Commit 1732a8b9 authored by Kubernetes Submit Queue's avatar Kubernetes Submit Queue Committed by GitHub

Merge pull request #51562 from nicksardo/gce-attempt-firewall

Automatic merge from submit-queue (batch tested with PRs 51915, 51294, 51562, 51911) GCE: Gracefully handle permission errors when attempting to create firewall rules Purpose of this PR is to raise events from the GCE cloud provider if the GCE service account does not have the permissions necessary to create/update/delete firewall rules. Fixes #51812 **Release note**: ```release-note NONE ``` Example Events: ``` Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 2m 2m 1 service-controller Normal EnsuringLoadBalancer Ensuring load balancer 2m 2m 1 gce-cloudprovider Normal LoadBalancerManualChange Firewall change required by network admin: `gcloud compute firewall-rules create aa8a1dd628ddb11e78ce042010a80000 --network https://www.googleapis.com/compute/v1/projects/playground/global/networks/e2e-test-nicksardo --description "{\"kubernetes.io/service-name\":\"default/myechosvc1\", \"kubernetes.io/service-ip\":\"\"}" --allow tcp:9000 --source-ranges 0.0.0.0/0 --target-tags e2e-test-nicksardo-minion --project playground` 2m 2m 1 gce-cloudprovider Normal LoadBalancerManualChange Firewall change required by network admin: `gcloud compute firewall-rules create k8s-1aee5045e658d174-node-hc --network https://www.googleapis.com/compute/v1/projects/playground/global/networks/e2e-test-nicksardo --description "" --allow tcp:10256 --source-ranges 130.211.0.0/22,35.191.0.0/16,209.85.152.0/22,209.85.204.0/22 --target-tags e2e-test-nicksardo-minion --project playground` 1m 1m 1 service-controller Normal EnsuredLoadBalancer Ensured load balancer ```
parents 4692555e 676b95e0
......@@ -77,7 +77,10 @@ go_library(
"//vendor/k8s.io/apiserver/pkg/server/options/encryptionconfig:go_default_library",
"//vendor/k8s.io/apiserver/pkg/storage/value/encrypt/envelope:go_default_library",
"//vendor/k8s.io/client-go/kubernetes:go_default_library",
"//vendor/k8s.io/client-go/kubernetes/scheme:go_default_library",
"//vendor/k8s.io/client-go/kubernetes/typed/core/v1:go_default_library",
"//vendor/k8s.io/client-go/tools/cache:go_default_library",
"//vendor/k8s.io/client-go/tools/record:go_default_library",
"//vendor/k8s.io/client-go/util/flowcontrol:go_default_library",
],
)
......
......@@ -30,10 +30,15 @@ import (
"cloud.google.com/go/compute/metadata"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/wait"
"k8s.io/apiserver/pkg/server/options/encryptionconfig"
"k8s.io/apiserver/pkg/storage/value/encrypt/envelope"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/client-go/kubernetes/scheme"
v1core "k8s.io/client-go/kubernetes/typed/core/v1"
"k8s.io/client-go/tools/record"
"k8s.io/client-go/util/flowcontrol"
"k8s.io/kubernetes/pkg/cloudprovider"
"k8s.io/kubernetes/pkg/controller"
......@@ -100,7 +105,10 @@ type GCECloud struct {
serviceAlpha *computealpha.Service
containerService *container.Service
cloudkmsService *cloudkms.Service
client clientset.Interface
clientBuilder controller.ControllerClientBuilder
eventBroadcaster record.EventBroadcaster
eventRecorder record.EventRecorder
projectID string
region string
localZone string // The zone in which we are running
......@@ -547,6 +555,14 @@ func tryConvertToProjectNames(configProject, configNetworkProject string, servic
// This must be called before utilizing the funcs of gce.ClusterID
func (gce *GCECloud) Initialize(clientBuilder controller.ControllerClientBuilder) {
gce.clientBuilder = clientBuilder
gce.client = clientBuilder.ClientOrDie("cloud-provider")
if gce.OnXPN() {
gce.eventBroadcaster = record.NewBroadcaster()
gce.eventBroadcaster.StartRecordingToSink(&v1core.EventSinkImpl{Interface: v1core.New(gce.client.Core().RESTClient()).Events("")})
gce.eventRecorder = gce.eventBroadcaster.NewRecorder(scheme.Scheme, v1.EventSource{Component: "gce-cloudprovider"})
}
go gce.watchClusterID()
}
......
......@@ -62,7 +62,7 @@ type ClusterID struct {
func (gce *GCECloud) watchClusterID() {
gce.ClusterID = ClusterID{
cfgMapKey: fmt.Sprintf("%v/%v", UIDNamespace, UIDConfigMapName),
client: gce.clientBuilder.ClientOrDie("cloud-provider"),
client: gce.client,
}
mapEventHandler := cache.ResourceEventHandlerFuncs{
......
