cmd/{containerboot,k8s-operator},k8s-operator: new options to expose user metrics (#14035)

containerboot:

Adds 3 new environment variables for containerboot, `TS_LOCAL_ADDR_PORT` (default
`"${POD_IP}:9002"`), `TS_METRICS_ENABLED` (default `false`), and `TS_DEBUG_ADDR_PORT`
(default `""`), to configure metrics and debug endpoints. In a follow-up PR, the
health check endpoint will be updated to use the `TS_LOCAL_ADDR_PORT` if
`TS_HEALTHCHECK_ADDR_PORT` hasn't been set.

Users previously only had access to internal debug metrics (which are unstable
and not recommended) via passing the `--debug` flag to tailscaled, but can now
set `TS_METRICS_ENABLED=true` to expose the stable metrics documented at
https://tailscale.com/kb/1482/client-metrics at `/metrics` on the addr/port
specified by `TS_LOCAL_ADDR_PORT`.

Users can also now configure a debug endpoint more directly via the
`TS_DEBUG_ADDR_PORT` environment variable. This is not recommended for production
use, but exposes an internal set of debug metrics and pprof endpoints.

operator:

The `ProxyClass` CRD's `.spec.metrics.enable` field now enables serving the
stable user metrics documented at https://tailscale.com/kb/1482/client-metrics
at `/metrics` on the same "metrics" container port that debug metrics were
previously served on. To smooth the transition for anyone relying on the way the
operator previously consumed this field, we also _temporarily_ serve tailscaled's
internal debug metrics on the same `/debug/metrics` path as before, until 1.82.0
when debug metrics will be turned off by default even if `.spec.metrics.enable`
is set. At that point, anyone who wishes to continue using the internal debug
metrics (not recommended) will need to set the new `ProxyClass` field
`.spec.statefulSet.pod.tailscaleContainer.debug.enable`.

Users who wish to opt out of the transitional behaviour, where enabling
`.spec.metrics.enable` also enables debug metrics, can set
`.spec.statefulSet.pod.tailscaleContainer.debug.enable` to false (recommended).

Separately but related, the operator will no longer specify a host port for the
"metrics" container port definition. This caused scheduling conflicts when k8s
needs to schedule more than one proxy per node, and was not necessary for allowing
the pod's port to be exposed to prometheus scrapers.

Updates #11292

---------

Co-authored-by: Kristoffer Dalby <kristoffer@tailscale.com>
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>

cmd/containerboot/healthz.go

@@ -39,7 +39,7 @@ func runHealthz(addr string, h *healthz) {
 		log.Fatalf("error listening on the provided health endpoint address %q: %v", addr, err)
 	}
 	mux := http.NewServeMux()
-	mux.Handle("/healthz", h)
+	mux.Handle("GET /healthz", h)
 	log.Printf("Running healthcheck endpoint at %s/healthz", addr)
 	hs := &http.Server{Handler: mux}
 

cmd/containerboot/main.go

@@ -178,6 +178,14 @@ func main() {
 	}
 	defer killTailscaled()
 
+	if cfg.LocalAddrPort != "" && cfg.MetricsEnabled {
+		m := &metrics{
+			lc:            client,
+			debugEndpoint: cfg.DebugAddrPort,
+		}
+		runMetrics(cfg.LocalAddrPort, m)
+	}
+
 	if cfg.EnableForwardingOptimizations {
 		if err := client.SetUDPGROForwarding(bootCtx); err != nil {
 			log.Printf("[unexpected] error enabling UDP GRO forwarding: %v", err)

