Revert "Fix the race between configuring cbr0 and restarting static pods"

cluster/saltbase/salt/docker/docker-defaults

@@ -2,5 +2,5 @@ DOCKER_OPTS=""
 {% if grains.docker_opts is defined and grains.docker_opts %}
 DOCKER_OPTS="${DOCKER_OPTS} {{grains.docker_opts}}"
 {% endif %}
-DOCKER_OPTS="${DOCKER_OPTS} --bridge=cbr0 --iptables=false --ip-masq=false"
+DOCKER_OPTS="${DOCKER_OPTS} --bridge cbr0 --iptables=false --ip-masq=false"
 DOCKER_NOFILE=1000000

cluster/saltbase/salt/docker/init.sls

@@ -48,6 +48,11 @@ net.ipv4.ip_forward:
   sysctl.present:
     - value: 1
 
+cbr0:
+  container_bridge.ensure:
+    - cidr: {{ grains['cbr-cidr'] }}
+    - mtu: 1460
+
 {{ environment_file }}:
   file.managed:
     - source: salt://docker/docker-defaults
@@ -119,6 +124,7 @@ docker:
     - enable: True
     - watch:
       - file: {{ environment_file }}
+      - container_bridge: cbr0
 {% if override_docker_ver != '' %}
     - require:
       - pkg: lxc-docker-{{ override_docker_ver }}

cluster/saltbase/salt/kube-master-addons/init.sls

@@ -37,26 +37,10 @@ master-docker-image-tags:
   file.touch:
     - name: /srv/pillar/docker-images.sls
 
-# Current containervm image by default has both docker and kubelet
-# running. But during cluster creation stage, docker and kubelet
-# could be overwritten completely, or restarted due to flag changes.
-# The ordering of salt states for service docker, kubelet and
-# master-addon below is very important to avoid the race between
-# salt restart docker or kubelet and kubelet start master components.
-# Without the ordering of salt states, when gce instance boot up,
-# configure-vm.sh will run and download the release. At the end of
-# boot, run-salt will run kube-master-addons service which installs
-# master component manifest files to kubelet config directory before
-# the installation of proper version kubelet. Please see
-# https://github.com/GoogleCloudPlatform/kubernetes/issues/10122#issuecomment-114566063
-# for detail explanation on this very issue.
 kube-master-addons:
   service.running:
     - enable: True
     - restart: True
-    - require:
-      - service: docker
-      - service: kubelet
     - watch:
       - file: master-docker-image-tags
       - file: /etc/kubernetes/kube-master-addons.sh

cluster/saltbase/salt/kubelet/default

@@ -76,9 +76,4 @@
   {% set cgroup_root = "--cgroup_root=/" -%}
 {% endif -%}
 
-{% set pod_cidr = "" %}
-{% if grains['roles'][0] == 'kubernetes-master' %}
-  {% set pod_cidr = "--pod-cidr=" + grains['cbr-cidr'] %}
-{% endif %} 
-
-DAEMON_ARGS="{{daemon_args}} {{api_servers_with_port}} {{debugging_handlers}} {{hostname_override}} {{cloud_provider}} {{config}} --allow_privileged={{pillar['allow_privileged']}} {{pillar['log_level']}} {{cluster_dns}} {{cluster_domain}} {{docker_root}} {{kubelet_root}} {{configure_cbr0}} {{cgroup_root}} {{system_container}} {{pod_cidr}}"
+DAEMON_ARGS="{{daemon_args}} {{api_servers_with_port}} {{debugging_handlers}} {{hostname_override}} {{cloud_provider}} {{config}} --allow_privileged={{pillar['allow_privileged']}} {{pillar['log_level']}} {{cluster_dns}} {{cluster_domain}} {{docker_root}} {{kubelet_root}} {{configure_cbr0}} {{cgroup_root}} {{system_container}}"

cmd/kubelet/app/server.go

@@ -115,7 +115,6 @@ type KubeletServer struct {
 	DockerDaemonContainer          string
 	SystemContainer                string
 	ConfigureCBR0                  bool
-	PodCIDR                        string
 	MaxPods                        int
 	DockerExecHandlerName          string
 