......@@ -34,8 +34,6 @@ const (
allInstances = "ALL"
)
type lbBalancingMode string
func (gce *GCECloud) ensureInternalLoadBalancer(clusterName, clusterID string, svc *v1.Service, existingFwdRule *compute.ForwardingRule, nodes []*v1.Node) (*v1.LoadBalancerStatus, error) {
nm := types.NamespacedName{Name: svc.Name, Namespace: svc.Namespace}
ports, protocol := getPortsAndProtocol(svc.Spec.Ports)
......@@ -90,12 +88,8 @@ func (gce *GCECloud) ensureInternalLoadBalancer(clusterName, clusterID string, s
glog.V(2).Infof("ensureInternalLoadBalancer(%v): reserved IP %q for the forwarding rule", loadBalancerName, ipToUse)
// Ensure firewall rules if necessary
if gce.OnXPN() {
glog.V(2).Infof("ensureInternalLoadBalancer: cluster is on a cross-project network (XPN) network project %v, compute project %v - skipping firewall creation", gce.networkProjectID, gce.projectID)
} else {
if err = gce.ensureInternalFirewalls(loadBalancerName, ipToUse, clusterID, nm, svc, strconv.Itoa(int(hcPort)), sharedHealthCheck, nodes); err != nil {
return nil, err
}
if err = gce.ensureInternalFirewalls(loadBalancerName, ipToUse, clusterID, nm, svc, strconv.Itoa(int(hcPort)), sharedHealthCheck, nodes); err != nil {
return nil, err
}
expectedFwdRule := &compute.ForwardingRule{
......@@ -141,7 +135,7 @@ func (gce *GCECloud) ensureInternalLoadBalancer(clusterName, clusterID string, s
// Delete the previous internal load balancer resources if necessary
if existingBackendService != nil {
gce.clearPreviousInternalResources(loadBalancerName, existingBackendService, backendServiceName, hcName)
gce.clearPreviousInternalResources(svc, loadBalancerName, existingBackendService, backendServiceName, hcName)
}
// Now that the controller knows the forwarding rule exists, we can release the address.
......@@ -154,7 +148,7 @@ func (gce *GCECloud) ensureInternalLoadBalancer(clusterName, clusterID string, s
return status, nil
}
func (gce *GCECloud) clearPreviousInternalResources(loadBalancerName string, existingBackendService *compute.BackendService, expectedBSName, expectedHCName string) {
func (gce *GCECloud) clearPreviousInternalResources(svc *v1.Service, loadBalancerName string, existingBackendService *compute.BackendService, expectedBSName, expectedHCName string) {
// If a new backend service was created, delete the old one.
if existingBackendService.Name != expectedBSName {
glog.V(2).Infof("clearPreviousInternalResources(%v): expected backend service %q does not match previous %q - deleting backend service", loadBalancerName, expectedBSName, existingBackendService.Name)
......@@ -168,7 +162,7 @@ func (gce *GCECloud) clearPreviousInternalResources(loadBalancerName string, exi
existingHCName := getNameFromLink(existingBackendService.HealthChecks[0])
if existingHCName != expectedHCName {
glog.V(2).Infof("clearPreviousInternalResources(%v): expected health check %q does not match previous %q - deleting health check", loadBalancerName, expectedHCName, existingHCName)
if err := gce.teardownInternalHealthCheckAndFirewall(existingHCName); err != nil {
if err := gce.teardownInternalHealthCheckAndFirewall(svc, existingHCName); err != nil {
glog.Warningf("clearPreviousInternalResources: could not delete existing healthcheck: %v, err: %v", existingHCName, err)
}
}
......@@ -224,12 +218,17 @@ func (gce *GCECloud) ensureInternalLoadBalancerDeleted(clusterName, clusterID st
glog.V(2).Infof("ensureInternalLoadBalancerDeleted(%v): deleting firewall for traffic", loadBalancerName)
if err := gce.DeleteFirewall(loadBalancerName); err != nil {
return err
if isForbidden(err) && gce.OnXPN() {
glog.V(2).Infof("ensureInternalLoadBalancerDeleted(%v): could not delete traffic firewall on XPN cluster. Raising event.", loadBalancerName)
gce.raiseFirewallChangeNeededEvent(svc, FirewallToGCloudDeleteCmd(loadBalancerName, gce.NetworkProjectID()))
} else {
return err
}
}
hcName := makeHealthCheckName(loadBalancerName, clusterID, sharedHealthCheck)
glog.V(2).Infof("ensureInternalLoadBalancerDeleted(%v): deleting health check %v and its firewall", loadBalancerName, hcName)
if err := gce.teardownInternalHealthCheckAndFirewall(hcName); err != nil {
if err := gce.teardownInternalHealthCheckAndFirewall(svc, hcName); err != nil {
return err
}
