cmd/k8s-operator,k8s-operator: optionally serve tailscaled metrics on Pod IP (#11699)

Adds a new .spec.metrics field to ProxyClass to allow users to optionally serve
client metrics (tailscaled --debug) on <Pod-IP>:9001.
Metrics cannot currently be enabled for proxies that egress traffic to tailnet
and for Ingress proxies with tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation
(because they currently forward all cluster traffic to their respective backends).

The assumption is that users will want to have these metrics enabled
continuously to be able to monitor proxy behaviour (as opposed to enabling
them temporarily for debugging). Hence we expose them on Pod IP to make it
easier to consume them i.e via Prometheus PodMonitor.

Updates tailscale/tailscale#11292

Signed-off-by: Irbe Krumina <irbe@tailscale.com>

cmd/k8s-operator/deploy/crds/tailscale.com_proxyclasses.yaml

@@ -37,9 +37,16 @@ spec:
             spec:
               description: Specification of the desired state of the ProxyClass resource. https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status
               type: object
-              required:
-                - statefulSet
               properties:
+                metrics:
+                  description: Configuration for proxy metrics. Metrics are currently not supported for egress proxies and for Ingress proxies that have been configured with tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation.
+                  type: object
+                  required:
+                    - enable
+                  properties:
+                    enable:
+                      description: Setting enable to true will make the proxy serve Tailscale metrics at <pod-ip>:9001/debug/metrics. Defaults to false.
+                      type: boolean
                 statefulSet:
                   description: Configuration parameters for the proxy's StatefulSet. Tailscale Kubernetes operator deploys a StatefulSet for each of the user configured proxies (Tailscale Ingress, Tailscale Service, Connector).
                   type: object

cmd/k8s-operator/deploy/examples/proxyclass.yaml

@@ -3,6 +3,8 @@ kind: ProxyClass
 metadata:
   name: prod
 spec:
+  metrics:
+    enable: true
   statefulSet:
     annotations:
       platform-component: infra
@@ -10,6 +12,6 @@ spec:
       labels:
         team: eng
       nodeSelector:
-        beta.kubernetes.io/os: "linux"
+        kubernetes.io/os: "linux"
       imagePullSecrets:
       - name: "foo"

cmd/k8s-operator/deploy/manifests/operator.yaml

@@ -193,6 +193,15 @@ spec:
                     spec:
                         description: Specification of the desired state of the ProxyClass resource. https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status
                         properties:
+                            metrics:
+                                description: Configuration for proxy metrics. Metrics are currently not supported for egress proxies and for Ingress proxies that have been configured with tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation.
+                                properties:
+                                    enable:
+                                        description: Setting enable to true will make the proxy serve Tailscale metrics at <pod-ip>:9001/debug/metrics. Defaults to false.
+                                        type: boolean
+                                required:
+                                    - enable
+                                type: object
                             statefulSet:
                                 description: Configuration parameters for the proxy's StatefulSet. Tailscale Kubernetes operator deploys a StatefulSet for each of the user configured proxies (Tailscale Ingress, Tailscale Service, Connector).
                                 properties:
@@ -1157,8 +1166,6 @@ spec:
                                                 type: array
                                         type: object
                                 type: object
-                        required:
-                            - statefulSet
                         type: object
                     status:
                         description: Status of the ProxyClass. This is set and managed automatically. https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status

cmd/k8s-operator/deploy/manifests/userspace-proxy.yaml

@@ -20,3 +20,7 @@ spec:
           env:
             - name: TS_USERSPACE
               value: "true"
+            - name: POD_IP
+              valueFrom:
+                fieldRef:
+                  fieldPath: status.podIP

cmd/k8s-operator/sts.go

@@ -582,7 +582,7 @@ func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.S
 	logger.Debugf("reconciling statefulset %s/%s", ss.GetNamespace(), ss.GetName())
 	if sts.ProxyClass != "" {
 		logger.Debugf("configuring proxy resources with ProxyClass %s", sts.ProxyClass)
-		ss = applyProxyClassToStatefulSet(proxyClass, ss)
+		ss = applyProxyClassToStatefulSet(proxyClass, ss, sts, logger)
 	}
 	updateSS := func(s *appsv1.StatefulSet) {
 		s.Spec = ss.Spec
@@ -613,8 +613,28 @@ func mergeStatefulSetLabelsOrAnnots(current, custom map[string]string, managed [
 	return custom
 }
 