cmd/containerboot/metrics.go

@@ -0,0 +1,91 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build linux
+
+package main
+
+import (
+	"fmt"
+	"io"
+	"log"
+	"net"
+	"net/http"
+
+	"tailscale.com/client/tailscale"
+	"tailscale.com/client/tailscale/apitype"
+)
+
+// metrics is a simple metrics HTTP server, if enabled it forwards requests to
+// the tailscaled's LocalAPI usermetrics endpoint at /localapi/v0/usermetrics.
+type metrics struct {
+	debugEndpoint string
+	lc            *tailscale.LocalClient
+}
+
+func proxy(w http.ResponseWriter, r *http.Request, url string, do func(*http.Request) (*http.Response, error)) {
+	req, err := http.NewRequestWithContext(r.Context(), r.Method, url, r.Body)
+	if err != nil {
+		http.Error(w, fmt.Sprintf("failed to construct request: %s", err), http.StatusInternalServerError)
+		return
+	}
+	req.Header = r.Header.Clone()
+
+	resp, err := do(req)
+	if err != nil {
+		http.Error(w, fmt.Sprintf("failed to proxy request: %s", err), http.StatusInternalServerError)
+		return
+	}
+	defer resp.Body.Close()
+
+	w.WriteHeader(resp.StatusCode)
+	for key, val := range resp.Header {
+		for _, v := range val {
+			w.Header().Add(key, v)
+		}
+	}
+	if _, err := io.Copy(w, resp.Body); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+	}
+}
+
+func (m *metrics) handleMetrics(w http.ResponseWriter, r *http.Request) {
+	localAPIURL := "http://" + apitype.LocalAPIHost + "/localapi/v0/usermetrics"
+	proxy(w, r, localAPIURL, m.lc.DoLocalRequest)
+}
+
+func (m *metrics) handleDebug(w http.ResponseWriter, r *http.Request) {
+	if m.debugEndpoint == "" {
+		http.Error(w, "debug endpoint not configured", http.StatusNotFound)
+		return
+	}
+
+	debugURL := "http://" + m.debugEndpoint + r.URL.Path
+	proxy(w, r, debugURL, http.DefaultClient.Do)
+}
+
+// runMetrics runs a simple HTTP metrics endpoint at <addr>/metrics, forwarding
+// requests to tailscaled's /localapi/v0/usermetrics API.
+//
+// In 1.78.x and 1.80.x, it also proxies debug paths to tailscaled's debug
+// endpoint if configured to ease migration for a breaking change serving user
+// metrics instead of debug metrics on the "metrics" port.
+func runMetrics(addr string, m *metrics) {
+	ln, err := net.Listen("tcp", addr)
+	if err != nil {
+		log.Fatalf("error listening on the provided metrics endpoint address %q: %v", addr, err)
+	}
+
+	mux := http.NewServeMux()
+	mux.HandleFunc("GET /metrics", m.handleMetrics)
+	mux.HandleFunc("/debug/", m.handleDebug) // TODO(tomhjp): Remove for 1.82.0 release.
+
+	log.Printf("Running metrics endpoint at %s/metrics", addr)
+	ms := &http.Server{Handler: mux}
+
+	go func() {
+		if err := ms.Serve(ln); err != nil {
+			log.Fatalf("failed running metrics endpoint: %v", err)
+		}
+	}()
+}

cmd/containerboot/settings.go

@@ -67,11 +67,18 @@ type settings struct {
 	PodIP               string
 	PodIPv4             string
 	PodIPv6             string
-	HealthCheckAddrPort string
+	HealthCheckAddrPort string // TODO(tomhjp): use the local addr/port instead.
+	LocalAddrPort       string
+	MetricsEnabled      bool
+	DebugAddrPort       string
 	EgressSvcsCfgPath   string
 }
 
 func configFromEnv() (*settings, error) {
+	defaultLocalAddrPort := ""
+	if v, ok := os.LookupEnv("POD_IP"); ok && v != "" {
+		defaultLocalAddrPort = fmt.Sprintf("%s:9002", v)
+	}
 	cfg := &settings{
 		AuthKey:                               defaultEnvs([]string{"TS_AUTHKEY", "TS_AUTH_KEY"}, ""),
 		Hostname:                              defaultEnv("TS_HOSTNAME", ""),
@@ -98,6 +105,9 @@ func configFromEnv() (*settings, error) {
 		PodIP:                                 defaultEnv("POD_IP", ""),
 		EnableForwardingOptimizations:         defaultBool("TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS", false),
 		HealthCheckAddrPort:                   defaultEnv("TS_HEALTHCHECK_ADDR_PORT", ""),
+		LocalAddrPort:                         defaultEnv("TS_LOCAL_ADDR_PORT", defaultLocalAddrPort),
+		MetricsEnabled:                        defaultBool("TS_METRICS_ENABLED", false),
+		DebugAddrPort:                         defaultEnv("TS_DEBUG_ADDR_PORT", ""),
 		EgressSvcsCfgPath:                     defaultEnv("TS_EGRESS_SERVICES_CONFIG_PATH", ""),
 	}
 	podIPs, ok := os.LookupEnv("POD_IPS")
@@ -175,6 +185,16 @@ func (s *settings) validate() error {
 			return fmt.Errorf("error parsing TS_HEALTH_CHECK_ADDR_PORT value %q: %w", s.HealthCheckAddrPort, err)
 		}
 	}
+	if s.LocalAddrPort != "" {
+		if _, err := netip.ParseAddrPort(s.LocalAddrPort); err != nil {
+			return fmt.Errorf("error parsing TS_LOCAL_ADDR_PORT value %q: %w", s.LocalAddrPort, err)
+		}
+	}
+	if s.DebugAddrPort != "" {
+		if _, err := netip.ParseAddrPort(s.DebugAddrPort); err != nil {
+			return fmt.Errorf("error parsing TS_DEBUG_ADDR_PORT value %q: %w", s.DebugAddrPort, err)
+		}
+	}
 	return nil
 }
 