......@@ -258,7 +257,7 @@ func (gce *GCECloud) teardownInternalBackendService(bsName string) error {
return nil
}
func (gce *GCECloud) teardownInternalHealthCheckAndFirewall(hcName string) error {
func (gce *GCECloud) teardownInternalHealthCheckAndFirewall(svc *v1.Service, hcName string) error {
if err := gce.DeleteHealthCheck(hcName); err != nil {
if isNotFound(err) {
glog.V(2).Infof("teardownInternalHealthCheckAndFirewall(%v): health check does not exist.", hcName)
......@@ -274,13 +273,19 @@ func (gce *GCECloud) teardownInternalHealthCheckAndFirewall(hcName string) error
hcFirewallName := makeHealthCheckFirewallNameFromHC(hcName)
if err := gce.DeleteFirewall(hcFirewallName); err != nil && !isNotFound(err) {
if isForbidden(err) && gce.OnXPN() {
glog.V(2).Infof("teardownInternalHealthCheckAndFirewall(%v): could not delete health check traffic firewall on XPN cluster. Raising Event.", hcName)
gce.raiseFirewallChangeNeededEvent(svc, FirewallToGCloudDeleteCmd(hcFirewallName, gce.NetworkProjectID()))
return nil
}
return fmt.Errorf("failed to delete health check firewall: %v, err: %v", hcFirewallName, err)
}
glog.V(2).Infof("teardownInternalHealthCheckAndFirewall(%v): health check firewall deleted", hcFirewallName)
return nil
}
func (gce *GCECloud) ensureInternalFirewall(fwName, fwDesc string, sourceRanges []string, ports []string, protocol v1.Protocol, nodes []*v1.Node) error {
func (gce *GCECloud) ensureInternalFirewall(svc *v1.Service, fwName, fwDesc string, sourceRanges []string, ports []string, protocol v1.Protocol, nodes []*v1.Node) error {
glog.V(2).Infof("ensureInternalFirewall(%v): checking existing firewall", fwName)
targetTags, err := gce.GetNodeTags(nodeNames(nodes))
if err != nil {
......@@ -308,7 +313,13 @@ func (gce *GCECloud) ensureInternalFirewall(fwName, fwDesc string, sourceRanges
if existingFirewall == nil {
glog.V(2).Infof("ensureInternalFirewall(%v): creating firewall", fwName)
return gce.CreateFirewall(expectedFirewall)
err = gce.CreateFirewall(expectedFirewall)
if err != nil && isForbidden(err) && gce.OnXPN() {
glog.V(2).Infof("ensureInternalFirewall(%v): do not have permission to create firewall rule (on XPN). Raising event.", fwName)
gce.raiseFirewallChangeNeededEvent(svc, FirewallToGCloudCreateCmd(expectedFirewall, gce.NetworkProjectID()))
return nil
}
return err
}
if firewallRuleEqual(expectedFirewall, existingFirewall) {
......@@ -316,7 +327,13 @@ func (gce *GCECloud) ensureInternalFirewall(fwName, fwDesc string, sourceRanges
}
glog.V(2).Infof("ensureInternalFirewall(%v): updating firewall", fwName)
return gce.UpdateFirewall(expectedFirewall)
err = gce.UpdateFirewall(expectedFirewall)
if err != nil && isForbidden(err) && gce.OnXPN() {
glog.V(2).Infof("ensureInternalFirewall(%v): do not have permission to update firewall rule (on XPN). Raising event.", fwName)
gce.raiseFirewallChangeNeededEvent(svc, FirewallToGCloudUpdateCmd(expectedFirewall, gce.NetworkProjectID()))
return nil
}
return err
}
func (gce *GCECloud) ensureInternalFirewalls(loadBalancerName, ipAddress, clusterID string, nm types.NamespacedName, svc *v1.Service, healthCheckPort string, sharedHealthCheck bool, nodes []*v1.Node) error {
......@@ -327,7 +344,7 @@ func (gce *GCECloud) ensureInternalFirewalls(loadBalancerName, ipAddress, cluste
if err != nil {
return err
}
err = gce.ensureInternalFirewall(loadBalancerName, fwDesc, sourceRanges.StringSlice(), ports, protocol, nodes)
err = gce.ensureInternalFirewall(svc, loadBalancerName, fwDesc, sourceRanges.StringSlice(), ports, protocol, nodes)
if err != nil {
return err
}
......@@ -335,7 +352,7 @@ func (gce *GCECloud) ensureInternalFirewalls(loadBalancerName, ipAddress, cluste
// Second firewall is for health checking nodes / services
fwHCName := makeHealthCheckFirewallName(loadBalancerName, clusterID, sharedHealthCheck)
hcSrcRanges := LoadBalancerSrcRanges()
return gce.ensureInternalFirewall(fwHCName, "", hcSrcRanges, []string{healthCheckPort}, v1.ProtocolTCP, nodes)
return gce.ensureInternalFirewall(svc, fwHCName, "", hcSrcRanges, []string{healthCheckPort}, v1.ProtocolTCP, nodes)
}
func (gce *GCECloud) ensureInternalHealthCheck(name string, svcName types.NamespacedName, shared bool, path string, port int32) (*compute.HealthCheck, error) {
......