@@ -242,7 +241,7 @@ func (s *KubeletServer) AddFlags(fs *pflag.FlagSet) {
 	fs.BoolVar(&s.ConfigureCBR0, "configure-cbr0", s.ConfigureCBR0, "If true, kubelet will configure cbr0 based on Node.Spec.PodCIDR.")
 	fs.IntVar(&s.MaxPods, "max-pods", 100, "Number of Pods that can run on this Kubelet.")
 	fs.StringVar(&s.DockerExecHandlerName, "docker-exec-handler", s.DockerExecHandlerName, "Handler to use when executing a command in a container. Valid values are 'native' and 'nsenter'. Defaults to 'native'.")
-	fs.StringVar(&s.PodCIDR, "pod-cidr", "", "The CIDR to use for pod IP addresses, only used in standalone mode.  In cluster mode, this is obtained from the master.")
+
 	// Flags intended for testing, not recommended used in production environments.
 	fs.BoolVar(&s.ReallyCrashForTesting, "really-crash-for-testing", s.ReallyCrashForTesting, "If true, when panics occur crash. Intended for testing.")
 	fs.Float64Var(&s.ChaosChance, "chaos-chance", s.ChaosChance, "If > 0.0, introduce random client errors and latency. Intended for testing. [default=0.0]")
@@ -362,7 +361,6 @@ func (s *KubeletServer) Run(_ []string) error {
 		DockerDaemonContainer:     s.DockerDaemonContainer,
 		SystemContainer:           s.SystemContainer,
 		ConfigureCBR0:             s.ConfigureCBR0,
-		PodCIDR:                   s.PodCIDR,
 		MaxPods:                   s.MaxPods,
 		DockerExecHandler:         dockerExecHandler,
 	}
@@ -716,7 +714,6 @@ type KubeletConfig struct {
 	DockerDaemonContainer          string
 	SystemContainer                string
 	ConfigureCBR0                  bool
-	PodCIDR                        string
 	MaxPods                        int
 	DockerExecHandler              dockertools.ExecHandler
 }
@@ -774,7 +771,6 @@ func createAndInitKubelet(kc *KubeletConfig) (k KubeletBootstrap, pc *config.Pod
 		kc.DockerDaemonContainer,
 		kc.SystemContainer,
 		kc.ConfigureCBR0,
-		kc.PodCIDR,
 		kc.MaxPods,
 		kc.DockerExecHandler)
 

contrib/mesos/pkg/executor/service/service.go

@@ -354,7 +354,6 @@ func (ks *KubeletExecutorServer) createAndInitKubelet(
 		kc.DockerDaemonContainer,
 		kc.SystemContainer,
 		kc.ConfigureCBR0,
-		kc.PodCIDR,
 		kc.MaxPods,
 		kc.DockerExecHandler,
 	)

pkg/kubelet/container_bridge.go

@@ -19,17 +19,27 @@ package kubelet
 import (
 	"bytes"
 	"net"
-	"os"
 	"os/exec"
 	"regexp"
 
-	"github.com/GoogleCloudPlatform/kubernetes/pkg/util"
 	"github.com/golang/glog"
 )
 
 var cidrRegexp = regexp.MustCompile(`inet ([0-9a-fA-F.:]*/[0-9]*)`)
 
