func ensureProcessInContainer(pid int, oomScoreAdj int, manager *fs.Manager) error {
if runningInHost, err := isProcessRunningInHost(pid); err != nil {
// Err on the side of caution. Avoid moving the docker daemon unless we are able to identify its context.
return err
} else if !runningInHost {
// Process is running inside a container. Don't touch that.
return nil
}
var errs []error
cont, err := getContainer(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to find container of PID %d: %v", pid, err))
}
if cont != manager.Cgroups.Name {
err = manager.Apply(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to move PID %d (in %q) to %q", pid, cont, manager.Cgroups.Name))
}
}
// Also apply oom-score-adj to processes
oomAdjuster := oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(pid, oomScoreAdj); err != nil {
errs = append(errs, fmt.Errorf("failed to apply oom score %d to PID %d", oomScoreAdj, pid))
}
return utilerrors.NewAggregate(errs)
}
// Ensures that the Docker daemon is in the desired container.
func ensureDockerInContainer(cadvisor cadvisor.Interface, oomScoreAdj int, manager *fs.Manager) error {
// What container is Docker in?
out, err := exec.Command("pidof", "docker").Output()
if err != nil {
return fmt.Errorf("failed to find pid of Docker container: %v", err)
}
// The output of pidof is a list of pids.
// Docker may be forking and thus there would be more than one result.
pids := []int{}
for _, pidStr := range strings.Split(strings.TrimSpace(string(out)), " ") {
pid, err := strconv.Atoi(pidStr)
if err != nil {
continue
}
pids = append(pids, pid)
}
// Move if the pid is not already in the desired container.
errs := []error{}
for _, pid := range pids {
if runningInHost, err := isProcessRunningInHost(pid); err != nil {
errs = append(errs, err)
// Err on the side of caution. Avoid moving the docker daemon unless we are able to identify its context.
continue
} else if !runningInHost {
// Docker daemon is running inside a container. Don't touch that.
continue
}
cont, err := getContainer(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to find container of PID %d: %v", pid, err))
}
if cont != manager.Cgroups.Name {
err = manager.Apply(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to move PID %d (in %q) to %q", pid, cont, manager.Cgroups.Name))
}
}
// Also apply oom-score-adj to processes
oomAdjuster := oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(pid, oomScoreAdj); err != nil {
errs = append(errs, fmt.Errorf("failed to apply oom score %d to PID %d", oomScoreAdj, pid))
}
}
return utilerrors.NewAggregate(errs)
}
// UnsecuredKubeletDeps returns a KubeletDeps suitable for being run, or an error if the server setup
// is not valid. It will not start any background processes, and does not include authentication/authorization
func UnsecuredKubeletDeps(s *options.KubeletServer) (*kubelet.KubeletDeps, error) {
// Initialize the TLS Options
tlsOptions, err := InitializeTLS(&s.KubeletConfiguration)
if err != nil {
return nil, err
}
mounter := mount.New(s.ExperimentalMounterPath)
var writer kubeio.Writer = &kubeio.StdWriter{}
if s.Containerized {
glog.V(2).Info("Running kubelet in containerized mode (experimental)")
mounter = mount.NewNsenterMounter()
writer = &kubeio.NsenterWriter{}
}
var dockerClient dockertools.DockerInterface
if s.ContainerRuntime == "docker" {
dockerClient = dockertools.ConnectToDockerOrDie(s.DockerEndpoint, s.RuntimeRequestTimeout.Duration,
s.ImagePullProgressDeadline.Duration)
} else {
dockerClient = nil
}
return &kubelet.KubeletDeps{
Auth: nil, // default does not enforce auth[nz]
CAdvisorInterface: nil, // cadvisor.New launches background processes (bg http.ListenAndServe, and some bg cleaners), not set here
Cloud: nil, // cloud provider might start background processes
ContainerManager: nil,
DockerClient: dockerClient,
KubeClient: nil,
ExternalKubeClient: nil,
Mounter: mounter,
NetworkPlugins: ProbeNetworkPlugins(s.NetworkPluginDir, s.CNIConfDir, s.CNIBinDir),
OOMAdjuster: oom.NewOOMAdjuster(),
OSInterface: kubecontainer.RealOS{},
Writer: writer,
VolumePlugins: ProbeVolumePlugins(s.VolumePluginDir),
TLSOptions: tlsOptions,
}, nil
}
// Ensures that the Docker daemon is in the desired container.
func ensureDockerInContainer(cadvisor cadvisor.Interface, oomScoreAdj int, manager *fs.Manager) error {
pids, err := getPidsForProcess("docker")
if err != nil {
return err
}
// Move if the pid is not already in the desired container.
errs := []error{}
for _, pid := range pids {
if runningInHost, err := isProcessRunningInHost(pid); err != nil {
errs = append(errs, err)
// Err on the side of caution. Avoid moving the docker daemon unless we are able to identify its context.
continue
} else if !runningInHost {
// Docker daemon is running inside a container. Don't touch that.
