func TestRepairEmpty(t *testing.T) {
_, cidr, _ := net.ParseCIDR("192.168.1.0/24")
previous := ipallocator.NewCIDRRange(cidr)
previous.Allocate(net.ParseIP("192.168.1.10"))
var dst api.RangeAllocation
err := previous.Snapshot(&dst)
if err != nil {
t.Fatal(err)
}
registry := registrytest.NewServiceRegistry()
ipregistry := &mockRangeRegistry{
item: &api.RangeAllocation{
ObjectMeta: api.ObjectMeta{
ResourceVersion: "1",
},
Range: dst.Range,
Data: dst.Data,
},
}
r := NewRepair(0, registry, cidr, ipregistry)
if err := r.RunOnce(); err != nil {
t.Fatal(err)
}
after := ipallocator.NewCIDRRange(cidr)
if err := after.Restore(cidr, ipregistry.updated.Data); err != nil {
t.Fatal(err)
}
if after.Has(net.ParseIP("192.168.1.10")) {
t.Errorf("unexpected ipallocator state: %#v", after)
}
}
func TestRepairWithExisting(t *testing.T) {
_, cidr, _ := net.ParseCIDR("192.168.1.0/24")
previous := ipallocator.NewCIDRRange(cidr)
var dst api.RangeAllocation
err := previous.Snapshot(&dst)
if err != nil {
t.Fatal(err)
}
registry := registrytest.NewServiceRegistry()
registry.List = api.ServiceList{
Items: []api.Service{
{
Spec: api.ServiceSpec{ClusterIP: "192.168.1.1"},
},
{
Spec: api.ServiceSpec{ClusterIP: "192.168.1.100"},
},
{ // outside CIDR, will be dropped
Spec: api.ServiceSpec{ClusterIP: "192.168.0.1"},
},
{ // empty, ignored
Spec: api.ServiceSpec{ClusterIP: ""},
},
{ // duplicate, dropped
Spec: api.ServiceSpec{ClusterIP: "192.168.1.1"},
},
{ // headless
Spec: api.ServiceSpec{ClusterIP: "None"},
},
},
}
ipregistry := &mockRangeRegistry{
item: &api.RangeAllocation{
ObjectMeta: api.ObjectMeta{
ResourceVersion: "1",
},
Range: dst.Range,
Data: dst.Data,
},
}
r := NewRepair(0, registry, cidr, ipregistry)
if err := r.RunOnce(); err != nil {
t.Fatal(err)
}
after := ipallocator.NewCIDRRange(cidr)
if err := after.Restore(cidr, ipregistry.updated.Data); err != nil {
t.Fatal(err)
}
if !after.Has(net.ParseIP("192.168.1.1")) || !after.Has(net.ParseIP("192.168.1.100")) {
t.Errorf("unexpected ipallocator state: %#v", after)
}
if after.Free() != 252 {
t.Errorf("unexpected ipallocator state: %#v", after)
}
}
func (c *Config) createIPAMConf() (*ipam.IPAMConfig, error) {
ipamSubnets := net.IPNet{
IP: c.NodeAddress.IPv6Address.IP(),
Mask: addressing.StateIPv6Mask,
}
ipamConf := &ipam.IPAMConfig{
IPAMConfig: hb.IPAMConfig{
Name: string(ipam.CNIIPAMType),
Subnet: cniTypes.IPNet(ipamSubnets),
Gateway: c.NodeAddress.IPv6Address.IP(),
Routes: []cniTypes.Route{
// IPv6
{
Dst: c.NodeAddress.IPv6Route,
},
{
Dst: addressing.IPv6DefaultRoute,
GW: c.NodeAddress.IPv6Address.IP(),
},
},
},
IPv6Allocator: ipallocator.NewCIDRRange(c.NodeAddress.IPv6AllocRange()),
}
if c.IPv4Enabled {
ipamConf.IPv4Allocator = ipallocator.NewCIDRRange(c.NodeAddress.IPv4AllocRange())
ipamConf.IPAMConfig.Routes = append(ipamConf.IPAMConfig.Routes,
// IPv4
cniTypes.Route{
Dst: c.NodeAddress.IPv4Route,
},
cniTypes.Route{
Dst: addressing.IPv4DefaultRoute,
GW: c.NodeAddress.IPv4Address.IP(),
})
// Reserve the IPv4 router IP in the IPv4 allocation range to ensure
// that we do not hand out the router IP to a container.