-func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet) *appsv1.StatefulSet {
-	if pc == nil || ss == nil || pc.Spec.StatefulSet == nil {
+func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet, stsCfg *tailscaleSTSConfig, logger *zap.SugaredLogger) *appsv1.StatefulSet {
+	if pc == nil || ss == nil {
+		return ss
+	}
+	if pc.Spec.Metrics != nil && pc.Spec.Metrics.Enable {
+		if stsCfg.TailnetTargetFQDN == "" && stsCfg.TailnetTargetIP == "" && !stsCfg.ForwardClusterTrafficViaL7IngressProxy {
+			enableMetrics(ss, pc)
+		} else if stsCfg.ForwardClusterTrafficViaL7IngressProxy {
+			// TODO (irbekrm): fix this
+			// For Ingress proxies that have been configured with
+			// tailscale.com/experimental-forward-cluster-traffic-via-ingress
+			// annotation, all cluster traffic is forwarded to the
+			// Ingress backend(s).
+			logger.Info("ProxyClass specifies that metrics should be enabled, but this is currently not supported for Ingress proxies that accept cluster traffic.")
+		} else {
+			// TODO (irbekrm): fix this
+			// For egress proxies, currently all cluster traffic is forwarded to the tailnet target.
+			logger.Info("ProxyClass specifies that metrics should be enabled, but this is currently not supported for Ingress proxies that accept cluster traffic.")
+		}
+	}
+
+	if pc.Spec.StatefulSet == nil {
 		return ss
 	}
 
@@ -681,6 +701,21 @@ func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet)
 	return ss
 }
 
+func enableMetrics(ss *appsv1.StatefulSet, pc *tsapi.ProxyClass) {
+	for i, c := range ss.Spec.Template.Spec.Containers {
+		if c.Name == "tailscale" {
+			// Serve metrics on on <pod-ip>:9001/debug/metrics. If
+			// we didn't specify Pod IP here, the proxy would, in
+			// some cases, also listen to its Tailscale IP- we don't
+			// want folks to start relying on this side-effect as a
+			// feature.
+			ss.Spec.Template.Spec.Containers[i].Env = append(ss.Spec.Template.Spec.Containers[i].Env, corev1.EnvVar{Name: "TS_TAILSCALED_EXTRA_ARGS", Value: "--debug=$(POD_IP):9001"})
+			ss.Spec.Template.Spec.Containers[i].Ports = append(ss.Spec.Template.Spec.Containers[i].Ports, corev1.ContainerPort{Name: "metrics", Protocol: "TCP", HostPort: 9001, ContainerPort: 9001})
+			break
+		}
+	}
+}
+
 // tailscaledConfig takes a proxy config, a newly generated auth key if
 // generated and a Secret with the previous proxy state and auth key and
 // produces returns tailscaled configuration and a hash of that configuration.

cmd/k8s-operator/sts_test.go

@@ -14,6 +14,7 @@ import (
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
+	"go.uber.org/zap"
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
@@ -51,6 +52,10 @@ func Test_statefulSetNameBase(t *testing.T) {
 }
 
 func Test_applyProxyClassToStatefulSet(t *testing.T) {
+	zl, err := zap.NewDevelopment()
+	if err != nil {
+		t.Fatal(err)
+	}
 	// Setup
 	proxyClassAllOpts := &tsapi.ProxyClass{
 		Spec: tsapi.ProxyClassSpec{
@@ -105,6 +110,12 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 			},
 		},
 	}
+	proxyClassMetrics := &tsapi.ProxyClass{
+		Spec: tsapi.ProxyClassSpec{
+			Metrics: &tsapi.Metrics{Enable: true},
+		},
+	}
+
 	var userspaceProxySS, nonUserspaceProxySS appsv1.StatefulSet
 	if err := yaml.Unmarshal(userspaceProxyYaml, &userspaceProxySS); err != nil {
 		t.Fatalf("unmarshaling userspace proxy template: %v", err)
@@ -149,7 +160,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Spec.InitContainers[0].Env = append(wantSS.Spec.Template.Spec.InitContainers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
 	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
 