continue
}
cont, err := getContainer(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to find container of PID %d: %v", pid, err))
}
if cont != manager.Cgroups.Name {
err = manager.Apply(pid)
if err != nil {
errs = append(errs, fmt.Errorf("failed to move PID %d (in %q) to %q", pid, cont, manager.Cgroups.Name))
}
}
// Also apply oom-score-adj to processes
oomAdjuster := oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(pid, oomScoreAdj); err != nil {
errs = append(errs, fmt.Errorf("failed to apply oom score %d to PID %d", oomScoreAdj, pid))
}
}
return utilerrors.NewAggregate(errs)
}
// Run runs the specified KubeletExecutorServer.
func (s *KubeletExecutorServer) Run(hks hyperkube.Interface, _ []string) error {
rand.Seed(time.Now().UTC().UnixNano())
oomAdjuster := oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(0, s.OOMScoreAdj); err != nil {
log.Info(err)
}
// empty string for the docker and system containers (= cgroup paths). This
// stops the kubelet taking any control over other system processes.
s.SystemContainer = ""
s.DockerDaemonContainer = ""
// create apiserver client
var apiclient *client.Client
clientConfig, err := s.CreateAPIServerClientConfig()
if err == nil {
apiclient, err = client.New(clientConfig)
}
if err != nil {
// required for k8sm since we need to send api.Binding information
// back to the apiserver
log.Fatalf("No API client: %v", err)
}
log.Infof("Using root directory: %v", s.RootDirectory)
credentialprovider.SetPreferredDockercfgPath(s.RootDirectory)
cAdvisorInterface, err := cadvisor.New(s.CAdvisorPort)
if err != nil {
return err
}
imageGCPolicy := kubelet.ImageGCPolicy{
HighThresholdPercent: s.ImageGCHighThresholdPercent,
LowThresholdPercent: s.ImageGCLowThresholdPercent,
}
diskSpacePolicy := kubelet.DiskSpacePolicy{
DockerFreeDiskMB: s.LowDiskSpaceThresholdMB,
RootFreeDiskMB: s.LowDiskSpaceThresholdMB,
}
//TODO(jdef) intentionally NOT initializing a cloud provider here since:
//(a) the kubelet doesn't actually use it
//(b) we don't need to create N-kubelet connections to zookeeper for no good reason
//cloud := cloudprovider.InitCloudProvider(s.CloudProvider, s.CloudConfigFile)
//log.Infof("Successfully initialized cloud provider: %q from the config file: %q\n", s.CloudProvider, s.CloudConfigFile)
hostNetworkSources, err := kubelet.GetValidatedSources(strings.Split(s.HostNetworkSources, ","))
if err != nil {
return err
}
hostPIDSources, err := kubelet.GetValidatedSources(strings.Split(s.HostPIDSources, ","))
if err != nil {
return err
}
hostIPCSources, err := kubelet.GetValidatedSources(strings.Split(s.HostIPCSources, ","))
if err != nil {
return err
}
tlsOptions, err := s.InitializeTLS()
if err != nil {
return err
}
mounter := mount.New()
if s.Containerized {
log.V(2).Info("Running kubelet in containerized mode (experimental)")
mounter = &mount.NsenterMounter{}
}
var writer utilio.Writer = &utilio.StdWriter{}
var dockerExecHandler dockertools.ExecHandler
switch s.DockerExecHandlerName {
case "native":
dockerExecHandler = &dockertools.NativeExecHandler{}
case "nsenter":
writer = &utilio.NsenterWriter{}
dockerExecHandler = &dockertools.NsenterExecHandler{}
default:
log.Warningf("Unknown Docker exec handler %q; defaulting to native", s.DockerExecHandlerName)
dockerExecHandler = &dockertools.NativeExecHandler{}
}
manifestURLHeader := make(http.Header)
if s.ManifestURLHeader != "" {
pieces := strings.Split(s.ManifestURLHeader, ":")
if len(pieces) != 2 {
return fmt.Errorf("manifest-url-header must have a single ':' key-value separator, got %q", s.ManifestURLHeader)
}
manifestURLHeader.Set(pieces[0], pieces[1])
}
kcfg := app.KubeletConfig{
Address: s.Address,
AllowPrivileged: s.AllowPrivileged,
CAdvisorInterface: cAdvisorInterface,
CgroupRoot: s.CgroupRoot,
Cloud: nil, // TODO(jdef) Cloud, specifying null here because we don't want all kubelets polling mesos-master; need to account for this in the cloudprovider impl
ClusterDNS: s.ClusterDNS,
ClusterDomain: s.ClusterDomain,
