metrics: scope metrics to active config, add optional per-host metrics (#6531)

* Add per host config

* Pass host label when option is enabled

* Test per host enabled

* metrics: scope metrics per loaded config

* doc and linter

Signed-off-by: Mohammed Al Sahaf <msaa1990@gmail.com>

* inject the custom registry into the admin handler

Co-Authored-By: Dave Henderson <dhenderson@gmail.com>

* remove `TODO` comment

* fixes

Signed-off-by: Mohammed Al Sahaf <msaa1990@gmail.com>

* refactor to delay metrics admin handler provision

Signed-off-by: Mohammed Al Sahaf <msaa1990@gmail.com>

---------

Signed-off-by: Mohammed Al Sahaf <msaa1990@gmail.com>
Co-authored-by: Hussam Almarzooq <me@hussam.io>
Co-authored-by: Dave Henderson <dhenderson@gmail.com>

admin.go

@@ -214,7 +214,7 @@ type AdminPermissions struct {
 
 // newAdminHandler reads admin's config and returns an http.Handler suitable
 // for use in an admin endpoint server, which will be listening on listenAddr.
-func (admin *AdminConfig) newAdminHandler(addr NetworkAddress, remote bool) adminHandler {
+func (admin *AdminConfig) newAdminHandler(addr NetworkAddress, remote bool, ctx Context) adminHandler {
 	muxWrap := adminHandler{mux: http.NewServeMux()}
 
 	// secure the local or remote endpoint respectively
@@ -270,7 +270,6 @@ func (admin *AdminConfig) newAdminHandler(addr NetworkAddress, remote bool) admi
 	// register third-party module endpoints
 	for _, m := range GetModules("admin.api") {
 		router := m.New().(AdminRouter)
-		handlerLabel := m.ID.Name()
 		for _, route := range router.Routes() {
 			addRoute(route.Pattern, handlerLabel, route.Handler)
 		}
@@ -382,7 +381,9 @@ func (admin AdminConfig) allowedOrigins(addr NetworkAddress) []*url.URL {
 // for the admin endpoint exists in cfg, a default one is used, so
 // that there is always an admin server (unless it is explicitly
 // configured to be disabled).
-func replaceLocalAdminServer(cfg *Config) error {
+// Critically note that some elements and functionality of the context
+// may not be ready, e.g. storage. Tread carefully.
+func replaceLocalAdminServer(cfg *Config, ctx Context) error {
 	// always* be sure to close down the old admin endpoint
 	// as gracefully as possible, even if the new one is
 	// disabled -- careful to use reference to the current
@@ -424,7 +425,7 @@ func replaceLocalAdminServer(cfg *Config) error {
 		return err
 	}
 
-	handler := cfg.Admin.newAdminHandler(addr, false)
+	handler := cfg.Admin.newAdminHandler(addr, false, ctx)
 
 	ln, err := addr.Listen(context.TODO(), 0, net.ListenConfig{})
 	if err != nil {
@@ -545,7 +546,7 @@ func replaceRemoteAdminServer(ctx Context, cfg *Config) error {
 
 	// make the HTTP handler but disable Host/Origin enforcement
 	// because we are using TLS authentication instead
-	handler := cfg.Admin.newAdminHandler(addr, true)
+	handler := cfg.Admin.newAdminHandler(addr, true, ctx)
 
 	// create client certificate pool for TLS mutual auth, and extract public keys
 	// so that we can enforce access controls at the application layer

caddy.go

@@ -399,6 +399,7 @@ func unsyncedDecodeAndRun(cfgJSON []byte, allowPersist bool) error {
 func run(newCfg *Config, start bool) (Context, error) {
 	ctx, err := provisionContext(newCfg, start)
 	if err != nil {
+		globalMetrics.configSuccess.Set(0)
 		return ctx, err
 	}
 
@@ -410,6 +411,7 @@ func run(newCfg *Config, start bool) (Context, error) {
 	// some of the other apps at runtime
 	err = ctx.cfg.Admin.provisionAdminRouters(ctx)
 	if err != nil {
+		globalMetrics.configSuccess.Set(0)
 		return ctx, err
 	}
 
@@ -435,9 +437,11 @@ func run(newCfg *Config, start bool) (Context, error) {
 		return nil
 	}()
 	if err != nil {
+		globalMetrics.configSuccess.Set(0)
 		return ctx, err
 	}
-
+	globalMetrics.configSuccess.Set(1)
+	globalMetrics.configSuccessTime.SetToCurrentTime()
 	// now that the user's config is running, finish setting up anything else,
 	// such as remote admin endpoint, config loader, etc.
 	return ctx, finishSettingUp(ctx, ctx.cfg)
@@ -471,6 +475,7 @@ func provisionContext(newCfg *Config, replaceAdminServer bool) (Context, error)
 	ctx, cancel := NewContext(Context{Context: context.Background(), cfg: newCfg})
 	defer func() {
 		if err != nil {
+			globalMetrics.configSuccess.Set(0)
 			// if there were any errors during startup,
 			// we should cancel the new context we created
 			// since the associated config won't be used;
@@ -497,7 +502,7 @@ func provisionContext(newCfg *Config, replaceAdminServer bool) (Context, error)
 
