github/kubernetes
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Added flag to set cluster class B network address for pods, add flag to disable allocation CIDRs for Pods. Fixed synchornization bug in NodeController registerNodes().

Browse Source
This commit is contained in:
Jerzy Szczepkowski 2015-04-28 17:02:45 +02:00 committed by CJ Cullen
parent 290c7b94ef
commit e967ffd522
12 changed files with 97 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
cluster/gce/config-default.sh
View File

@ -42,6 +42,7 @@ MINION_SCOPES=("storage-ro" "compute-rw" "https://www.googleapis.com/auth/monito
# Increase the sleep interval value if concerned about API rate limits. 3, in seconds, is the default.
POLL_SLEEP_INTERVAL=3
PORTAL_NET="10.0.0.0/16"
ALLOCATE_NODE_CIDRS=true
# When set to true, Docker Cache is enabled by default as part of the cluster bring up.
ENABLE_DOCKER_REGISTRY_CACHE=true

2
cluster/gce/configure-vm.sh
View File

@ -234,6 +234,8 @@ function create-salt-pillar() {
cat <<EOF >/srv/salt-overlay/pillar/cluster-params.sls
instance_prefix: '$(echo "$INSTANCE_PREFIX" | sed -e "s/'/''/g")'
node_instance_prefix: '$(echo "$NODE_INSTANCE_PREFIX" | sed -e "s/'/''/g")'
cluster_class_b: '$(echo "$KUBE_GCE_CLUSTER_CLASS_B" | sed -e "s/'/''/g")'
allocate_node_cidrs: '$(echo "$ALLOCATE_NODE_CIDRS" | sed -e "s/'/''/g")'
portal_net: '$(echo "$PORTAL_NET" | sed -e "s/'/''/g")'
enable_cluster_monitoring: '$(echo "$ENABLE_CLUSTER_MONITORING" | sed -e "s/'/''/g")'
enable_node_monitoring: '$(echo "$ENABLE_NODE_MONITORING" | sed -e "s/'/''/g")'

2
cluster/gce/coreos/helper.sh
View File

@ -28,9 +28,11 @@ function build-kube-env {
ENV_TIMESTAMP: $(yaml-quote $(date -u +%Y-%m-%dT%T%z))
INSTANCE_PREFIX: $(yaml-quote ${INSTANCE_PREFIX})
NODE_INSTANCE_PREFIX: $(yaml-quote ${NODE_INSTANCE_PREFIX})
KUBE_GCE_CLUSTER_CLASS_B: $(yaml-quote ${KUBE_GCE_CLUSTER_CLASS_B:-10.244})
SERVER_BINARY_TAR_URL: $(yaml-quote ${SERVER_BINARY_TAR_URL})
SALT_TAR_URL: $(yaml-quote ${SALT_TAR_URL})
PORTAL_NET: $(yaml-quote ${PORTAL_NET})
ALLOCATE_NODE_CIDRS: $(yaml-quote ${ALLOCATE_NODE_CIDRS:-false})
ENABLE_CLUSTER_MONITORING: $(yaml-quote ${ENABLE_CLUSTER_MONITORING:-false})
ENABLE_NODE_MONITORING: $(yaml-quote ${ENABLE_NODE_MONITORING:-false})
ENABLE_CLUSTER_LOGGING: $(yaml-quote ${ENABLE_CLUSTER_LOGGING:-false})

2
cluster/gce/debian/helper.sh
View File

@ -26,9 +26,11 @@ function build-kube-env {
ENV_TIMESTAMP: $(yaml-quote $(date -u +%Y-%m-%dT%T%z))
INSTANCE_PREFIX: $(yaml-quote ${INSTANCE_PREFIX})
NODE_INSTANCE_PREFIX: $(yaml-quote ${NODE_INSTANCE_PREFIX})
KUBE_GCE_CLUSTER_CLASS_B: $(yaml-quote ${KUBE_GCE_CLUSTER_CLASS_B:-10.244})
SERVER_BINARY_TAR_URL: $(yaml-quote ${SERVER_BINARY_TAR_URL})
SALT_TAR_URL: $(yaml-quote ${SALT_TAR_URL})
PORTAL_NET: $(yaml-quote ${PORTAL_NET})
ALLOCATE_NODE_CIDRS: $(yaml-quote ${ALLOCATE_NODE_CIDRS:-false})
ENABLE_CLUSTER_MONITORING: $(yaml-quote ${ENABLE_CLUSTER_MONITORING:-false})
ENABLE_NODE_MONITORING: $(yaml-quote ${ENABLE_NODE_MONITORING:-false})
ENABLE_CLUSTER_LOGGING: $(yaml-quote ${ENABLE_CLUSTER_LOGGING:-false})

2
cluster/gce/util.sh
View File

@ -32,6 +32,8 @@ fi
NODE_INSTANCE_PREFIX="${INSTANCE_PREFIX}-minion"
ALLOCATE_NODE_CIDRS=true
KUBE_PROMPT_FOR_UPDATE=y
KUBE_SKIP_UPDATE=${KUBE_SKIP_UPDATE-"n"}

9
cluster/saltbase/salt/kube-controller-manager/kube-controller-manager.manifest
View File

@ -1,5 +1,6 @@
{% set machines = ""-%}
{% set cluster_name = "" -%}
{% set cluster_class_b = "" -%}
{% set minion_regexp = "--minion_regexp=.*" -%}
{% set sync_nodes = "--sync_nodes=true" -%}
@ -9,6 +10,12 @@
{% if pillar['instance_prefix'] is defined -%}
{% set cluster_name = "--cluster_name=" + pillar['instance_prefix'] -%}
{% endif -%}
{% if pillar['cluster_class_b'] is defined -%}
{% set cluster_class_b = "--cluster-class-b=" + pillar['cluster_class_b'] -%}
{% endif -%}
{% if pillar['allocate_node_cidrs'] is defined -%}
{% set allocate_node_cidrs = "--allocate-node-cidrs=" + pillar['allocate_node_cidrs'] -%}
{% endif -%}
{% set cloud_provider = "" -%}
{% set cloud_config = "" -%}
@ -47,7 +54,7 @@
{% endif -%}
{% endif -%}
{% set params = "--master=127.0.0.1:8080" + " " + machines + " " + cluster_name + " " + minion_regexp + " " + cloud_provider + " " + sync_nodes + " " + cloud_config + " " + pillar['log_level'] -%}
{% set params = "--master=127.0.0.1:8080" + " " + machines + " " + cluster_name + " " + cluster_class_b + " " + allocate_node_cidrs + " " + minion_regexp + " " + cloud_provider + " " + sync_nodes + " " + cloud_config + " " + pillar['log_level'] -%}
{
"apiVersion": "v1beta3",

2
cmd/integration/integration.go
View File

@ -225,7 +225,7 @@ func startComponents(firstManifestURL, secondManifestURL, apiVersion string) (st
api.ResourceName(api.ResourceMemory): resource.MustParse("10G"),
}}
nodeController := nodecontroller.NewNodeController(nil, "", machineList, nodeResources, cl, 10, 5*time.Minute, util.NewFakeRateLimiter(), 40*time.Second, 60*time.Second, 5*time.Second, "")
nodeController := nodecontroller.NewNodeController(nil, "", machineList, nodeResources, cl, 10, 5*time.Minute, util.NewFakeRateLimiter(), 40*time.Second, 60*time.Second, 5*time.Second, "", "", false)
nodeController.Run(5*time.Second, true)
cadvisorInterface := new(cadvisor.Fake)

6
cmd/kube-controller-manager/app/controllermanager.go
View File

@ -79,6 +79,8 @@ type CMServer struct {
NodeMemory resource.Quantity
ClusterName string
ClusterClassB string
AllocateNodeCIDRs bool
EnableProfiling bool
Master string
@ -145,6 +147,8 @@ func (s *CMServer) AddFlags(fs *pflag.FlagSet) {
fs.Var(resource.NewQuantityFlagValue(&s.NodeMemory), "node-memory", "The amount of memory (in bytes) provisioned on each node")
fs.StringVar(&s.ClusterName, "cluster-name", s.ClusterName, "The instance prefix for the cluster")
fs.BoolVar(&s.EnableProfiling, "profiling", true, "Enable profiling via web interface host:port/debug/pprof/")
fs.StringVar(&s.ClusterClassB, "cluster-class-b", "10.244", "Class B network address for Pods in cluster.")
fs.BoolVar(&s.AllocateNodeCIDRs, "allocate-node-cidrs", false, "Should CIDRs for Pods be allocated and set on the cloud provider.")
fs.StringVar(&s.Master, "master", s.Master, "The address of the Kubernetes API server (overrides any value in kubeconfig)")
fs.StringVar(&s.Kubeconfig, "kubeconfig", s.Kubeconfig, "Path to kubeconfig file with authorization and master location information.")
}
@ -226,7 +230,7 @@ func (s *CMServer) Run(_ []string) error {
nodeController := nodecontroller.NewNodeController(cloud, s.MinionRegexp, s.MachineList, nodeResources,
kubeClient, s.RegisterRetryCount, s.PodEvictionTimeout, util.NewTokenBucketRateLimiter(s.DeletingPodsQps, s.DeletingPodsBurst),
s.NodeMonitorGracePeriod, s.NodeStartupGracePeriod, s.NodeMonitorPeriod, s.ClusterName)
s.NodeMonitorGracePeriod, s.NodeStartupGracePeriod, s.NodeMonitorPeriod, s.ClusterName, s.ClusterClassB, s.AllocateNodeCIDRs)
nodeController.Run(s.NodeSyncPeriod, s.SyncNodeList)
serviceController := servicecontroller.New(cloud, kubeClient, s.ClusterName)

2
cmd/kubernetes/kubernetes.go
View File

@ -132,7 +132,7 @@ func runControllerManager(machineList []string, cl *client.Client, nodeMilliCPU,
const nodeSyncPeriod = 10 * time.Second
nodeController := nodecontroller.NewNodeController(
nil, "", machineList, nodeResources, cl, 10, 5*time.Minute, util.NewTokenBucketRateLimiter(*deletingPodsQps, *deletingPodsBurst), 40*time.Second, 60*time.Second, 5*time.Second, "")
nil, "", machineList, nodeResources, cl, 10, 5*time.Minute, util.NewTokenBucketRateLimiter(*deletingPodsQps, *deletingPodsBurst), 40*time.Second, 60*time.Second, 5*time.Second, "", "", false)
nodeController.Run(nodeSyncPeriod, true)
serviceController := servicecontroller.New(nil, cl, "kubernetes")

4
pkg/cloudprovider/gce/gce.go
View File

@ -53,6 +53,8 @@ type GCECloud struct {
projectID string
zone string
instanceID string
// We assume here that nodes and master are in the same network. TODO(cjcullen) Fix it.
networkName string
// Used for accessing the metadata server
@ -243,6 +245,7 @@ func waitForOp(op *compute.Operation, getOperation func() (*compute.Operation, e
pollOp := op
for pollOp.Status != "DONE" {
var err error
// TODO: add some backoff here.
time.Sleep(time.Second)
pollOp, err = getOperation()
if err != nil {
@ -569,6 +572,7 @@ func (gce *GCECloud) Configure(name string, spec *api.NodeSpec) error {
}
return ErrMetadataConflict
}
// We are setting the metadata, so they can be picked-up by the configure-vm.sh script to start docker with the given CIDR for Pods.
instance.Metadata.Items = append(instance.Metadata.Items,
&compute.MetadataItems{
Key: podCIDRMetadataKey,