// ConfigFile: ""
ConfigureCBR0: s.ConfigureCBR0,
ContainerRuntime: s.ContainerRuntime,
CPUCFSQuota: s.CPUCFSQuota,
DiskSpacePolicy: diskSpacePolicy,
DockerClient: dockertools.ConnectToDockerOrDie(s.DockerEndpoint),
DockerDaemonContainer: s.DockerDaemonContainer,
DockerExecHandler: dockerExecHandler,
EnableDebuggingHandlers: s.EnableDebuggingHandlers,
EnableServer: s.EnableServer,
EventBurst: s.EventBurst,
EventRecordQPS: s.EventRecordQPS,
FileCheckFrequency: s.FileCheckFrequency,
HostnameOverride: s.HostnameOverride,
HostNetworkSources: hostNetworkSources,
HostPIDSources: hostPIDSources,
HostIPCSources: hostIPCSources,
// HTTPCheckFrequency
ImageGCPolicy: imageGCPolicy,
KubeClient: apiclient,
// ManifestURL: ""
ManifestURLHeader: manifestURLHeader,
MasterServiceNamespace: s.MasterServiceNamespace,
MaxContainerCount: s.MaxContainerCount,
MaxOpenFiles: s.MaxOpenFiles,
MaxPerPodContainerCount: s.MaxPerPodContainerCount,
MaxPods: s.MaxPods,
MinimumGCAge: s.MinimumGCAge,
Mounter: mounter,
NetworkPluginName: s.NetworkPluginName,
NetworkPlugins: app.ProbeNetworkPlugins(s.NetworkPluginDir),
NodeStatusUpdateFrequency: s.NodeStatusUpdateFrequency,
OOMAdjuster: oomAdjuster,
OSInterface: kubecontainer.RealOS{},
PodCIDR: s.PodCIDR,
PodInfraContainerImage: s.PodInfraContainerImage,
Port: s.Port,
ReadOnlyPort: s.ReadOnlyPort,
RegisterNode: s.RegisterNode,
RegistryBurst: s.RegistryBurst,
RegistryPullQPS: s.RegistryPullQPS,
ResolverConfig: s.ResolverConfig,
ResourceContainer: s.ResourceContainer,
RootDirectory: s.RootDirectory,
Runonce: s.RunOnce,
// StandaloneMode: false
StreamingConnectionIdleTimeout: s.StreamingConnectionIdleTimeout,
SyncFrequency: s.SyncFrequency,
SystemContainer: s.SystemContainer,
TLSOptions: tlsOptions,
VolumePlugins: app.ProbeVolumePlugins(),
Writer: writer,
}
kcfg.NodeName = kcfg.Hostname
kcfg.Builder = app.KubeletBuilder(func(kc *app.KubeletConfig) (app.KubeletBootstrap, *kconfig.PodConfig, error) {
return s.createAndInitKubelet(kc, hks, clientConfig)
})
err = app.RunKubelet(&kcfg)
if err != nil {
return err
}
if s.HealthzPort > 0 {
healthz.DefaultHealthz()
go util.Until(func() {
err := http.ListenAndServe(net.JoinHostPort(s.HealthzBindAddress.String(), strconv.Itoa(s.HealthzPort)), nil)
if err != nil {
log.Errorf("Starting health server failed: %v", err)
}
}, 5*time.Second, util.NeverStop)
}
// block until executor is shut down or commits shutdown
select {}
}
// UnsecuredKubeletConfig returns a KubeletConfig suitable for being run, or an error if the server setup
// is not valid. It will not start any background processes, and does not include authentication/authorization
func UnsecuredKubeletConfig(s *options.KubeletServer) (*KubeletConfig, error) {
hostNetworkSources, err := kubetypes.GetValidatedSources(strings.Split(s.HostNetworkSources, ","))
if err != nil {
return nil, err
}
hostPIDSources, err := kubetypes.GetValidatedSources(strings.Split(s.HostPIDSources, ","))
if err != nil {
return nil, err
}
hostIPCSources, err := kubetypes.GetValidatedSources(strings.Split(s.HostIPCSources, ","))
if err != nil {
return nil, err
}
mounter := mount.New()
var writer io.Writer = &io.StdWriter{}
if s.Containerized {
glog.V(2).Info("Running kubelet in containerized mode (experimental)")
mounter = mount.NewNsenterMounter()
writer = &io.NsenterWriter{}
}
chmodRunner := chmod.New()
chownRunner := chown.New()
tlsOptions, err := InitializeTLS(s)
if err != nil {
return nil, err
}
var dockerExecHandler dockertools.ExecHandler
switch s.DockerExecHandlerName {
case "native":
dockerExecHandler = &dockertools.NativeExecHandler{}
case "nsenter":
dockerExecHandler = &dockertools.NsenterExecHandler{}
default:
glog.Warningf("Unknown Docker exec handler %q; defaulting to native", s.DockerExecHandlerName)
dockerExecHandler = &dockertools.NativeExecHandler{}
}
imageGCPolicy := kubelet.ImageGCPolicy{
HighThresholdPercent: s.ImageGCHighThresholdPercent,
LowThresholdPercent: s.ImageGCLowThresholdPercent,
}
diskSpacePolicy := kubelet.DiskSpacePolicy{