cmd/containerboot/tailscaled.go

@@ -90,6 +90,12 @@ func tailscaledArgs(cfg *settings) []string {
 	if cfg.TailscaledConfigFilePath != "" {
 		args = append(args, "--config="+cfg.TailscaledConfigFilePath)
 	}
+	// Once enough proxy versions have been released for all the supported
+	// versions to understand this cfg setting, the operator can stop
+	// setting TS_TAILSCALED_EXTRA_ARGS for the debug flag.
+	if cfg.DebugAddrPort != "" && !strings.Contains(cfg.DaemonExtraArgs, cfg.DebugAddrPort) {
+		args = append(args, "--debug="+cfg.DebugAddrPort)
+	}
 	if cfg.DaemonExtraArgs != "" {
 		args = append(args, strings.Fields(cfg.DaemonExtraArgs)...)
 	}

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

@@ -73,7 +73,12 @@ spec:
                     enable:
                       description: |-
                         Setting enable to true will make the proxy serve Tailscale metrics
-                        at <pod-ip>:9001/debug/metrics.
+                        at <pod-ip>:9002/metrics.
+
+                        In 1.78.x and 1.80.x, this field also serves as the default value for
+                        .spec.statefulSet.pod.tailscaleContainer.debug.enable. From 1.82.0, both
+                        fields will independently default to false.
+
                         Defaults to false.
                       type: boolean
                 statefulSet:
@@ -1249,6 +1254,25 @@ spec:
                           description: Configuration for the proxy container running tailscale.
                           type: object
                           properties:
+                            debug:
+                              description: |-
+                                Configuration for enabling extra debug information in the container.
+                                Not recommended for production use.
+                              type: object
+                              properties:
+                                enable:
+                                  description: |-
+                                    Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/
+                                    and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where
+                                    9001 is a container port named "debug". The endpoints and their responses
+                                    may change in backwards incompatible ways in the future, and should not
+                                    be considered stable.
+
+                                    In 1.78.x and 1.80.x, this setting will default to the value of
+                                    .spec.metrics.enable, and requests to the "metrics" port matching the
+                                    mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,
+                                    this setting will default to false, and no requests will be proxied.
+                                  type: boolean
                             env:
                               description: |-
                                 List of environment variables to set in the container.
@@ -1553,6 +1577,25 @@ spec:
                           description: Configuration for the proxy init container that enables forwarding.
                           type: object
                           properties:
+                            debug:
+                              description: |-
+                                Configuration for enabling extra debug information in the container.
+                                Not recommended for production use.
+                              type: object
+                              properties:
+                                enable:
+                                  description: |-
+                                    Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/
+                                    and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where
+                                    9001 is a container port named "debug". The endpoints and their responses
+                                    may change in backwards incompatible ways in the future, and should not
+                                    be considered stable.
+
+                                    In 1.78.x and 1.80.x, this setting will default to the value of
+                                    .spec.metrics.enable, and requests to the "metrics" port matching the
+                                    mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,
+                                    this setting will default to false, and no requests will be proxied.
+                                  type: boolean
                             env:
                               description: |-
                                 List of environment variables to set in the container.