-func createCBR0(wantCIDR *net.IPNet) error {
+func ensureCbr0(wantCIDR *net.IPNet) error {
+	if !cbr0CidrCorrect(wantCIDR) {
+		glog.V(2).Infof("Attempting to recreate cbr0 with address range: %s", wantCIDR)
+
+		// delete cbr0
+		if err := exec.Command("ip", "link", "set", "dev", "cbr0", "down").Run(); err != nil {
+			glog.Error(err)
+			return err
+		}
+		if err := exec.Command("brctl", "delbr", "cbr0").Run(); err != nil {
+			glog.Error(err)
+			return err
+		}
 		// recreate cbr0 with wantCIDR
 		if err := exec.Command("brctl", "addbr", "cbr0").Run(); err != nil {
 			glog.Error(err)
@@ -44,60 +54,16 @@ func createCBR0(wantCIDR *net.IPNet) error {
 			return err
 		}
 		// restart docker
-	// For now just log the error. The containerRuntime check will catch docker failures.
-	// TODO (dawnchen) figure out what we should do for rkt here.
-	if util.UsingSystemdInitSystem() {
-		if err := exec.Command("systemctl", "restart", "docker").Run(); err != nil {
-			glog.Error(err)
-		}
-	} else {
 		if err := exec.Command("service", "docker", "restart").Run(); err != nil {
 			glog.Error(err)
-		}
+			// For now just log the error. The containerRuntime check will catch docker failures.
+			// TODO (dawnchen) figure out what we should do for rkt here.
 		}
 		glog.V(2).Info("Recreated cbr0 and restarted docker")
-	return nil
-}
-
-func ensureCbr0(wantCIDR *net.IPNet) error {
-	exists, err := cbr0Exists()
-	if err != nil {
-		return err
-	}
-	if !exists {
-		glog.V(2).Infof("CBR0 doesn't exist, attempting to create it with range: %s", wantCIDR)
-		return createCBR0(wantCIDR)
-	}
-	if !cbr0CidrCorrect(wantCIDR) {
-		glog.V(2).Infof("Attempting to recreate cbr0 with address range: %s", wantCIDR)
-
-		// delete cbr0
-		if err := exec.Command("ip", "link", "set", "dev", "cbr0", "down").Run(); err != nil {
-			glog.Error(err)
-			return err
-		}
-		if err := exec.Command("brctl", "delbr", "cbr0").Run(); err != nil {
-			glog.Error(err)
-			return err
-		}
-		return createCBR0(wantCIDR)
 	}
 	return nil
 }
 
-// Check if cbr0 network interface is configured or not, and take action
-// when the configuration is missing on the node, and propagate the rest
-// error to kubelet to handle.
-func cbr0Exists() (bool, error) {
-	if _, err := os.Stat("/sys/class/net/cbr0"); err != nil {
-		if os.IsNotExist(err) {
-			return false, nil
-		}
-		return false, err
-	}
-	return true, nil
-}
-
 func cbr0CidrCorrect(wantCIDR *net.IPNet) bool {
 	output, err := exec.Command("ip", "addr", "show", "cbr0").Output()
 	if err != nil {
@@ -113,7 +79,6 @@ func cbr0CidrCorrect(wantCIDR *net.IPNet) bool {
 		return false
 	}
 	cbr0CIDR.IP = cbr0IP
-
 	glog.V(5).Infof("Want cbr0 CIDR: %s, have cbr0 CIDR: %s", wantCIDR, cbr0CIDR)
 	return wantCIDR.IP.Equal(cbr0IP) && bytes.Equal(wantCIDR.Mask, cbr0CIDR.Mask)
 }

pkg/kubelet/kubelet.go

@@ -147,7 +147,6 @@ func NewMainKubelet(
 	dockerDaemonContainer string,
 	systemContainer string,
 	configureCBR0 bool,
-	podCIDR string,
 	pods int,
 	dockerExecHandler dockertools.ExecHandler) (*Kubelet, error) {
 	if rootDirectory == "" {
@@ -262,7 +261,6 @@ func NewMainKubelet(
 		cgroupRoot:                     cgroupRoot,
 		mounter:                        mounter,
 		configureCBR0:                  configureCBR0,
-		podCIDR:                        podCIDR,
 		pods:                           pods,
 		syncLoopMonitor:                util.AtomicValue{},
 	}
@@ -320,10 +318,6 @@ func NewMainKubelet(
 	}
 	klet.containerManager = containerManager
 
-	// Start syncing node status immediately, this may set up things the runtime needs to run.
-	go util.Until(klet.syncNetworkStatus, 30*time.Second, util.NeverStop)
-	go klet.syncNodeStatus()
-
 	// Wait for the runtime to be up with a timeout.
 	if err := waitUntilRuntimeIsUp(klet.containerRuntime, maxWaitForContainerRuntime); err != nil {
 		return nil, fmt.Errorf("timed out waiting for %q to come up: %v", containerRuntime, err)
@@ -418,10 +412,6 @@ type Kubelet struct {
 	runtimeUpThreshold     time.Duration
 	lastTimestampRuntimeUp time.Time
 
-	// Network Status information
-	networkConfigMutex sync.Mutex
-	networkConfigured  bool
-
 	// Volume plugins.
 	volumePluginMgr volume.VolumePluginMgr
 
@@ -499,7 +489,6 @@ type Kubelet struct {
 	// Whether or not kubelet should take responsibility for keeping cbr0 in
 	// the correct state.
 	configureCBR0 bool
-	podCIDR       string
 
 	// Number of Pods which can be run by this Kubelet
 	pods int
@@ -718,7 +707,7 @@ func (kl *Kubelet) Run(updates <-chan PodUpdate) {
 	}
 
 	go util.Until(kl.updateRuntimeUp, 5*time.Second, util.NeverStop)
-
+	go kl.syncNodeStatus()
 	// Run the system oom watcher forever.
 	kl.statusManager.Start()
 	kl.syncLoop(updates, kl)
@@ -1716,11 +1705,6 @@ func (kl *Kubelet) syncLoopIteration(updates <-chan PodUpdate, handler SyncHandl
 		glog.Infof("Skipping pod synchronization, container runtime is not up.")
 		return
 	}
-	if !kl.doneNetworkConfigure() {
-		time.Sleep(5 * time.Second)
-		glog.Infof("Skipping pod synchronization, network is not configured")
-		return
-	}
 	unsyncedPod := false
 	podSyncTypes := make(map[types.UID]SyncPodType)
 	select {
@@ -1877,7 +1861,6 @@ func (kl *Kubelet) reconcileCBR0(podCIDR string) error {
 		glog.V(5).Info("PodCIDR not set. Will not configure cbr0.")
 		return nil
 	}
-	glog.V(5).Infof("PodCIDR is set to %q", podCIDR)
 	_, cidr, err := net.ParseCIDR(podCIDR)
 	if err != nil {
 		return err
@@ -1912,22 +1895,6 @@ func (kl *Kubelet) recordNodeStatusEvent(event string) {
 // Maintains Node.Spec.Unschedulable value from previous run of tryUpdateNodeStatus()
 var oldNodeUnschedulable bool
 
-func (kl *Kubelet) syncNetworkStatus() {
-	kl.networkConfigMutex.Lock()
-	defer kl.networkConfigMutex.Unlock()
-
-	networkConfigured := true
-	if kl.configureCBR0 {
-		if len(kl.podCIDR) == 0 {
-			networkConfigured = false
-		} else if err := kl.reconcileCBR0(kl.podCIDR); err != nil {
-			networkConfigured = false
-			glog.Errorf("Error configuring cbr0: %v", err)
-		}
-	}
-	kl.networkConfigured = networkConfigured
-}
-
 // setNodeStatus fills in the Status fields of the given Node, overwriting
 // any fields that are currently set.
 func (kl *Kubelet) setNodeStatus(node *api.Node) error {
@@ -1961,6 +1928,16 @@ func (kl *Kubelet) setNodeStatus(node *api.Node) error {
 		}
 	}
 