DockerFreeDiskMB: s.LowDiskSpaceThresholdMB,
RootFreeDiskMB: s.LowDiskSpaceThresholdMB,
}
manifestURLHeader := make(http.Header)
if s.ManifestURLHeader != "" {
pieces := strings.Split(s.ManifestURLHeader, ":")
if len(pieces) != 2 {
return nil, fmt.Errorf("manifest-url-header must have a single ':' key-value separator, got %q", s.ManifestURLHeader)
}
manifestURLHeader.Set(pieces[0], pieces[1])
}
reservation, err := parseReservation(s.KubeReserved, s.SystemReserved)
if err != nil {
return nil, err
}
return &KubeletConfig{
Address: s.Address,
AllowPrivileged: s.AllowPrivileged,
Auth: nil, // default does not enforce auth[nz]
CAdvisorInterface: nil, // launches background processes, not set here
CgroupRoot: s.CgroupRoot,
Cloud: nil, // cloud provider might start background processes
ClusterDNS: s.ClusterDNS,
ClusterDomain: s.ClusterDomain,
ConfigFile: s.Config,
ConfigureCBR0: s.ConfigureCBR0,
ContainerManager: nil,
ContainerRuntime: s.ContainerRuntime,
CPUCFSQuota: s.CPUCFSQuota,
DiskSpacePolicy: diskSpacePolicy,
DockerClient: dockertools.ConnectToDockerOrDie(s.DockerEndpoint),
DockerDaemonContainer: s.DockerDaemonContainer,
DockerExecHandler: dockerExecHandler,
EnableDebuggingHandlers: s.EnableDebuggingHandlers,
EnableServer: s.EnableServer,
EventBurst: s.EventBurst,
EventRecordQPS: s.EventRecordQPS,
FileCheckFrequency: s.FileCheckFrequency,
HostnameOverride: s.HostnameOverride,
HostNetworkSources: hostNetworkSources,
HostPIDSources: hostPIDSources,
HostIPCSources: hostIPCSources,
HTTPCheckFrequency: s.HTTPCheckFrequency,
ImageGCPolicy: imageGCPolicy,
KubeClient: nil,
ManifestURL: s.ManifestURL,
ManifestURLHeader: manifestURLHeader,
MasterServiceNamespace: s.MasterServiceNamespace,
MaxContainerCount: s.MaxContainerCount,
MaxOpenFiles: s.MaxOpenFiles,
MaxPerPodContainerCount: s.MaxPerPodContainerCount,
MaxPods: s.MaxPods,
MinimumGCAge: s.MinimumGCAge,
Mounter: mounter,
ChownRunner: chownRunner,
ChmodRunner: chmodRunner,
NetworkPluginName: s.NetworkPluginName,
NetworkPlugins: ProbeNetworkPlugins(s.NetworkPluginDir),
NodeLabels: s.NodeLabels,
NodeStatusUpdateFrequency: s.NodeStatusUpdateFrequency,
OOMAdjuster: oom.NewOOMAdjuster(),
OSInterface: kubecontainer.RealOS{},
PodCIDR: s.PodCIDR,
ReconcileCIDR: s.ReconcileCIDR,
PodInfraContainerImage: s.PodInfraContainerImage,
Port: s.Port,
ReadOnlyPort: s.ReadOnlyPort,
RegisterNode: s.RegisterNode,
RegisterSchedulable: s.RegisterSchedulable,
RegistryBurst: s.RegistryBurst,
RegistryPullQPS: s.RegistryPullQPS,
ResolverConfig: s.ResolverConfig,
Reservation: *reservation,
ResourceContainer: s.ResourceContainer,
RktPath: s.RktPath,
RktStage1Image: s.RktStage1Image,
RootDirectory: s.RootDirectory,
Runonce: s.RunOnce,
SerializeImagePulls: s.SerializeImagePulls,
StandaloneMode: (len(s.APIServerList) == 0),
StreamingConnectionIdleTimeout: s.StreamingConnectionIdleTimeout,
SyncFrequency: s.SyncFrequency,
SystemContainer: s.SystemContainer,
TLSOptions: tlsOptions,
Writer: writer,
VolumePlugins: ProbeVolumePlugins(s.VolumePluginDir),
OutOfDiskTransitionFrequency: s.OutOfDiskTransitionFrequency,
ExperimentalFlannelOverlay: s.ExperimentalFlannelOverlay,
NodeIP: s.NodeIP,
}, nil
}
// NewProxyServerDefault creates a new ProxyServer object with default parameters.
func NewProxyServerDefault(config *options.ProxyServerConfig) (*ProxyServer, error) {
if c, err := configz.New("componentconfig"); err == nil {
c.Set(config.KubeProxyConfiguration)
} else {
glog.Errorf("unable to register configz: %s", err)
}
protocol := utiliptables.ProtocolIpv4
if net.ParseIP(config.BindAddress).To4() == nil {
protocol = utiliptables.ProtocolIpv6
}
// Create a iptables utils.
execer := exec.New()
dbus := utildbus.New()
iptInterface := utiliptables.New(execer, dbus, protocol)
// We omit creation of pretty much everything if we run in cleanup mode
if config.CleanupAndExit {
return &ProxyServer{
Config: config,
IptInterface: iptInterface,
}, nil
}
// TODO(vmarmol): Use container config for this.
var oomAdjuster *oom.OOMAdjuster
if config.OOMScoreAdj != nil {
oomAdjuster = oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(0, int(*config.OOMScoreAdj)); err != nil {
glog.V(2).Info(err)
}
}
if config.ResourceContainer != "" {
// Run in its own container.
if err := resourcecontainer.RunInResourceContainer(config.ResourceContainer); err != nil {
glog.Warningf("Failed to start in resource-only container %q: %v", config.ResourceContainer, err)
} else {
glog.V(2).Infof("Running in resource-only container %q", config.ResourceContainer)
}
}
// Create a Kube Client
// define api config source
if config.Kubeconfig == "" && config.Master == "" {
glog.Warningf("Neither --kubeconfig nor --master was specified. Using default API client. This might not work.")
}
// This creates a client, first loading any specified kubeconfig
// file, and then overriding the Master flag, if non-empty.
kubeconfig, err := clientcmd.NewNonInteractiveDeferredLoadingClientConfig(
&clientcmd.ClientConfigLoadingRules{ExplicitPath: config.Kubeconfig},
&clientcmd.ConfigOverrides{ClusterInfo: clientcmdapi.Cluster{Server: config.Master}}).ClientConfig()
if err != nil {
return nil, err
}
kubeconfig.ContentType = config.ContentType
// Override kubeconfig qps/burst settings from flags
kubeconfig.QPS = config.KubeAPIQPS
kubeconfig.Burst = int(config.KubeAPIBurst)
client, err := clientset.NewForConfig(kubeconfig)
if err != nil {
glog.Fatalf("Invalid API configuration: %v", err)
}
// Create event recorder
hostname := nodeutil.GetHostname(config.HostnameOverride)
eventBroadcaster := record.NewBroadcaster()
recorder := eventBroadcaster.NewRecorder(api.EventSource{Component: "kube-proxy", Host: hostname})
var proxier proxy.ProxyProvider
var endpointsHandler proxyconfig.EndpointsConfigHandler
proxyMode := getProxyMode(string(config.Mode), client.Core().Nodes(), hostname, iptInterface, iptables.LinuxKernelCompatTester{})
if proxyMode == proxyModeIPTables {
glog.V(0).Info("Using iptables Proxier.")
if config.IPTablesMasqueradeBit == nil {
// IPTablesMasqueradeBit must be specified or defaulted.
return nil, fmt.Errorf("Unable to read IPTablesMasqueradeBit from config")
}
proxierIPTables, err := iptables.NewProxier(iptInterface, utilsysctl.New(), execer, config.IPTablesSyncPeriod.Duration, config.MasqueradeAll, int(*config.IPTablesMasqueradeBit), config.ClusterCIDR, hostname, getNodeIP(client, hostname))
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierIPTables
endpointsHandler = proxierIPTables
// No turning back. Remove artifacts that might still exist from the userspace Proxier.
glog.V(0).Info("Tearing down userspace rules.")
userspace.CleanupLeftovers(iptInterface)
} else {
glog.V(0).Info("Using userspace Proxier.")
// This is a proxy.LoadBalancer which NewProxier needs but has methods we don't need for
// our config.EndpointsConfigHandler.
loadBalancer := userspace.NewLoadBalancerRR()
// set EndpointsConfigHandler to our loadBalancer
endpointsHandler = loadBalancer
proxierUserspace, err := userspace.NewProxier(
loadBalancer,
net.ParseIP(config.BindAddress),
iptInterface,
*utilnet.ParsePortRangeOrDie(config.PortRange),
config.IPTablesSyncPeriod.Duration,
config.UDPIdleTimeout.Duration,
)
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierUserspace
// Remove artifacts from the pure-iptables Proxier.
glog.V(0).Info("Tearing down pure-iptables proxy rules.")
iptables.CleanupLeftovers(iptInterface)
}
iptInterface.AddReloadFunc(proxier.Sync)
// Create configs (i.e. Watches for Services and Endpoints)
// Note: RegisterHandler() calls need to happen before creation of Sources because sources
// only notify on changes, and the initial update (on process start) may be lost if no handlers
// are registered yet.