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

@@ -540,7 +540,12 @@ spec:
                                     enable:
                                         description: |-
                                             Setting enable to true will make the proxy serve Tailscale metrics
-                                            at <pod-ip>:9001/debug/metrics.
+                                            at <pod-ip>:9002/metrics.
+
+                                            In 1.78.x and 1.80.x, this field also serves as the default value for
+                                            .spec.statefulSet.pod.tailscaleContainer.debug.enable. From 1.82.0, both
+                                            fields will independently default to false.
+
                                             Defaults to false.
                                         type: boolean
                                 required:
@@ -1716,6 +1721,25 @@ spec:
                                             tailscaleContainer:
                                                 description: Configuration for the proxy container running tailscale.
                                                 properties:
+                                                    debug:
+                                                        description: |-
+                                                            Configuration for enabling extra debug information in the container.
+                                                            Not recommended for production use.
+                                                        properties:
+                                                            enable:
+                                                                description: |-
+                                                                    Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/
+                                                                    and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where
+                                                                    9001 is a container port named "debug". The endpoints and their responses
+                                                                    may change in backwards incompatible ways in the future, and should not
+                                                                    be considered stable.
+
+                                                                    In 1.78.x and 1.80.x, this setting will default to the value of
+                                                                    .spec.metrics.enable, and requests to the "metrics" port matching the
+                                                                    mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,
+                                                                    this setting will default to false, and no requests will be proxied.
+                                                                type: boolean
+                                                        type: object
                                                     env:
                                                         description: |-
                                                             List of environment variables to set in the container.
@@ -2020,6 +2044,25 @@ spec:
                                             tailscaleInitContainer:
                                                 description: Configuration for the proxy init container that enables forwarding.
                                                 properties:
+                                                    debug:
+                                                        description: |-
+                                                            Configuration for enabling extra debug information in the container.
+                                                            Not recommended for production use.
+                                                        properties:
+                                                            enable:
+                                                                description: |-
+                                                                    Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/
+                                                                    and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where
+                                                                    9001 is a container port named "debug". The endpoints and their responses
+                                                                    may change in backwards incompatible ways in the future, and should not
+                                                                    be considered stable.
+
+                                                                    In 1.78.x and 1.80.x, this setting will default to the value of
+                                                                    .spec.metrics.enable, and requests to the "metrics" port matching the
+                                                                    mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,
+                                                                    this setting will default to false, and no requests will be proxied.
+                                                                type: boolean
+                                                        type: object
                                                     env:
                                                         description: |-
                                                             List of environment variables to set in the container.

cmd/k8s-operator/proxyclass.go

@@ -160,6 +160,10 @@ func (pcr *ProxyClassReconciler) validate(pc *tsapi.ProxyClass) (violations fiel
 						violations = append(violations, field.TypeInvalid(field.NewPath("spec", "statefulSet", "pod", "tailscaleInitContainer", "image"), tc.Image, err.Error()))
 					}
 				}
+
+				if tc.Debug != nil {
+					violations = append(violations, field.TypeInvalid(field.NewPath("spec", "statefulSet", "pod", "tailscaleInitContainer", "debug"), tc.Debug, "debug settings cannot be configured on the init container"))
+				}
 			}
 		}
 	}

cmd/k8s-operator/proxyclass_test.go

@@ -135,3 +135,56 @@ func TestProxyClass(t *testing.T) {
 	expectReconciled(t, pcr, "", "test")
 	expectEvents(t, fr, expectedEvents)
 }
+
+func TestValidateProxyClass(t *testing.T) {
+	for name, tc := range map[string]struct {
+		pc    *tsapi.ProxyClass
+		valid bool
+	}{
+		"empty": {
+			valid: true,
+			pc:    &tsapi.ProxyClass{},
+		},
+		"debug_enabled_for_main_container": {
+			valid: true,
+			pc: &tsapi.ProxyClass{
+				Spec: tsapi.ProxyClassSpec{
+					StatefulSet: &tsapi.StatefulSet{
+						Pod: &tsapi.Pod{
+							TailscaleContainer: &tsapi.Container{
+								Debug: &tsapi.Debug{
+									Enable: true,
+								},
+							},
+						},
+					},
+				},
+			},
+		},
+		"debug_enabled_for_init_container": {
+			valid: false,
+			pc: &tsapi.ProxyClass{
+				Spec: tsapi.ProxyClassSpec{
+					StatefulSet: &tsapi.StatefulSet{
+						Pod: &tsapi.Pod{
+							TailscaleInitContainer: &tsapi.Container{
+								Debug: &tsapi.Debug{
+									Enable: true,
+								},
+							},
+						},
+					},
+				},
+			},
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			pcr := &ProxyClassReconciler{}
+			err := pcr.validate(tc.pc)
+			valid := err == nil
+			if valid != tc.valid {
+				t.Errorf("expected valid=%v, got valid=%v, err=%v", tc.valid, valid, err)
+			}
+		})
+	}
+}

cmd/k8s-operator/sts.go

@@ -476,7 +476,7 @@ func newAuthKey(ctx context.Context, tsClient tsClient, tags []string) (string,
 //go:embed deploy/manifests/userspace-proxy.yaml
 var userspaceProxyYaml []byte
 
-func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.SugaredLogger, sts *tailscaleSTSConfig, headlessSvc *corev1.Service, proxySecret, tsConfigHash string, configs map[tailcfg.CapabilityVersion]ipn.ConfigVAlpha) (*appsv1.StatefulSet, error) {
+func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.SugaredLogger, sts *tailscaleSTSConfig, headlessSvc *corev1.Service, proxySecret, tsConfigHash string, _ map[tailcfg.CapabilityVersion]ipn.ConfigVAlpha) (*appsv1.StatefulSet, error) {
 	ss := new(appsv1.StatefulSet)
 	if sts.ServeConfig != nil && sts.ForwardClusterTrafficViaL7IngressProxy != true { // If forwarding cluster traffic via is required we need non-userspace + NET_ADMIN + forwarding
 		if err := yaml.Unmarshal(userspaceProxyYaml, &ss); err != nil {
@@ -666,24 +666,42 @@ func mergeStatefulSetLabelsOrAnnots(current, custom map[string]string, managed [
 	return custom
 }
 
+func debugSetting(pc *tsapi.ProxyClass) bool {
+	if pc == nil ||
+		pc.Spec.StatefulSet == nil ||
+		pc.Spec.StatefulSet.Pod == nil ||
+		pc.Spec.StatefulSet.Pod.TailscaleContainer == nil ||
+		pc.Spec.StatefulSet.Pod.TailscaleContainer.Debug == nil {
+		// This default will change to false in 1.82.0.
+		return pc.Spec.Metrics != nil && pc.Spec.Metrics.Enable
+	}
+
+	return pc.Spec.StatefulSet.Pod.TailscaleContainer.Debug.Enable
+}
+
 func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet, stsCfg *tailscaleSTSConfig, logger *zap.SugaredLogger) *appsv1.StatefulSet {
 	if pc == nil || ss == nil {
 		return ss
 	}
-	if stsCfg != nil && pc.Spec.Metrics != nil && pc.Spec.Metrics.Enable {
-		if stsCfg.TailnetTargetFQDN == "" && stsCfg.TailnetTargetIP == "" && !stsCfg.ForwardClusterTrafficViaL7IngressProxy {
-			enableMetrics(ss)
-		} else if stsCfg.ForwardClusterTrafficViaL7IngressProxy {
+
+	metricsEnabled := pc.Spec.Metrics != nil && pc.Spec.Metrics.Enable
+	debugEnabled := debugSetting(pc)
+	if metricsEnabled || debugEnabled {
+		isEgress := stsCfg != nil && (stsCfg.TailnetTargetFQDN != "" || stsCfg.TailnetTargetIP != "")
+		isForwardingL7Ingress := stsCfg != nil && stsCfg.ForwardClusterTrafficViaL7IngressProxy
+		if isEgress {
 			// 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 {
+			logger.Info("ProxyClass specifies that metrics should be enabled, but this is currently not supported for egress proxies.")
+		} else if isForwardingL7Ingress {
 			// 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.")
+		} else {
+			enableEndpoints(ss, metricsEnabled, debugEnabled)
 		}
 	}
 
@@ -761,16 +779,58 @@ func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet,
 	return ss
 }
 
-func enableMetrics(ss *appsv1.StatefulSet) {
+func enableEndpoints(ss *appsv1.StatefulSet, metrics, debug bool) {
 	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=$(POD_IP):9001"})
-			ss.Spec.Template.Spec.Containers[i].Ports = append(ss.Spec.Template.Spec.Containers[i].Ports, corev1.ContainerPort{Name: "metrics", Protocol: "TCP", HostPort: 9001, ContainerPort: 9001})
+			if debug {
+				ss.Spec.Template.Spec.Containers[i].Env = append(ss.Spec.Template.Spec.Containers[i].Env,
+					// Serve tailscaled's debug metrics on on
+					// <pod-ip>:9001/debug/metrics. If we didn't specify Pod IP
+					// here, the proxy would, in some cases, also listen to its
+					// Tailscale IP- we don't want folks to start relying on this
+					// side-effect as a feature.
+					corev1.EnvVar{
+						Name:  "TS_DEBUG_ADDR_PORT",
+						Value: "$(POD_IP):9001",
+					},
+					// TODO(tomhjp): Can remove this env var once 1.76.x is no
+					// longer supported.
+					corev1.EnvVar{
+						Name:  "TS_TAILSCALED_EXTRA_ARGS",
+						Value: "--debug=$(TS_DEBUG_ADDR_PORT)",
+					},
+				)
+
+				ss.Spec.Template.Spec.Containers[i].Ports = append(ss.Spec.Template.Spec.Containers[i].Ports,
+					corev1.ContainerPort{
+						Name:          "debug",
+						Protocol:      "TCP",
+						ContainerPort: 9001,
+					},
+				)
+			}
+
+			if metrics {
+				ss.Spec.Template.Spec.Containers[i].Env = append(ss.Spec.Template.Spec.Containers[i].Env,
+					// Serve client metrics on <pod-ip>:9002/metrics.
+					corev1.EnvVar{
+						Name:  "TS_LOCAL_ADDR_PORT",
+						Value: "$(POD_IP):9002",
+					},
+					corev1.EnvVar{
+						Name:  "TS_METRICS_ENABLED",
+						Value: "true",
+					},
+				)
+				ss.Spec.Template.Spec.Containers[i].Ports = append(ss.Spec.Template.Spec.Containers[i].Ports,
+					corev1.ContainerPort{
+						Name:          "metrics",
+						Protocol:      "TCP",
+						ContainerPort: 9002,
+					},
+				)
+			}
+
 			break
 		}
 	}

cmd/k8s-operator/sts_test.go

@@ -125,10 +125,26 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 			},
 		},
 	}
-	proxyClassMetrics := &tsapi.ProxyClass{
+
+	proxyClassWithMetricsDebug := func(metrics bool, debug *bool) *tsapi.ProxyClass {
+		return &tsapi.ProxyClass{
 			Spec: tsapi.ProxyClassSpec{
-			Metrics: &tsapi.Metrics{Enable: true},
+				Metrics: &tsapi.Metrics{Enable: metrics},
+				StatefulSet: func() *tsapi.StatefulSet {
+					if debug == nil {
+						return nil
+					}
+
+					return &tsapi.StatefulSet{
+						Pod: &tsapi.Pod{
+							TailscaleContainer: &tsapi.Container{
+								Debug: &tsapi.Debug{Enable: *debug},
 							},
+						},
+					}
+				}(),
+			},
+		}
 	}
 
 	var userspaceProxySS, nonUserspaceProxySS appsv1.StatefulSet
@@ -184,7 +200,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 
 	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)
+		t.Errorf("Unexpected result applying ProxyClass with all fields set to a StatefulSet for non-userspace proxy (-got +want):\n%s", diff)
 	}
 
 	// 2. Test that a ProxyClass with custom labels and annotations for
@@ -197,7 +213,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Annotations = proxyClassJustLabels.Spec.StatefulSet.Pod.Annotations
 	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)
+		t.Errorf("Unexpected result applying ProxyClass with custom labels and annotations to a StatefulSet for non-userspace proxy (-got +want):\n%s", diff)
 	}
 
 	// 3. Test that a ProxyClass with all fields set gets correctly applied
@@ -221,7 +237,7 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Spec.Containers[0].Image = "ghcr.io/my-repo/tailscale:v0.01testsomething"
 	gotSS = applyProxyClassToStatefulSet(proxyClassAllOpts, userspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
-		t.Fatalf("Unexpected result applying ProxyClass with all options to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
+		t.Errorf("Unexpected result applying ProxyClass with all options to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
 	}
 
 	// 4. Test that a ProxyClass with custom labels and annotations gets correctly applied
@@ -233,16 +249,48 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Annotations = proxyClassJustLabels.Spec.StatefulSet.Pod.Annotations
 	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)
+		t.Errorf("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.
+	// 5. Metrics enabled defaults to enabling both metrics and debug.
 	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=$(POD_IP):9001"})
-	wantSS.Spec.Template.Spec.Containers[0].Ports = []corev1.ContainerPort{{Name: "metrics", Protocol: "TCP", ContainerPort: 9001, HostPort: 9001}}
-	gotSS = applyProxyClassToStatefulSet(proxyClassMetrics, nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
+	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env,
+		corev1.EnvVar{Name: "TS_DEBUG_ADDR_PORT", Value: "$(POD_IP):9001"},
+		corev1.EnvVar{Name: "TS_TAILSCALED_EXTRA_ARGS", Value: "--debug=$(TS_DEBUG_ADDR_PORT)"},
+		corev1.EnvVar{Name: "TS_LOCAL_ADDR_PORT", Value: "$(POD_IP):9002"},
+		corev1.EnvVar{Name: "TS_METRICS_ENABLED", Value: "true"},
+	)
+	wantSS.Spec.Template.Spec.Containers[0].Ports = []corev1.ContainerPort{
+		{Name: "debug", Protocol: "TCP", ContainerPort: 9001},
+		{Name: "metrics", Protocol: "TCP", ContainerPort: 9002},
+	}
+	gotSS = applyProxyClassToStatefulSet(proxyClassWithMetricsDebug(true, nil), 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)
+		t.Errorf("Unexpected result applying ProxyClass with metrics enabled to a StatefulSet (-got +want):\n%s", diff)
+	}
+
+	// 6. Enable _just_ metrics by explicitly disabling debug.
+	wantSS = nonUserspaceProxySS.DeepCopy()
+	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env,
+		corev1.EnvVar{Name: "TS_LOCAL_ADDR_PORT", Value: "$(POD_IP):9002"},
+		corev1.EnvVar{Name: "TS_METRICS_ENABLED", Value: "true"},
+	)
+	wantSS.Spec.Template.Spec.Containers[0].Ports = []corev1.ContainerPort{{Name: "metrics", Protocol: "TCP", ContainerPort: 9002}}
+	gotSS = applyProxyClassToStatefulSet(proxyClassWithMetricsDebug(true, ptr.To(false)), nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
+	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
+		t.Errorf("Unexpected result applying ProxyClass with metrics enabled to a StatefulSet (-got +want):\n%s", diff)
+	}
+
+	// 7. Enable _just_ debug without metrics.
+	wantSS = nonUserspaceProxySS.DeepCopy()
+	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env,
+		corev1.EnvVar{Name: "TS_DEBUG_ADDR_PORT", Value: "$(POD_IP):9001"},
+		corev1.EnvVar{Name: "TS_TAILSCALED_EXTRA_ARGS", Value: "--debug=$(TS_DEBUG_ADDR_PORT)"},
+	)
+	wantSS.Spec.Template.Spec.Containers[0].Ports = []corev1.ContainerPort{{Name: "debug", Protocol: "TCP", ContainerPort: 9001}}
+	gotSS = applyProxyClassToStatefulSet(proxyClassWithMetricsDebug(false, ptr.To(true)), nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
+	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
+		t.Errorf("Unexpected result applying ProxyClass with metrics enabled to a StatefulSet (-got +want):\n%s", diff)
 	}
 }
 

k8s-operator/api.md

@@ -146,6 +146,7 @@ _Appears in:_
 | `imagePullPolicy` _[PullPolicy](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.3/#pullpolicy-v1-core)_ | Image pull policy. One of Always, Never, IfNotPresent. Defaults to Always.<br />https://kubernetes.io/docs/reference/kubernetes-api/workload-resources/pod-v1/#image |  | Enum: [Always Never IfNotPresent] <br /> |
 | `resources` _[ResourceRequirements](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.3/#resourcerequirements-v1-core)_ | Container resource requirements.<br />By default Tailscale Kubernetes operator does not apply any resource<br />requirements. The amount of resources required wil depend on the<br />amount of resources the operator needs to parse, usage patterns and<br />cluster size.<br />https://kubernetes.io/docs/reference/kubernetes-api/workload-resources/pod-v1/#resources |  |  |
 | `securityContext` _[SecurityContext](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.3/#securitycontext-v1-core)_ | Container security context.<br />Security context specified here will override the security context by the operator.<br />By default the operator:<br />- sets 'privileged: true' for the init container<br />- set NET_ADMIN capability for tailscale container for proxies that<br />are created for Services or Connector.<br />https://kubernetes.io/docs/reference/kubernetes-api/workload-resources/pod-v1/#security-context |  |  |
+| `debug` _[Debug](#debug)_ | Configuration for enabling extra debug information in the container.<br />Not recommended for production use. |  |  |
 
 
 #### DNSConfig
@@ -248,6 +249,22 @@ _Appears in:_
 | `nameserver` _[NameserverStatus](#nameserverstatus)_ | Nameserver describes the status of nameserver cluster resources. |  |  |
 
 
+#### Debug
+
+
+
+
+
+
+
+_Appears in:_
+- [Container](#container)
+
+| Field | Description | Default | Validation |
+| --- | --- | --- | --- |
+| `enable` _boolean_ | Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/<br />and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where<br />9001 is a container port named "debug". The endpoints and their responses<br />may change in backwards incompatible ways in the future, and should not<br />be considered stable.<br />In 1.78.x and 1.80.x, this setting will default to the value of<br />.spec.metrics.enable, and requests to the "metrics" port matching the<br />mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,<br />this setting will default to false, and no requests will be proxied. |  |  |
+
+
 #### Env
 
 
@@ -309,7 +326,7 @@ _Appears in:_
 
 | Field | Description | Default | Validation |
 | --- | --- | --- | --- |
-| `enable` _boolean_ | Setting enable to true will make the proxy serve Tailscale metrics<br />at <pod-ip>:9001/debug/metrics.<br />Defaults to false. |  |  |
+| `enable` _boolean_ | Setting enable to true will make the proxy serve Tailscale metrics<br />at <pod-ip>:9002/metrics.<br />In 1.78.x and 1.80.x, this field also serves as the default value for<br />.spec.statefulSet.pod.tailscaleContainer.debug.enable. From 1.82.0, both<br />fields will independently default to false.<br />Defaults to false. |  |  |
 
 
 #### Name

k8s-operator/apis/v1alpha1/types_proxyclass.go

@@ -163,7 +163,12 @@ type Pod struct {
 
 type Metrics struct {
 	// Setting enable to true will make the proxy serve Tailscale metrics
-	// at <pod-ip>:9001/debug/metrics.
+	// at <pod-ip>:9002/metrics.
+	//
+	// In 1.78.x and 1.80.x, this field also serves as the default value for
+	// .spec.statefulSet.pod.tailscaleContainer.debug.enable. From 1.82.0, both
+	// fields will independently default to false.
+	//
 	// Defaults to false.
 	Enable bool `json:"enable"`
 }
@@ -209,6 +214,26 @@ type Container struct {
 	// https://kubernetes.io/docs/reference/kubernetes-api/workload-resources/pod-v1/#security-context
 	// +optional
 	SecurityContext *corev1.SecurityContext `json:"securityContext,omitempty"`
+	// Configuration for enabling extra debug information in the container.
+	// Not recommended for production use.
+	// +optional
+	Debug *Debug `json:"debug,omitempty"`
+}
+
+type Debug struct {
+	// Enable tailscaled's HTTP pprof endpoints at <pod-ip>:9001/debug/pprof/
+	// and internal debug metrics endpoint at <pod-ip>:9001/debug/metrics, where
+	// 9001 is a container port named "debug". The endpoints and their responses
+	// may change in backwards incompatible ways in the future, and should not
+	// be considered stable.
+	//
+	// In 1.78.x and 1.80.x, this setting will default to the value of
+	// .spec.metrics.enable, and requests to the "metrics" port matching the
+	// mux pattern /debug/ will be forwarded to the "debug" port. In 1.82.x,
+	// this setting will default to false, and no requests will be proxied.
+	//
+	// +optional
+	Enable bool `json:"enable"`
 }
 
 type Env struct {

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

@@ -163,6 +163,11 @@ func (in *Container) DeepCopyInto(out *Container) {
 		*out = new(corev1.SecurityContext)
 		(*in).DeepCopyInto(*out)
 	}
+	if in.Debug != nil {
+		in, out := &in.Debug, &out.Debug
+		*out = new(Debug)
+		**out = **in
+	}
 }
 
 // DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Container.
@@ -281,6 +286,21 @@ func (in *DNSConfigStatus) DeepCopy() *DNSConfigStatus {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *Debug) DeepCopyInto(out *Debug) {
+	*out = *in
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Debug.
+func (in *Debug) DeepCopy() *Debug {
+	if in == nil {
+		return nil
+	}
+	out := new(Debug)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *Env) DeepCopyInto(out *Env) {
 	*out = *in