72
pkg/cloudprovider/nodecontroller/nodecontroller.go
View File

@ -22,6 +22,7 @@ import (
"net"
"strings"
"sync"
"sync/atomic"
"time"
"github.com/GoogleCloudPlatform/kubernetes/pkg/api"
@ -88,6 +89,8 @@ type NodeController struct {
// TODO: Change node status monitor to watch based.
nodeMonitorPeriod time.Duration
clusterName string
clusterClassB string
allocateNodeCIDRs bool
// Method for easy mocking in unittest.
lookupIP func(host string) ([]net.IP, error)
now func() util.Time
@ -106,7 +109,9 @@ func NewNodeController(
nodeMonitorGracePeriod time.Duration,
nodeStartupGracePeriod time.Duration,
nodeMonitorPeriod time.Duration,
clusterName string) *NodeController {
clusterName string,
clusterClassB string,
allocateNodeCIDRs bool) *NodeController {
eventBroadcaster := record.NewBroadcaster()
recorder := eventBroadcaster.NewRecorder(api.EventSource{Component: "controllermanager"})
if kubeClient != nil {
@ -132,24 +137,26 @@ func NewNodeController(
lookupIP: net.LookupIP,
now: util.Now,
clusterName: clusterName,
clusterClassB: clusterClassB,
allocateNodeCIDRs: allocateNodeCIDRs,
}
}
// Generates num pod CIDRs that could be assigned to nodes.
func generateCIDRs(num int) util.StringSet {
func (nc *NodeController) generateCIDRs(num int) util.StringSet {
res := util.NewStringSet()
for i := 0; i < num; i++ {
// TODO: Make the CIDRs configurable.
res.Insert(fmt.Sprintf("10.244.%v.0/24", i))
res.Insert(fmt.Sprintf("%v.%v.0/24", nc.clusterClassB, i))
}
return res
}
// For each node from newNodes, finds its current spec in registeredNodes.
// If it is not there, it gets a new valid CIDR assigned.
func reconcilePodCIDRs(newNodes, registeredNodes *api.NodeList) *api.NodeList {
func (nc *NodeController) reconcilePodCIDRs(newNodes, registeredNodes *api.NodeList) *api.NodeList {
registeredCIDRs := make(map[string]string)
availableCIDRs := generateCIDRs(len(newNodes.Items) + len(registeredNodes.Items))
availableCIDRs := nc.generateCIDRs(len(newNodes.Items) + len(registeredNodes.Items))
for _, node := range registeredNodes.Items {
registeredCIDRs[node.Name] = node.Spec.PodCIDR
availableCIDRs.Delete(node.Spec.PodCIDR)
@ -220,8 +227,8 @@ func (nc *NodeController) Run(period time.Duration, syncNodeList bool) {
if nodes, err = nc.populateAddresses(nodes); err != nil {
glog.Errorf("Error getting nodes ips: %v", err)
}
if nc.isRunningCloudProvider() {
reconcilePodCIDRs(nodes, &api.NodeList{})
if nc.isRunningCloudProvider() && nc.allocateNodeCIDRs {
nc.reconcilePodCIDRs(nodes, &api.NodeList{})
}
if err := nc.registerNodes(nodes, nc.registerRetryCount, period); err != nil {
glog.Errorf("Error registering node list %+v: %v", nodes, err)
@ -249,38 +256,37 @@ func (nc *NodeController) registerNodes(nodes *api.NodeList, retryCount int, ret
if len(nodes.Items) == 0 {
return nil
}
registered := util.NewStringSet()
nodes = nc.canonicalizeName(nodes)
for i := 0; i < retryCount; i++ {
toRegister := util.NewStringSet()
var wg sync.WaitGroup
wg.Add(len(nodes.Items))
var successfullyRegistered int32 = 0
for i := range nodes.Items {
go func(node *api.Node) {
node := &nodes.Items[i]
if !toRegister.Has(node.Name) {
wg.Add(1)
toRegister.Insert(node.Name)
go func(n *api.Node) {
defer wg.Done()
if registered.Has(node.Name) {
return
for i := 0; i < retryCount; i++ {
if nc.isRunningCloudProvider() && nc.allocateNodeCIDRs {
nc.configureNodeCIDR(n)
}
if nc.isRunningCloudProvider() {
nc.configureNodeCIDR(node)
}
_, err := nc.kubeClient.Nodes().Create(node)
_, err := nc.kubeClient.Nodes().Create(n)
if err == nil || apierrors.IsAlreadyExists(err) {
registered.Insert(node.Name)
glog.Infof("Registered node in registry: %s", node.Name)
glog.Infof("Registered node in registry: %v", n.Name)
atomic.AddInt32(&successfullyRegistered, 1)
return
} else {
glog.Errorf("Error registering node %s, retrying: %s", node.Name, err)
}
}(&nodes.Items[i])
}
wg.Wait()
if registered.Len() == len(nodes.Items) {
glog.Infof("Successfully registered all nodes")
return nil
glog.Errorf("Error registering node %v (retries left: %v): %v", n.Name, retryCount-i-1, err)
}
time.Sleep(retryInterval)
}
if registered.Len() != len(nodes.Items) {
glog.Errorf("Unable to register node %v", n.Name)
}(node)
}
}
wg.Wait()
if int32(toRegister.Len()) != atomic.LoadInt32(&successfullyRegistered) {
return ErrRegistration
} else {
return nil
@ -302,7 +308,9 @@ func (nc *NodeController) syncCloudNodes() error {
node := nodes.Items[i]
nodeMap[node.Name] = &node
}
reconcilePodCIDRs(matches, nodes)
if nc.allocateNodeCIDRs {
nc.reconcilePodCIDRs(matches, nodes)
}
var wg sync.WaitGroup
wg.Add(len(matches.Items))
// Create nodes which have been created in cloud, but not in kubernetes cluster
@ -320,7 +328,9 @@ func (nc *NodeController) syncCloudNodes() error {
return
}
node.Status.Addresses = nodeList.Items[0].Status.Addresses
if nc.allocateNodeCIDRs {
nc.configureNodeCIDR(node)
}
glog.Infof("Create node in registry: %s", node.Name)
_, err = nc.kubeClient.Nodes().Create(node)
if err != nil {
@ -337,7 +347,9 @@ func (nc *NodeController) syncCloudNodes() error {
for nodeID := range nodeMap {
go func(nodeID string) {
defer wg.Done()
if nc.allocateNodeCIDRs {
nc.unassignNodeCIDR(nodeID)
}
glog.Infof("Delete node from registry: %s", nodeID)
err = nc.kubeClient.Nodes().Delete(nodeID)
if err != nil {

26
pkg/cloudprovider/nodecontroller/nodecontroller_test.go
View File

@ -21,6 +21,7 @@ import (
"fmt"
"reflect"
"sort"
"sync"
"testing"
"time"
@ -59,6 +60,9 @@ type FakeNodeHandler struct {
UpdatedNodes []*api.Node
UpdatedNodeStatuses []*api.Node
RequestCount int
// Synchronization
createLock sync.Mutex
}
func (c *FakeNodeHandler) Nodes() client.NodeInterface {
@ -66,7 +70,11 @@ func (c *FakeNodeHandler) Nodes() client.NodeInterface {
}
func (m *FakeNodeHandler) Create(node *api.Node) (*api.Node, error) {
defer func() { m.RequestCount++ }()
m.createLock.Lock()
defer func() {
m.RequestCount++
m.createLock.Unlock()
}()
for _, n := range m.Existing {
if n.Name == node.Name {
return nil, apierrors.NewAlreadyExists("Minion", node.Name)
@ -238,7 +246,7 @@ func TestRegisterNodes(t *testing.T) {
nodes.Items = append(nodes.Items, *newNode(machine))
}
nodeController := NewNodeController(nil, "", item.machines, &api.NodeResources{}, item.fakeNodeHandler, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
err := nodeController.registerNodes(&nodes, item.retryCount, time.Millisecond)
if !item.expectedFail && err != nil {
t.Errorf("unexpected error: %v", err)
@ -324,7 +332,7 @@ func TestCreateGetStaticNodesWithSpec(t *testing.T) {
}
for _, item := range table {
nodeController := NewNodeController(nil, "", item.machines, &resources, nil, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
nodes, err := nodeController.getStaticNodesWithSpec()
if err != nil {
t.Errorf("unexpected error: %v", err)
@ -386,7 +394,7 @@ func TestCreateGetCloudNodesWithSpec(t *testing.T) {
for _, item := range table {
nodeController := NewNodeController(item.fakeCloud, ".*", nil, &api.NodeResources{}, nil, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
nodes, err := nodeController.getCloudNodesWithSpec()
if err != nil {
t.Errorf("unexpected error: %v", err)
@ -496,7 +504,7 @@ func TestSyncCloudNodes(t *testing.T) {
item.fakeNodeHandler.Fake = testclient.NewSimpleFake()
}
nodeController := NewNodeController(item.fakeCloud, item.matchRE, nil, &api.NodeResources{}, item.fakeNodeHandler, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
if err := nodeController.syncCloudNodes(); err != nil {
t.Errorf("unexpected error: %v", err)
}
@ -580,7 +588,7 @@ func TestSyncCloudNodesEvictPods(t *testing.T) {
item.fakeNodeHandler.Fake = testclient.NewSimpleFake()
}
nodeController := NewNodeController(item.fakeCloud, item.matchRE, nil, &api.NodeResources{}, item.fakeNodeHandler, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
if err := nodeController.syncCloudNodes(); err != nil {
t.Errorf("unexpected error: %v", err)
}
@ -620,7 +628,7 @@ func TestPopulateNodeAddresses(t *testing.T) {
for _, item := range table {
nodeController := NewNodeController(item.fakeCloud, ".*", nil, nil, nil, 10, time.Minute,
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
util.NewFakeRateLimiter(), testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
result, err := nodeController.populateAddresses(item.nodes)
// In case of IP querying error, we should continue.
if err != nil {
@ -820,7 +828,7 @@ func TestMonitorNodeStatusEvictPods(t *testing.T) {
for _, item := range table {
nodeController := NewNodeController(nil, "", []string{"node0"}, nil, item.fakeNodeHandler, 10,
evictionTimeout, util.NewFakeRateLimiter(), testNodeMonitorGracePeriod,
testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
nodeController.now = func() util.Time { return fakeNow }
if err := nodeController.monitorNodeStatus(); err != nil {
t.Errorf("unexpected error: %v", err)
@ -1022,7 +1030,7 @@ func TestMonitorNodeStatusUpdateStatus(t *testing.T) {
for _, item := range table {
nodeController := NewNodeController(nil, "", []string{"node0"}, nil, item.fakeNodeHandler, 10, 5*time.Minute, util.NewFakeRateLimiter(),
testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "")
testNodeMonitorGracePeriod, testNodeStartupGracePeriod, testNodeMonitorPeriod, "", "", false)
nodeController.now = func() util.Time { return fakeNow }
if err := nodeController.monitorNodeStatus(); err != nil {
t.Errorf("unexpected error: %v", err)