serviceConfig := proxyconfig.NewServiceConfig()
serviceConfig.RegisterHandler(proxier)
endpointsConfig := proxyconfig.NewEndpointsConfig()
endpointsConfig.RegisterHandler(endpointsHandler)
proxyconfig.NewSourceAPI(
client.Core().RESTClient(),
config.ConfigSyncPeriod,
serviceConfig.Channel("api"),
endpointsConfig.Channel("api"),
)
config.NodeRef = &api.ObjectReference{
Kind: "Node",
Name: hostname,
UID: types.UID(hostname),
Namespace: "",
}
conntracker := realConntracker{}
return NewProxyServer(client, config, iptInterface, proxier, eventBroadcaster, recorder, conntracker, proxyMode)
}
// UnsecuredKubeletConfig returns a KubeletConfig suitable for being run, or an error if the server setup
// is not valid. It will not start any background processes, and does not include authentication/authorization
func UnsecuredKubeletConfig(s *options.KubeletServer) (*KubeletConfig, error) {
hostNetworkSources, err := kubetypes.GetValidatedSources(s.HostNetworkSources)
if err != nil {
return nil, err
}
hostPIDSources, err := kubetypes.GetValidatedSources(s.HostPIDSources)
if err != nil {
return nil, err
}
hostIPCSources, err := kubetypes.GetValidatedSources(s.HostIPCSources)
if err != nil {
return nil, err
}
mounter := mount.New()
var writer io.Writer = &io.StdWriter{}
if s.Containerized {
glog.V(2).Info("Running kubelet in containerized mode (experimental)")
mounter = mount.NewNsenterMounter()
writer = &io.NsenterWriter{}
}
tlsOptions, err := InitializeTLS(s)
if err != nil {
return nil, err
}
var dockerExecHandler dockertools.ExecHandler
switch s.DockerExecHandlerName {
case "native":
dockerExecHandler = &dockertools.NativeExecHandler{}
case "nsenter":
dockerExecHandler = &dockertools.NsenterExecHandler{}
default:
glog.Warningf("Unknown Docker exec handler %q; defaulting to native", s.DockerExecHandlerName)
dockerExecHandler = &dockertools.NativeExecHandler{}
}
imageGCPolicy := images.ImageGCPolicy{
MinAge: s.ImageMinimumGCAge.Duration,
HighThresholdPercent: int(s.ImageGCHighThresholdPercent),
LowThresholdPercent: int(s.ImageGCLowThresholdPercent),
}
diskSpacePolicy := kubelet.DiskSpacePolicy{
DockerFreeDiskMB: int(s.LowDiskSpaceThresholdMB),
RootFreeDiskMB: int(s.LowDiskSpaceThresholdMB),
}
manifestURLHeader := make(http.Header)
if s.ManifestURLHeader != "" {
pieces := strings.Split(s.ManifestURLHeader, ":")
if len(pieces) != 2 {
return nil, fmt.Errorf("manifest-url-header must have a single ':' key-value separator, got %q", s.ManifestURLHeader)
}
manifestURLHeader.Set(pieces[0], pieces[1])
}
reservation, err := parseReservation(s.KubeReserved, s.SystemReserved)
if err != nil {
return nil, err
}
thresholds, err := eviction.ParseThresholdConfig(s.EvictionHard, s.EvictionSoft, s.EvictionSoftGracePeriod, s.EvictionMinimumReclaim)
if err != nil {
return nil, err
}
evictionConfig := eviction.Config{
PressureTransitionPeriod: s.EvictionPressureTransitionPeriod.Duration,
MaxPodGracePeriodSeconds: int64(s.EvictionMaxPodGracePeriod),
Thresholds: thresholds,
}
return &KubeletConfig{
Address: net.ParseIP(s.Address),
AllowPrivileged: s.AllowPrivileged,
Auth: nil, // default does not enforce auth[nz]
CAdvisorInterface: nil, // launches background processes, not set here
VolumeStatsAggPeriod: s.VolumeStatsAggPeriod.Duration,
CgroupRoot: s.CgroupRoot,
Cloud: nil, // cloud provider might start background processes
ClusterDNS: net.ParseIP(s.ClusterDNS),
ClusterDomain: s.ClusterDomain,
ConfigFile: s.Config,
ConfigureCBR0: s.ConfigureCBR0,
ContainerManager: nil,
ContainerRuntime: s.ContainerRuntime,
RuntimeRequestTimeout: s.RuntimeRequestTimeout.Duration,
CPUCFSQuota: s.CPUCFSQuota,
DiskSpacePolicy: diskSpacePolicy,
DockerClient: dockertools.ConnectToDockerOrDie(s.DockerEndpoint, s.RuntimeRequestTimeout.Duration), // TODO(random-liu): Set RuntimeRequestTimeout for rkt.