......@@ -23,6 +23,7 @@ import (
"regexp"
"strings"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/types"
"k8s.io/apimachinery/pkg/util/sets"
......@@ -58,6 +59,43 @@ func getProjectAndZone() (string, string, error) {
return projectID, zone, nil
}
func (gce *GCECloud) raiseFirewallChangeNeededEvent(svc *v1.Service, cmd string) {
msg := fmt.Sprintf("Firewall change required by network admin: `%v`", cmd)
if gce.eventRecorder != nil && svc != nil {
gce.eventRecorder.Event(svc, v1.EventTypeNormal, "LoadBalancerManualChange", msg)
}
}
// FirewallToGCloudCreateCmd generates a gcloud command to create a firewall with specified params
func FirewallToGCloudCreateCmd(fw *compute.Firewall, projectID string) string {
args := firewallToGcloudArgs(fw, projectID)
return fmt.Sprintf("gcloud compute firewall-rules create %v --network %v %v", fw.Name, getNameFromLink(fw.Network), args)
}
// FirewallToGCloudCreateCmd generates a gcloud command to update a firewall to specified params
func FirewallToGCloudUpdateCmd(fw *compute.Firewall, projectID string) string {
args := firewallToGcloudArgs(fw, projectID)
return fmt.Sprintf("gcloud compute firewall-rules update %v %v", fw.Name, args)
}
// FirewallToGCloudCreateCmd generates a gcloud command to delete a firewall to specified params
func FirewallToGCloudDeleteCmd(fwName, projectID string) string {
return fmt.Sprintf("gcloud compute firewall-rules delete %v --project %v", fwName, projectID)
}
func firewallToGcloudArgs(fw *compute.Firewall, projectID string) string {
var allPorts []string
for _, a := range fw.Allowed {
for _, p := range a.Ports {
allPorts = append(allPorts, fmt.Sprintf("%v:%v", a.IPProtocol, p))
}
}
allow := strings.Join(allPorts, ",")
srcRngs := strings.Join(fw.SourceRanges, ",")
targets := strings.Join(fw.TargetTags, ",")
return fmt.Sprintf("--description %q --allow %v --source-ranges %v --target-tags %v --project %v", fw.Description, allow, srcRngs, targets, projectID)
}
// Take a GCE instance 'hostname' and break it down to something that can be fed
// to the GCE API client library. Basically this means reducing 'kubernetes-
// node-2.c.my-proj.internal' to 'kubernetes-node-2' if necessary.
......@@ -150,6 +188,10 @@ func isNotFoundOrInUse(err error) bool {
return isNotFound(err) || isInUsedByError(err)
}
func isForbidden(err error) bool {
return isHTTPErrorCode(err, http.StatusForbidden)
}
func makeGoogleAPINotFoundError(message string) error {
return &googleapi.Error{Code: http.StatusNotFound, Message: message}
}
......@@ -158,10 +200,6 @@ func makeGoogleAPIError(code int, message string) error {
return &googleapi.Error{Code: code, Message: message}
}
func isForbidden(err error) bool {
return isHTTPErrorCode(err, http.StatusForbidden)
}
// TODO(#51665): Remove this once Network Tiers becomes Beta in GCP.
func handleAlphaNetworkTierGetError(err error) (string, error) {
if isForbidden(err) {
......
......@@ -4802,7 +4802,7 @@ func CleanupGCEResources(c clientset.Interface, loadBalancerName, zone string) (
retErr = fmt.Errorf("%v\n%v", retErr, err)
return
}
if err := gceCloud.DeleteExternalTargetPoolAndChecks(loadBalancerName, region, clusterID, hcNames...); err != nil &&
if err := gceCloud.DeleteExternalTargetPoolAndChecks(nil, loadBalancerName, region, clusterID, hcNames...); err != nil &&
!IsGoogleAPIHTTPErrorCode(err, http.StatusNotFound) {
retErr = fmt.Errorf("%v\n%v", retErr, err)
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment