go-units:go_default_library",
         "//vendor/github.com/golang/glog:go_default_library",
+        "//vendor/github.com/google/cadvisor/accelerators:go_default_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/manager/container.go

@@ -29,6 +29,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/google/cadvisor/accelerators"
 	"github.com/google/cadvisor/cache/memory"
 	"github.com/google/cadvisor/collector"
 	"github.com/google/cadvisor/container"
@@ -78,6 +79,9 @@ type containerData struct {
 
 	// Runs custom metric collectors.
 	collectorManager collector.CollectorManager
+
+	// nvidiaCollector updates stats for Nvidia GPUs attached to the container.
+	nvidiaCollector accelerators.AcceleratorCollector
 }
 
 // jitter returns a time.Duration between duration and duration + maxFactor * duration,
@@ -557,6 +561,12 @@ func (c *containerData) updateStats() error {
 		}
 	}
 
+	var nvidiaStatsErr error
+	if c.nvidiaCollector != nil {
+		// This updates the Accelerators field of the stats struct
+		nvidiaStatsErr = c.nvidiaCollector.UpdateStats(stats)
+	}
+
 	ref, err := c.handler.ContainerReference()
 	if err != nil {
 		// Ignore errors if the container is dead.
@@ -572,6 +582,9 @@ func (c *containerData) updateStats() error {
 	if statsErr != nil {
 		return statsErr
 	}
+	if nvidiaStatsErr != nil {
+		return nvidiaStatsErr
+	}
 	return customStatsErr
 }
 

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

@@ -18,6 +18,7 @@ package manager
 import (
 	"flag"
 	"fmt"
+	"net/http"
 	"os"
 	"path"
 	"strconv"
@@ -25,6 +26,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/google/cadvisor/accelerators"
 	"github.com/google/cadvisor/cache/memory"
 	"github.com/google/cadvisor/collector"
 	"github.com/google/cadvisor/container"
@@ -45,8 +47,6 @@ import (
 	"github.com/google/cadvisor/utils/sysfs"
 	"github.com/google/cadvisor/version"
 
-	"net/http"
-
 	"github.com/golang/glog"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
 )
@@ -148,13 +148,19 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 	}
 	glog.Infof("cAdvisor running in container: %q", selfContainer)
 
-	dockerStatus, err := docker.Status()
-	if err != nil {
+	var (
+		dockerStatus info.DockerStatus
+		rktPath      string
+	)
+	if tempDockerStatus, err := docker.Status(); err != nil {
 		glog.Warningf("Unable to connect to Docker: %v", err)
+	} else {
+		dockerStatus = tempDockerStatus
 	}
-	rktPath, err := rkt.RktPath()
-	if err != nil {
+	if tmpRktPath, err := rkt.RktPath(); err != nil {
 		glog.Warningf("unable to connect to Rkt api service: %v", err)
+	} else {
+		rktPath = tmpRktPath
 	}
 
 	crioClient, err := crio.Client()
@@ -206,6 +212,7 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 		containerWatchers:        []watcher.ContainerWatcher{},
 		eventsChannel:            eventsChannel,
 		collectorHttpClient:      collectorHttpClient,
+		nvidiaManager:            &accelerators.NvidiaManager{},
 	}
 
 	machineInfo, err := machine.Info(sysfs, fsInfo, inHostNamespace)
@@ -251,6 +258,7 @@ type manager struct {
 	containerWatchers        []watcher.ContainerWatcher
 	eventsChannel            chan watcher.ContainerEvent
 	collectorHttpClient      *http.Client
+	nvidiaManager            accelerators.AcceleratorManager
 }
 
 // Start the container manager.
@@ -303,6 +311,9 @@ func (self *manager) Start() error {
 		return nil
 	}
 
+	// Setup collection of nvidia GPU metrics if any of them are attached to the machine.
+	self.nvidiaManager.Setup()
+
 	// Create root and then recover all containers.
 	err = self.createContainer("/", watcher.Raw)
 	if err != nil {
@@ -332,6 +343,7 @@ func (self *manager) Start() error {
 }
 
 func (self *manager) Stop() error {
+	defer self.nvidiaManager.Destroy()
 	// Stop and wait on all quit channels.
 	for i, c := range self.quitChannels {
 		// Send the exit signal and wait on the thread to exit (by closing the channel).
@@ -911,6 +923,15 @@ func (m *manager) createContainerLocked(containerName string, watchSource watche
 	if err != nil {
 		return err
 	}
+	devicesCgroupPath, err := handler.GetCgroupPath("devices")
+	if err != nil {
+		glog.Infof("Error getting devices cgroup path: %v", err)
+	} else {
+		cont.nvidiaCollector, err = m.nvidiaManager.GetCollector(devicesCgroupPath)
+		if err != nil {
+			glog.Infof("GPU metrics may be unavailable/incomplete for container %q: %v", cont.info.Name, err)
+		}
+	}
 
 	// Add collectors
 	labels := handler.GetContainerLabels()

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 inotify watch for %q: %v", dir, err)
+				glog.Warningf("Failed to remove fsnotify watch for %q: %v", dir, err)
 			}
 		}
 	}()
@@ -152,18 +152,16 @@ func (self *rawContainerWatcher) watchDirectory(dir string, containerName string
 	return alreadyWatching, nil
 }
 
-func (self *rawContainerWatcher) processEvent(event *inotify.Event, events chan watcher.ContainerEvent) error {
-	// Convert the inotify event type to a container create or delete.
+func (self *rawContainerWatcher) processEvent(event fsnotify.Event, events chan watcher.ContainerEvent) error {
+	// Convert the fsnotify event type to a container create or delete.
 	var eventType watcher.ContainerEventType
 	switch {
-	case (event.Mask & inotify.IN_CREATE) > 0:
+	case event.Op == fsnotify.Create:
 		eventType = watcher.ContainerAdd
-	case (event.Mask & inotify.IN_DELETE) > 0:
+	case event.Op == fsnotify.Remove:
 		eventType = watcher.ContainerDelete
-	case (event.Mask & inotify.IN_MOVED_FROM) > 0:
+	case event.Op == fsnotify.Rename:
 		eventType = watcher.ContainerDelete
-	case (event.Mask & inotify.IN_MOVED_TO) > 0:
-		eventType = watcher.ContainerAdd
 	default:
 		// Ignore other events.
 		return nil