RuntimeCgroups: s.RuntimeCgroups,
DockerExecHandler: dockerExecHandler,
EnableControllerAttachDetach: s.EnableControllerAttachDetach,
EnableCustomMetrics: s.EnableCustomMetrics,
EnableDebuggingHandlers: s.EnableDebuggingHandlers,
CgroupsPerQOS: s.CgroupsPerQOS,
EnableServer: s.EnableServer,
EventBurst: int(s.EventBurst),
EventRecordQPS: float32(s.EventRecordQPS),
FileCheckFrequency: s.FileCheckFrequency.Duration,
HostnameOverride: s.HostnameOverride,
HostNetworkSources: hostNetworkSources,
HostPIDSources: hostPIDSources,
HostIPCSources: hostIPCSources,
HTTPCheckFrequency: s.HTTPCheckFrequency.Duration,
ImageGCPolicy: imageGCPolicy,
KubeClient: nil,
ManifestURL: s.ManifestURL,
ManifestURLHeader: manifestURLHeader,
MasterServiceNamespace: s.MasterServiceNamespace,
MaxContainerCount: int(s.MaxContainerCount),
MaxOpenFiles: uint64(s.MaxOpenFiles),
MaxPerPodContainerCount: int(s.MaxPerPodContainerCount),
MaxPods: int(s.MaxPods),
NvidiaGPUs: int(s.NvidiaGPUs),
MinimumGCAge: s.MinimumGCAge.Duration,
Mounter: mounter,
NetworkPluginName: s.NetworkPluginName,
NetworkPlugins: ProbeNetworkPlugins(s.NetworkPluginDir),
NodeLabels: s.NodeLabels,
NodeStatusUpdateFrequency: s.NodeStatusUpdateFrequency.Duration,
NonMasqueradeCIDR: s.NonMasqueradeCIDR,
OOMAdjuster: oom.NewOOMAdjuster(),
OSInterface: kubecontainer.RealOS{},
PodCIDR: s.PodCIDR,
ReconcileCIDR: s.ReconcileCIDR,
PodInfraContainerImage: s.PodInfraContainerImage,
Port: uint(s.Port),
ReadOnlyPort: uint(s.ReadOnlyPort),
RegisterNode: s.RegisterNode,
RegisterSchedulable: s.RegisterSchedulable,
RegistryBurst: int(s.RegistryBurst),
RegistryPullQPS: float64(s.RegistryPullQPS),
ResolverConfig: s.ResolverConfig,
Reservation: *reservation,
KubeletCgroups: s.KubeletCgroups,
RktPath: s.RktPath,
RktAPIEndpoint: s.RktAPIEndpoint,
RktStage1Image: s.RktStage1Image,
RootDirectory: s.RootDirectory,
SeccompProfileRoot: s.SeccompProfileRoot,
Runonce: s.RunOnce,
SerializeImagePulls: s.SerializeImagePulls,
StandaloneMode: (len(s.APIServerList) == 0),
StreamingConnectionIdleTimeout: s.StreamingConnectionIdleTimeout.Duration,
SyncFrequency: s.SyncFrequency.Duration,
SystemCgroups: s.SystemCgroups,
TLSOptions: tlsOptions,
Writer: writer,
VolumePlugins: ProbeVolumePlugins(s.VolumePluginDir),
OutOfDiskTransitionFrequency: s.OutOfDiskTransitionFrequency.Duration,
HairpinMode: s.HairpinMode,
BabysitDaemons: s.BabysitDaemons,
ExperimentalFlannelOverlay: s.ExperimentalFlannelOverlay,
NodeIP: net.ParseIP(s.NodeIP),
EvictionConfig: evictionConfig,
PodsPerCore: int(s.PodsPerCore),
}, nil
}
// NewProxyServerDefault creates a new ProxyServer object with default parameters.
func NewProxyServerDefault(config *ProxyServerConfig) (*ProxyServer, error) {
protocol := utiliptables.ProtocolIpv4
if config.BindAddress.To4() == nil {
protocol = utiliptables.ProtocolIpv6
}
// Create a iptables utils.
execer := exec.New()
dbus := utildbus.New()
iptInterface := utiliptables.New(execer, dbus, protocol)
// We ommit creation of pretty much everything if we run in cleanup mode
if config.CleanupAndExit {
return &ProxyServer{
Config: config,
IptInterface: iptInterface,
}, nil
}
// TODO(vmarmol): Use container config for this.
var oomAdjuster *oom.OOMAdjuster
if config.OOMScoreAdj != 0 {
oomAdjuster = oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(0, config.OOMScoreAdj); err != nil {
glog.V(2).Info(err)
}
}
if config.ResourceContainer != "" {
// Run in its own container.
if err := util.RunInResourceContainer(config.ResourceContainer); err != nil {
glog.Warningf("Failed to start in resource-only container %q: %v", config.ResourceContainer, err)
} else {
glog.V(2).Infof("Running in resource-only container %q", config.ResourceContainer)
}
}
// Create a Kube Client
// define api config source
if config.Kubeconfig == "" && config.Master == "" {
glog.Warningf("Neither --kubeconfig nor --master was specified. Using default API client. This might not work.")
}
// This creates a client, first loading any specified kubeconfig
// file, and then overriding the Master flag, if non-empty.
kubeconfig, err := clientcmd.NewNonInteractiveDeferredLoadingClientConfig(
&clientcmd.ClientConfigLoadingRules{ExplicitPath: config.Kubeconfig},
&clientcmd.ConfigOverrides{ClusterInfo: clientcmdapi.Cluster{Server: config.Master}}).ClientConfig()
if err != nil {
return nil, err
}
// Override kubeconfig qps/burst settings from flags
kubeconfig.QPS = config.KubeApiQps
kubeconfig.Burst = config.KubeApiBurst
client, err := kubeclient.New(kubeconfig)
if err != nil {
glog.Fatalf("Invalid API configuration: %v", err)
}
// Create event recorder
hostname := nodeutil.GetHostname(config.HostnameOverride)
eventBroadcaster := record.NewBroadcaster()
recorder := eventBroadcaster.NewRecorder(api.EventSource{Component: "kube-proxy", Host: hostname})
eventBroadcaster.StartRecordingToSink(client.Events(""))
var proxier proxy.ProxyProvider
var endpointsHandler proxyconfig.EndpointsConfigHandler
useIptablesProxy := false
if mayTryIptablesProxy(config.ProxyMode, client.Nodes(), hostname) {
var err error
// guaranteed false on error, error only necessary for debugging
useIptablesProxy, err = iptables.ShouldUseIptablesProxier()
if err != nil {
glog.Errorf("Can't determine whether to use iptables proxy, using userspace proxier: %v", err)
}
}
if useIptablesProxy {
glog.V(2).Info("Using iptables Proxier.")