-	gotSS := applyProxyClassToStatefulSet(proxyClassAllOpts, nonUserspaceProxySS.DeepCopy())
+	gotSS := applyProxyClassToStatefulSet(proxyClassAllOpts, nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
 		t.Fatalf("Unexpected result applying ProxyClass with all fields set to a StatefulSet for non-userspace proxy (-got +want):\n%s", diff)
 	}
@@ -162,7 +173,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.ObjectMeta.Annotations = mergeMapKeys(wantSS.ObjectMeta.Annotations, proxyClassJustLabels.Spec.StatefulSet.Annotations)
 	wantSS.Spec.Template.Labels = proxyClassJustLabels.Spec.StatefulSet.Pod.Labels
 	wantSS.Spec.Template.Annotations = proxyClassJustLabels.Spec.StatefulSet.Pod.Annotations
-	gotSS = applyProxyClassToStatefulSet(proxyClassJustLabels, nonUserspaceProxySS.DeepCopy())
+	gotSS = applyProxyClassToStatefulSet(proxyClassJustLabels, nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
 		t.Fatalf("Unexpected result applying ProxyClass with custom labels and annotations to a StatefulSet for non-userspace proxy (-got +want):\n%s", diff)
 	}
@@ -183,7 +194,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Spec.Containers[0].SecurityContext = proxyClassAllOpts.Spec.StatefulSet.Pod.TailscaleContainer.SecurityContext
 	wantSS.Spec.Template.Spec.Containers[0].Resources = proxyClassAllOpts.Spec.StatefulSet.Pod.TailscaleContainer.Resources
 	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
-	gotSS = applyProxyClassToStatefulSet(proxyClassAllOpts, userspaceProxySS.DeepCopy())
+	gotSS = applyProxyClassToStatefulSet(proxyClassAllOpts, userspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
 		t.Fatalf("Unexpected result applying ProxyClass with custom labels and annotations to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
 	}
@@ -195,10 +206,19 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.ObjectMeta.Annotations = mergeMapKeys(wantSS.ObjectMeta.Annotations, proxyClassJustLabels.Spec.StatefulSet.Annotations)
 	wantSS.Spec.Template.Labels = proxyClassJustLabels.Spec.StatefulSet.Pod.Labels
 	wantSS.Spec.Template.Annotations = proxyClassJustLabels.Spec.StatefulSet.Pod.Annotations
-	gotSS = applyProxyClassToStatefulSet(proxyClassJustLabels, userspaceProxySS.DeepCopy())
+	gotSS = applyProxyClassToStatefulSet(proxyClassJustLabels, userspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
 		t.Fatalf("Unexpected result applying ProxyClass with custom labels and annotations to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
 	}
+
+	// 5. Test that a ProxyClass with metrics enabled gets correctly applied to a StatefulSet.
+	wantSS = nonUserspaceProxySS.DeepCopy()
+	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env, corev1.EnvVar{Name: "TS_TAILSCALED_EXTRA_ARGS", Value: "--debug=$(POD_IP):9001"})
+	wantSS.Spec.Template.Spec.Containers[0].Ports = []corev1.ContainerPort{{Name: "metrics", Protocol: "TCP", ContainerPort: 9001, HostPort: 9001}}
+	gotSS = applyProxyClassToStatefulSet(proxyClassMetrics, nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
+	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
+		t.Fatalf("Unexpected result applying ProxyClass with metrics enabled to a StatefulSet (-got +want):\n%s", diff)
+	}
 }
 
 func mergeMapKeys(a, b map[string]string) map[string]string {

cmd/k8s-operator/testutils_test.go

@@ -15,6 +15,7 @@ import (
 	"time"
 
 	"github.com/google/go-cmp/cmp"
+	"go.uber.org/zap"
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
@@ -54,6 +55,10 @@ type configOpts struct {
 
 func expectedSTS(t *testing.T, cl client.Client, opts configOpts) *appsv1.StatefulSet {
 	t.Helper()
+	zl, err := zap.NewDevelopment()
+	if err != nil {
+		t.Fatal(err)
+	}
 	tsContainer := corev1.Container{
 		Name:  "tailscale",
 		Image: "tailscale/tailscale",
@@ -205,18 +210,23 @@ func expectedSTS(t *testing.T, cl client.Client, opts configOpts) *appsv1.Statef
 		if err := cl.Get(context.Background(), types.NamespacedName{Name: opts.proxyClass}, proxyClass); err != nil {
 			t.Fatalf("error getting ProxyClass: %v", err)
 		}
-		return applyProxyClassToStatefulSet(proxyClass, ss)
+		return applyProxyClassToStatefulSet(proxyClass, ss, new(tailscaleSTSConfig), zl.Sugar())
 	}
 	return ss
 }
 
 func expectedSTSUserspace(t *testing.T, cl client.Client, opts configOpts) *appsv1.StatefulSet {
 	t.Helper()
+	zl, err := zap.NewDevelopment()
+	if err != nil {
+		t.Fatal(err)
+	}
 	tsContainer := corev1.Container{
 		Name:  "tailscale",
 		Image: "tailscale/tailscale",
 		Env: []corev1.EnvVar{
 			{Name: "TS_USERSPACE", Value: "true"},
+			{Name: "POD_IP", ValueFrom: &corev1.EnvVarSource{FieldRef: &corev1.ObjectFieldSelector{APIVersion: "", FieldPath: "status.podIP"}, ResourceFieldRef: nil, ConfigMapKeyRef: nil, SecretKeyRef: nil}},
 			{Name: "TS_KUBE_SECRET", Value: opts.secretName},
 			{Name: "EXPERIMENTAL_TS_CONFIGFILE_PATH", Value: "/etc/tsconfig/tailscaled"},
 			{Name: "TS_SERVE_CONFIG", Value: "/etc/tailscaled/serve-config"},
@@ -301,7 +311,7 @@ func expectedSTSUserspace(t *testing.T, cl client.Client, opts configOpts) *apps
 		if err := cl.Get(context.Background(), types.NamespacedName{Name: opts.proxyClass}, proxyClass); err != nil {
 			t.Fatalf("error getting ProxyClass: %v", err)
 		}
-		return applyProxyClassToStatefulSet(proxyClass, ss)
+		return applyProxyClassToStatefulSet(proxyClass, ss, new(tailscaleSTSConfig), zl.Sugar())
 	}
 	return ss
 }