err := ipamConf.IPv4Allocator.Allocate(c.NodeAddress.IPv4Address.IP())
if err != nil {
return nil, fmt.Errorf("Unable to reserve IPv4 router address %s: %s",
c.NodeAddress.IPv4Address.String(), err)
}
}
return ipamConf, nil
}
func NewTestREST(t *testing.T, endpoints *api.EndpointsList) (*REST, *registrytest.ServiceRegistry) {
registry := registrytest.NewServiceRegistry()
endpointRegistry := ®istrytest.EndpointRegistry{
Endpoints: endpoints,
}
r := ipallocator.NewCIDRRange(makeIPNet(t))
portRange := utilnet.PortRange{Base: 30000, Size: 1000}
portAllocator := portallocator.NewPortAllocator(portRange)
storage := NewStorage(registry, endpointRegistry, r, portAllocator, nil)
return storage.Service, registry
}
func NewTestREST(t *testing.T, endpoints *api.EndpointsList) (*REST, *registrytest.ServiceRegistry) {
registry := registrytest.NewServiceRegistry()
machines := []string{"foo", "bar", "baz"}
endpointRegistry := ®istrytest.EndpointRegistry{
Endpoints: endpoints,
}
nodeRegistry := registrytest.NewMinionRegistry(machines, api.NodeResources{})
r := ipallocator.NewCIDRRange(makeIPNet(t))
portRange := util.PortRange{Base: 30000, Size: 1000}
portAllocator := portallocator.NewPortAllocator(portRange)
storage := NewStorage(registry, nodeRegistry, endpointRegistry, r, portAllocator, "kubernetes")
return storage, registry
}
// runOnce verifies the state of the cluster IP allocations and returns an error if an unrecoverable problem occurs.
func (c *Repair) runOnce() error {
// TODO: (per smarterclayton) if Get() or ListServices() is a weak consistency read,
// or if they are executed against different leaders,
// the ordering guarantee required to ensure no IP is allocated twice is violated.
// ListServices must return a ResourceVersion higher than the etcd index Get triggers,
// and the release code must not release services that have had IPs allocated but not yet been created
// See #8295
// If etcd server is not running we should wait for some time and fail only then. This is particularly
// important when we start apiserver and etcd at the same time.
var latest *api.RangeAllocation
var err error
err = wait.PollImmediate(time.Second, 10*time.Second, func() (bool, error) {
latest, err = c.alloc.Get()
return err == nil, err
})
if err != nil {
return fmt.Errorf("unable to refresh the service IP block: %v", err)
}
ctx := api.WithNamespace(api.NewDefaultContext(), api.NamespaceAll)
// We explicitly send no resource version, since the resource version
// of 'latest' is from a different collection, it's not comparable to
// the service collection. The caching layer keeps per-collection RVs,
// and this is proper, since in theory the collections could be hosted
// in separate etcd (or even non-etcd) instances.
list, err := c.registry.ListServices(ctx, nil)
if err != nil {
return fmt.Errorf("unable to refresh the service IP block: %v", err)
}
r := ipallocator.NewCIDRRange(c.network)
for _, svc := range list.Items {
if !api.IsServiceIPSet(&svc) {
continue
}
ip := net.ParseIP(svc.Spec.ClusterIP)
if ip == nil {
// cluster IP is broken, reallocate
runtime.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s is not a valid IP; please recreate", svc.Spec.ClusterIP, svc.Name, svc.Namespace))
continue
}
switch err := r.Allocate(ip); err {
case nil:
case ipallocator.ErrAllocated:
// TODO: send event
// cluster IP is broken, reallocate
runtime.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s was assigned to multiple services; please recreate", ip, svc.Name, svc.Namespace))
case ipallocator.ErrNotInRange:
// TODO: send event
// cluster IP is broken, reallocate
runtime.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s is not within the service CIDR %s; please recreate", ip, svc.Name, svc.Namespace, c.network))
case ipallocator.ErrFull:
// TODO: send event
return fmt.Errorf("the service CIDR %v is full; you must widen the CIDR in order to create new services", r)
default:
return fmt.Errorf("unable to allocate cluster IP %s for service %s/%s due to an unknown error, exiting: %v", ip, svc.Name, svc.Namespace, err)
}
}
if err := r.Snapshot(latest); err != nil {
return fmt.Errorf("unable to snapshot the updated service IP allocations: %v", err)
}
if err := c.alloc.CreateOrUpdate(latest); err != nil {
if errors.IsConflict(err) {
return err
}
return fmt.Errorf("unable to persist the updated service IP allocations: %v", err)
}
return nil
}
// RunOnce verifies the state of the cluster IP allocations and returns an error if an unrecoverable problem occurs.
func (c *Repair) RunOnce() error {
// TODO: (per smarterclayton) if Get() or ListServices() is a weak consistency read,
// or if they are executed against different leaders,
// the ordering guarantee required to ensure no IP is allocated twice is violated.
// ListServices must return a ResourceVersion higher than the etcd index Get triggers,
// and the release code must not release services that have had IPs allocated but not yet been created
// See #8295
// If etcd server is not running we should wait for some time and fail only then. This is particularly
// important when we start apiserver and etcd at the same time.
var latest *api.RangeAllocation
var err error
for i := 0; i < 10; i++ {
if latest, err = c.alloc.Get(); err != nil {
time.Sleep(time.Second)
} else {
break
}
}
if err != nil {
return fmt.Errorf("unable to refresh the service IP block: %v", err)
}
ctx := api.WithNamespace(api.NewDefaultContext(), api.NamespaceAll)
list, err := c.registry.ListServices(ctx, labels.Everything(), fields.Everything())
if err != nil {
return fmt.Errorf("unable to refresh the service IP block: %v", err)
}
r := ipallocator.NewCIDRRange(c.network)
for _, svc := range list.Items {
if !api.IsServiceIPSet(&svc) {
continue
}
ip := net.ParseIP(svc.Spec.ClusterIP)
if ip == nil {
// cluster IP is broken, reallocate
util.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s is not a valid IP; please recreate", svc.Spec.ClusterIP, svc.Name, svc.Namespace))
continue
}
switch err := r.Allocate(ip); err {
case nil:
case ipallocator.ErrAllocated:
// TODO: send event
// cluster IP is broken, reallocate
util.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s was assigned to multiple services; please recreate", ip, svc.Name, svc.Namespace))
case ipallocator.ErrNotInRange:
// TODO: send event
// cluster IP is broken, reallocate
util.HandleError(fmt.Errorf("the cluster IP %s for service %s/%s is not within the service CIDR %s; please recreate", ip, svc.Name, svc.Namespace, c.network))
case ipallocator.ErrFull:
// TODO: send event
return fmt.Errorf("the service CIDR %v is full; you must widen the CIDR in order to create new services", r)
default:
return fmt.Errorf("unable to allocate cluster IP %s for service %s/%s due to an unknown error, exiting: %v", ip, svc.Name, svc.Namespace, err)
}
}
err = r.Snapshot(latest)
if err != nil {
return fmt.Errorf("unable to persist the updated service IP allocations: %v", err)
}
if err := c.alloc.CreateOrUpdate(latest); err != nil {
return fmt.Errorf("unable to persist the updated service IP allocations: %v", err)
}
return nil
}