proxierIptables, err := iptables.NewProxier(iptInterface, execer, config.IptablesSyncPeriod, config.MasqueradeAll)
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierIptables
endpointsHandler = proxierIptables
// No turning back. Remove artifacts that might still exist from the userspace Proxier.
glog.V(2).Info("Tearing down userspace rules. Errors here are acceptable.")
userspace.CleanupLeftovers(iptInterface)
} else {
glog.V(2).Info("Using userspace Proxier.")
// This is a proxy.LoadBalancer which NewProxier needs but has methods we don't need for
// our config.EndpointsConfigHandler.
loadBalancer := userspace.NewLoadBalancerRR()
// set EndpointsConfigHandler to our loadBalancer
endpointsHandler = loadBalancer
proxierUserspace, err := userspace.NewProxier(loadBalancer, config.BindAddress, iptInterface, config.PortRange, config.IptablesSyncPeriod)
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierUserspace
// Remove artifacts from the pure-iptables Proxier.
glog.V(2).Info("Tearing down pure-iptables proxy rules. Errors here are acceptable.")
iptables.CleanupLeftovers(iptInterface)
}
iptInterface.AddReloadFunc(proxier.Sync)
// Create configs (i.e. Watches for Services and Endpoints)
// Note: RegisterHandler() calls need to happen before creation of Sources because sources
// only notify on changes, and the initial update (on process start) may be lost if no handlers
// are registered yet.
serviceConfig := proxyconfig.NewServiceConfig()
serviceConfig.RegisterHandler(proxier)
endpointsConfig := proxyconfig.NewEndpointsConfig()
endpointsConfig.RegisterHandler(endpointsHandler)
proxyconfig.NewSourceAPI(
client,
config.ConfigSyncPeriod,
serviceConfig.Channel("api"),
endpointsConfig.Channel("api"),
)
config.nodeRef = &api.ObjectReference{
Kind: "Node",
Name: hostname,
UID: types.UID(hostname),
Namespace: "",
}
return NewProxyServer(config, iptInterface, proxier, recorder)
}
// Run runs the specified KubeletServer for the given KubeletConfig. This should never exit.
// The kcfg argument may be nil - if so, it is initialized from the settings on KubeletServer.
// Otherwise, the caller is assumed to have set up the KubeletConfig object and all defaults
// will be ignored.
func (s *KubeletServer) Run(kcfg *KubeletConfig) error {
if kcfg == nil {
cfg, err := s.KubeletConfig()
if err != nil {
return err
}
kcfg = cfg
clientConfig, err := s.CreateAPIServerClientConfig()
if err == nil {
kcfg.KubeClient, err = client.New(clientConfig)
}
if err != nil && len(s.APIServerList) > 0 {
glog.Warningf("No API client: %v", err)
}
cloud, err := cloudprovider.InitCloudProvider(s.CloudProvider, s.CloudConfigFile)
if err != nil {
return err
}
glog.V(2).Infof("Successfully initialized cloud provider: %q from the config file: %q\n", s.CloudProvider, s.CloudConfigFile)
kcfg.Cloud = cloud
}
if kcfg.CAdvisorInterface == nil {
ca, err := cadvisor.New(s.CAdvisorPort)
if err != nil {
return err
}
kcfg.CAdvisorInterface = ca
}
util.ReallyCrash = s.ReallyCrashForTesting
rand.Seed(time.Now().UTC().UnixNano())
credentialprovider.SetPreferredDockercfgPath(s.RootDirectory)
glog.V(2).Infof("Using root directory: %v", s.RootDirectory)
// TODO(vmarmol): Do this through container config.
oomAdjuster := oom.NewOOMAdjuster()
if err := oomAdjuster.ApplyOOMScoreAdj(0, s.OOMScoreAdj); err != nil {
glog.Warning(err)
}
if err := RunKubelet(kcfg, nil); err != nil {
return err
}
if s.HealthzPort > 0 {
healthz.DefaultHealthz()
go util.Until(func() {
err := http.ListenAndServe(net.JoinHostPort(s.HealthzBindAddress.String(), strconv.Itoa(s.HealthzPort)), nil)
if err != nil {
glog.Errorf("Starting health server failed: %v", err)
}
}, 5*time.Second, util.NeverStop)
}
if s.RunOnce {
return nil
}
// run forever
select {}
}