k8s-operator/api.md

@@ -330,11 +330,45 @@ Specification of the desired state of the ProxyClass resource. https://git.k8s.i
         </tr>
     </thead>
     <tbody><tr>
+        <td><b><a href="#proxyclassspecmetrics">metrics</a></b></td>
+        <td>object</td>
+        <td>
+          Configuration for proxy metrics. Metrics are currently not supported for egress proxies and for Ingress proxies that have been configured with tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation.<br/>
+        </td>
+        <td>false</td>
+      </tr><tr>
         <td><b><a href="#proxyclassspecstatefulset">statefulSet</a></b></td>
         <td>object</td>
         <td>
           Configuration parameters for the proxy's StatefulSet. Tailscale Kubernetes operator deploys a StatefulSet for each of the user configured proxies (Tailscale Ingress, Tailscale Service, Connector).<br/>
         </td>
+        <td>false</td>
+      </tr></tbody>
+</table>
+
+
+### ProxyClass.spec.metrics
+<sup><sup>[↩ Parent](#proxyclassspec)</sup></sup>
+
+
+
+Configuration for proxy metrics. Metrics are currently not supported for egress proxies and for Ingress proxies that have been configured with tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation.
+
+<table>
+    <thead>
+        <tr>
+            <th>Name</th>
+            <th>Type</th>
+            <th>Description</th>
+            <th>Required</th>
+        </tr>
+    </thead>
+    <tbody><tr>
+        <td><b>enable</b></td>
+        <td>boolean</td>
+        <td>
+          Setting enable to true will make the proxy serve Tailscale metrics at <pod-ip>:9001/debug/metrics. Defaults to false.<br/>
+        </td>
         <td>true</td>
       </tr></tbody>
 </table>

k8s-operator/apis/v1alpha1/types_proxyclass.go

@@ -52,7 +52,14 @@ type ProxyClassSpec struct {
 	// Configuration parameters for the proxy's StatefulSet. Tailscale
 	// Kubernetes operator deploys a StatefulSet for each of the user
 	// configured proxies (Tailscale Ingress, Tailscale Service, Connector).
+	// +optional
 	StatefulSet *StatefulSet `json:"statefulSet"`
+	// Configuration for proxy metrics. Metrics are currently not supported
+	// for egress proxies and for Ingress proxies that have been configured
+	// with tailscale.com/experimental-forward-cluster-traffic-via-ingress
+	// annotation.
+	// +optional
+	Metrics *Metrics `json:"metrics,omitempty"`
 }
 
 type StatefulSet struct {
@@ -131,6 +138,14 @@ type Pod struct {
 	// https://kubernetes.io/docs/reference/kubernetes-api/workload-resources/pod-v1/#scheduling
 	// +optional
 	Tolerations []corev1.Toleration `json:"tolerations,omitempty"`
+	// +optional
+}
+
+type Metrics struct {
+	// Setting enable to true will make the proxy serve Tailscale metrics
+	// at <pod-ip>:9001/debug/metrics.
+	// Defaults to false.
+	Enable bool `json:"enable"`
 }
 
 type Container struct {

k8s-operator/apis/v1alpha1/zz_generated.deepcopy.go

@@ -178,6 +178,21 @@ func (in *Env) DeepCopy() *Env {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *Metrics) DeepCopyInto(out *Metrics) {
+	*out = *in
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Metrics.
+func (in *Metrics) DeepCopy() *Metrics {
+	if in == nil {
+		return nil
+	}
+	out := new(Metrics)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *Pod) DeepCopyInto(out *Pod) {
 	*out = *in
@@ -313,6 +328,11 @@ func (in *ProxyClassSpec) DeepCopyInto(out *ProxyClassSpec) {
 		*out = new(StatefulSet)
 		(*in).DeepCopyInto(*out)
 	}
+	if in.Metrics != nil {
+		in, out := &in.Metrics, &out.Metrics
+		*out = new(Metrics)
+		**out = **in
+	}
 }
 
 // DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ProxyClassSpec.