Merge pull request #59595 from Random-Liu/update-cadvisor

vendor/BUILD

@@ -273,7 +273,6 @@ filegroup(
         "//vendor/github.com/jonboulle/clockwork:all-srcs",
         "//vendor/github.com/json-iterator/go:all-srcs",
         "//vendor/github.com/jteeuwen/go-bindata:all-srcs",
-        "//vendor/github.com/juju/ratelimit:all-srcs",
         "//vendor/github.com/kardianos/osext:all-srcs",
         "//vendor/github.com/kr/fs:all-srcs",
         "//vendor/github.com/kr/pretty:all-srcs",

vendor/github.com/google/cadvisor/container/containerd/client.go

@@ -16,6 +16,7 @@ package containerd
 
 import (
 	"context"
+	"sync"
 	"time"
 
 	containersapi "github.com/containerd/containerd/api/services/containers/v1"
@@ -45,32 +46,38 @@ type containerdClient interface {
 	Version(ctx context.Context) (string, error)
 }
 
+var once sync.Once
+var ctrdClient containerdClient = nil
+
 // Client creates a containerd client
 func Client() (containerdClient, error) {
-	gopts := []grpc.DialOption{
-		grpc.WithInsecure(),
-		grpc.FailOnNonTempDialError(true),
-		grpc.WithDialer(dialer.Dialer),
-		grpc.WithBlock(),
-		grpc.WithTimeout(2 * time.Second),
-		grpc.WithBackoffMaxDelay(3 * time.Second),
-	}
-	unary, stream := newNSInterceptors(k8sNamespace)
-	gopts = append(gopts,
-		grpc.WithUnaryInterceptor(unary),
-		grpc.WithStreamInterceptor(stream),
-	)
+	var retErr error
+	once.Do(func() {
+		gopts := []grpc.DialOption{
+			grpc.WithInsecure(),
+			grpc.WithDialer(dialer.Dialer),
+			grpc.WithBlock(),
+			grpc.WithTimeout(2 * time.Second),
+			grpc.WithBackoffMaxDelay(3 * time.Second),
+		}
+		unary, stream := newNSInterceptors(k8sNamespace)
+		gopts = append(gopts,
+			grpc.WithUnaryInterceptor(unary),
+			grpc.WithStreamInterceptor(stream),
+		)
 
-	conn, err := grpc.Dial(dialer.DialAddress("/var/run/containerd/containerd.sock"), gopts...)
-	if err != nil {
-		return nil, err
-	}
-	c := &client{
-		containerService: containersapi.NewContainersClient(conn),
-		taskService:      tasksapi.NewTasksClient(conn),
-		versionService:   versionapi.NewVersionClient(conn),
-	}
-	return c, err
+		conn, err := grpc.Dial(dialer.DialAddress("/var/run/containerd/containerd.sock"), gopts...)
+		if err != nil {
+			retErr = err
+			return
+		}
+		ctrdClient = &client{
+			containerService: containersapi.NewContainersClient(conn),
+			taskService:      tasksapi.NewTasksClient(conn),
+			versionService:   versionapi.NewVersionClient(conn),
+		}
+	})
+	return ctrdClient, retErr
 }
 
 func (c *client) LoadContainer(ctx context.Context, id string) (*containers.Container, error) {

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

@@ -29,19 +29,27 @@ import (
 	"github.com/google/cadvisor/machine"
 )
 
-const defaultTimeout = time.Second * 5
+var dockerTimeout = 10 * time.Second
 
 func defaultContext() context.Context {
-	ctx, _ := context.WithTimeout(context.Background(), defaultTimeout)
+	ctx, _ := context.WithTimeout(context.Background(), dockerTimeout)
 	return ctx
 }
 
+func SetTimeout(timeout time.Duration) {
+	dockerTimeout = timeout
+}
+
 func Status() (v1.DockerStatus, error) {
+	return StatusWithContext(defaultContext())
+}
+
+func StatusWithContext(ctx context.Context) (v1.DockerStatus, error) {
 	client, err := Client()
 	if err != nil {
 		return v1.DockerStatus{}, fmt.Errorf("unable to communicate with docker daemon: %v", err)
 	}
-	dockerInfo, err := client.Info(defaultContext())
+	dockerInfo, err := client.Info(ctx)
 	if err != nil {
 		return v1.DockerStatus{}, err
 	}

vendor/github.com/google/cadvisor/container/docker/handler.go

@@ -44,6 +44,7 @@ import (
 const (
 	// The read write layers exist here.
 	aufsRWLayer     = "diff"
+	overlayRWLayer  = "upper"
 	overlay2RWLayer = "diff"
 
 	// Path to the directory where docker stores log files if the json logging driver is enabled.
@@ -197,7 +198,7 @@ func newDockerContainerHandler(
 	case aufsStorageDriver:
 		rootfsStorageDir = path.Join(storageDir, string(aufsStorageDriver), aufsRWLayer, rwLayerID)
 	case overlayStorageDriver:
-		rootfsStorageDir = path.Join(storageDir, string(storageDriver), rwLayerID)
+		rootfsStorageDir = path.Join(storageDir, string(storageDriver), rwLayerID, overlayRWLayer)
 	case overlay2StorageDriver:
 		rootfsStorageDir = path.Join(storageDir, string(storageDriver), rwLayerID, overlay2RWLayer)
 	case zfsStorageDriver:

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

@@ -114,7 +114,9 @@ func NewFsInfo(context Context) (FsInfo, error) {
 
 	fsUUIDToDeviceName, err := getFsUUIDToDeviceNameMap()
 	if err != nil {
-		return nil, err
+		// UUID is not always avaiable across different OS distributions.
+		// Do not fail if there is an error.
+		glog.Warningf("Failed to get disk UUID mapping, getting disk info by uuid will not work: %v", err)
 	}
 
 	// Avoid devicemapper container mounts - these are tracked by the ThinPoolWatcher

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

@@ -35,6 +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/golang.org/x/net/context:go_default_library",
         "//vendor/k8s.io/utils/clock:go_default_library",
     ],
 )

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

@@ -504,7 +504,7 @@ func (c *containerData) housekeepingTick(timer <-chan time.Time, longHousekeepin
 	err := c.updateStats()
 	if err != nil {
 		if c.allowErrorLogging() {
-			glog.Warning("Failed to update stats for container \"%s\": %s", c.info.Name, err)
+			glog.Warningf("Failed to update stats for container \"%s\": %s", c.info.Name, err)
 		}
 	}
 	// Log if housekeeping took too long.

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

@@ -50,6 +50,7 @@ import (
 
 	"github.com/golang/glog"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"golang.org/x/net/context"
 	"k8s.io/utils/clock"
 )
 
@@ -59,6 +60,8 @@ var eventStorageAgeLimit = flag.String("event_storage_age_limit", "default=24h",
 var eventStorageEventLimit = flag.String("event_storage_event_limit", "default=100000", "Max number of events to store (per type). Value is a comma separated list of key values, where the keys are event types (e.g.: creation, oom) or \"default\" and the value is an integer. Default is applied to all non-specified event types")
 var applicationMetricsCountLimit = flag.Int("application_metrics_count_limit", 100, "Max number of application metrics to store (per container)")
 
+const dockerClientTimeout = 10 * time.Second
+
 // The Manager interface defines operations for starting a manager and getting
 // container and machine information.
 type Manager interface {
@@ -154,11 +157,10 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 		dockerStatus info.DockerStatus
 		rktPath      string
 	)
-	if tempDockerStatus, err := docker.Status(); err != nil {
-		glog.V(5).Infof("Docker not connected: %v", err)
-	} else {
-		dockerStatus = tempDockerStatus
-	}
+	docker.SetTimeout(dockerClientTimeout)
+	// Try to connect to docker indefinitely on startup.
+	dockerStatus = retryDockerStatus()
+
 	if tmpRktPath, err := rkt.RktPath(); err != nil {
 		glog.V(5).Infof("Rkt not connected: %v", err)
 	} else {
@@ -234,6 +236,31 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 	return newManager, nil
 }
 
+func retryDockerStatus() info.DockerStatus {
+	startupTimeout := dockerClientTimeout
+	maxTimeout := 4 * startupTimeout
+	for {
+		ctx, _ := context.WithTimeout(context.Background(), startupTimeout)
+		dockerStatus, err := docker.StatusWithContext(ctx)
+		if err != nil {
+			return dockerStatus
+		}
+
+		switch err {
+		case context.DeadlineExceeded:
+			glog.Warningf("Timeout trying to communicate with docker during initialization, will retry")
+		default:
+			glog.V(5).Infof("Docker not connected: %v", err)
+			return info.DockerStatus{}
+		}
+
+		startupTimeout = 2 * startupTimeout
+		if startupTimeout > maxTimeout {
+			startupTimeout = maxTimeout
+		}
+	}
+}
+
 // A namespaced container name.
 type namespacedContainerName struct {
 	// The namespace of the container. Can be empty for the root namespace.

vendor/github.com/juju/ratelimit/BUILD

-load("@io_bazel_rules_go//go:def.bzl", "go_library")
-
-go_library(
-    name = "go_default_library",
-    srcs = [
-        "ratelimit.go",
-        "reader.go",
-    ],
-    importpath = "github.com/juju/ratelimit",
-    visibility = ["//visibility:public"],
-)
-
-filegroup(
-    name = "package-srcs",
-    srcs = glob(["**"]),
-    tags = ["automanaged"],
-    visibility = ["//visibility:private"],
-)
-
-filegroup(
-    name = "all-srcs",
-    srcs = [":package-srcs"],
-    tags = ["automanaged"],
-    visibility = ["//visibility:public"],
-)