+	networkConfigured := true
+	if kl.configureCBR0 {
+		if len(node.Spec.PodCIDR) == 0 {
+			networkConfigured = false
+		} else if err := kl.reconcileCBR0(node.Spec.PodCIDR); err != nil {
+			networkConfigured = false
+			glog.Errorf("Error configuring cbr0: %v", err)
+		}
+	}
+
 	// TODO: Post NotReady if we cannot get MachineInfo from cAdvisor. This needs to start
 	// cAdvisor locally, e.g. for test-cmd.sh, and in integration test.
 	info, err := kl.GetCachedMachineInfo()
@@ -2004,8 +1981,6 @@ func (kl *Kubelet) setNodeStatus(node *api.Node) error {
 
 	// Check whether container runtime can be reported as up.
 	containerRuntimeUp := kl.containerRuntimeUp()
-	// Check whether network is configured properly
-	networkConfigured := kl.doneNetworkConfigure()
 
 	currentTime := util.Now()
 	var newNodeReadyCondition api.NodeCondition
@@ -2074,12 +2049,6 @@ func (kl *Kubelet) containerRuntimeUp() bool {
 	return kl.lastTimestampRuntimeUp.Add(kl.runtimeUpThreshold).After(time.Now())
 }
 
-func (kl *Kubelet) doneNetworkConfigure() bool {
-	kl.networkConfigMutex.Lock()
-	defer kl.networkConfigMutex.Unlock()
-	return kl.networkConfigured
-}
-
 // tryUpdateNodeStatus tries to update node status to master. If ReconcileCBR0
 // is set, this function will also confirm that cbr0 is configured correctly.
 func (kl *Kubelet) tryUpdateNodeStatus() error {
@@ -2090,8 +2059,6 @@ func (kl *Kubelet) tryUpdateNodeStatus() error {
 	if node == nil {
 		return fmt.Errorf("no node instance returned for %q", kl.nodeName)
 	}
-	kl.podCIDR = node.Spec.PodCIDR
-
 	if err := kl.setNodeStatus(node); err != nil {
 		return err
 	}

pkg/kubelet/kubelet_test.go

@@ -127,7 +127,6 @@ func newTestKubelet(t *testing.T) *TestKubelet {
 	}
 	kubelet.volumeManager = newVolumeManager()
 	kubelet.containerManager, _ = newContainerManager(mockCadvisor, "", "", "")
-	kubelet.networkConfigured = true
 	return &TestKubelet{kubelet, fakeRuntime, mockCadvisor, fakeKubeClient, fakeMirrorClient}
 }
 

pkg/kubelet/status_manager.go

@@ -17,7 +17,6 @@ limitations under the License.
 package kubelet
 
 import (
-	"errors"
 	"fmt"
 	"reflect"
 	"sort"
@@ -143,9 +142,6 @@ func (s *statusManager) RemoveOrphanedStatuses(podFullNames map[string]bool) {
 
 // syncBatch syncs pods statuses with the apiserver.
 func (s *statusManager) syncBatch() error {
-	if s.kubeClient == nil {
-		return errors.New("Kubernetes client is nil, skipping pod status updates")
-	}
 	syncRequest := <-s.podStatusChannel
 	pod := syncRequest.pod
 	podFullName := kubecontainer.GetPodFullName(pod)

pkg/util/util.go

@@ -198,20 +198,6 @@ func CompileRegexps(regexpStrings []string) ([]*regexp.Regexp, error) {
 	return regexps, nil
 }
 
-// Detects if using systemd as the init system
-// Please note that simply reading /proc/1/cmdline can be misleading because
-// some installation of various init programs can automatically make /sbin/init
-// a symlink or even a renamed version of their main program.
-// TODO(dchen1107): realiably detects the init system using on the system:
-// systemd, upstart, initd, etc.
-func UsingSystemdInitSystem() bool {
-	if _, err := os.Stat("/run/systemd/system"); err != nil {
-		return true
-	}
-
-	return false
-}
-
 // Writes 'value' to /proc/<pid>/oom_score_adj. PID = 0 means self
 func ApplyOomScoreAdj(pid int, value int) error {
 	if value < -1000 || value > 1000 {