 	// start the admin endpoint (and stop any prior one)
 	if replaceAdminServer {
-		err = replaceLocalAdminServer(newCfg)
+		err = replaceLocalAdminServer(newCfg, ctx)
 		if err != nil {
 			return ctx, fmt.Errorf("starting caddy administration endpoint: %v", err)
 		}

caddyconfig/httpcaddyfile/serveroptions.go

@@ -240,13 +240,13 @@ func unmarshalCaddyfileServerOptions(d *caddyfile.Dispenser) (any, error) {
 			}
 
 		case "metrics":
-			if d.NextArg() {
-				return nil, d.ArgErr()
-			}
-			if nesting := d.Nesting(); d.NextBlock(nesting) {
-				return nil, d.ArgErr()
-			}
 			serverOpts.Metrics = new(caddyhttp.Metrics)
+			for nesting := d.Nesting(); d.NextBlock(nesting); {
+				switch d.Val() {
+				case "per_host":
+					serverOpts.Metrics.PerHost = true
+				}
+			}
 
 		case "trace":
 			if d.NextArg() {

caddytest/integration/caddyfile_adapt/metrics_perhost.caddyfiletest

@@ -0,0 +1,37 @@
+{
+	servers :80 {
+		metrics {
+			per_host
+		}
+	}
+}
+:80 {
+	respond "Hello"
+}
+----------
+{
+	"apps": {
+		"http": {
+			"servers": {
+				"srv0": {
+					"listen": [
+						":80"
+					],
+					"routes": [
+						{
+							"handle": [
+								{
+									"body": "Hello",
+									"handler": "static_response"
+								}
+							]
+						}
+					],
+					"metrics": {
+						"per_host": true
+					}
+				}
+			}
+		}
+	}
+}

context.go

@@ -23,6 +23,8 @@ import (
 	"reflect"
 
 	"github.com/caddyserver/certmagic"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/collectors"
 	"go.uber.org/zap"
 	"go.uber.org/zap/exp/zapslog"
 
@@ -47,6 +49,7 @@ type Context struct {
 	ancestry        []Module
 	cleanupFuncs    []func()                // invoked at every config unload
 	exitFuncs       []func(context.Context) // invoked at config unload ONLY IF the process is exiting (EXPERIMENTAL)
+	metricsRegistry *prometheus.Registry
 }
 
 // NewContext provides a new context derived from the given
@@ -58,7 +61,7 @@ type Context struct {
 // modules which are loaded will be properly unloaded.
 // See standard library context package's documentation.
 func NewContext(ctx Context) (Context, context.CancelFunc) {
-	newCtx := Context{moduleInstances: make(map[string][]Module), cfg: ctx.cfg}
+	newCtx := Context{moduleInstances: make(map[string][]Module), cfg: ctx.cfg, metricsRegistry: prometheus.NewPedanticRegistry()}
 	c, cancel := context.WithCancel(ctx.Context)
 	wrappedCancel := func() {
 		cancel()
@@ -79,6 +82,7 @@ func NewContext(ctx Context) (Context, context.CancelFunc) {
 		}
 	}
 	newCtx.Context = c
+	newCtx.initMetrics()
 	return newCtx, wrappedCancel
 }
 
@@ -97,6 +101,22 @@ func (ctx *Context) Filesystems() FileSystems {
 	return ctx.cfg.filesystems
 }
 
+// Returns the active metrics registry for the context
+// EXPERIMENTAL: This API is subject to change.
+func (ctx *Context) GetMetricsRegistry() *prometheus.Registry {
+	return ctx.metricsRegistry
+}
+
+func (ctx *Context) initMetrics() {
+	ctx.metricsRegistry.MustRegister(
+		collectors.NewBuildInfoCollector(),
+		adminMetrics.requestCount,
+		adminMetrics.requestErrors,
+		globalMetrics.configSuccess,
+		globalMetrics.configSuccessTime,
+	)
+}
+
 // OnExit executes f when the process exits gracefully.
 // The function is only executed if the process is gracefully
 // shut down while this context is active.

metrics.go

@@ -4,30 +4,33 @@ import (
 	"net/http"
 
 	"github.com/prometheus/client_golang/prometheus"
-	"github.com/prometheus/client_golang/prometheus/collectors"
-	"github.com/prometheus/client_golang/prometheus/promauto"
 
 	"github.com/caddyserver/caddy/v2/internal/metrics"
 )
 
 // define and register the metrics used in this package.
 func init() {
-	prometheus.MustRegister(collectors.NewBuildInfoCollector())
-
 	const ns, sub = "caddy", "admin"
-
+	adminMetrics.requestCount = prometheus.NewCounterVec(prometheus.CounterOpts{
-	adminMetrics.requestCount = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "http_requests_total",
 		Help:      "Counter of requests made to the Admin API's HTTP endpoints.",
 	}, []string{"handler", "path", "code", "method"})
-	adminMetrics.requestErrors = promauto.NewCounterVec(prometheus.CounterOpts{
+	adminMetrics.requestErrors = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "http_request_errors_total",
 		Help:      "Number of requests resulting in middleware errors.",
 	}, []string{"handler", "path", "method"})
+	globalMetrics.configSuccess = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "caddy_config_last_reload_successful",
+		Help: "Whether the last configuration reload attempt was successful.",
+	})
+	globalMetrics.configSuccessTime = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "caddy_config_last_reload_success_timestamp_seconds",
+		Help: "Timestamp of the last successful configuration reload.",
+	})
 }
 
 // adminMetrics is a collection of metrics that can be tracked for the admin API.
@@ -36,6 +39,12 @@ var adminMetrics = struct {
 	requestErrors *prometheus.CounterVec
 }{}
 
+// globalMetrics is a collection of metrics that can be tracked for Caddy global state
+var globalMetrics = struct {
+	configSuccess     prometheus.Gauge
+	configSuccessTime prometheus.Gauge
+}{}
+
 // Similar to promhttp.InstrumentHandlerCounter, but upper-cases method names
 // instead of lower-casing them.
 //

modules/caddyhttp/app.go

@@ -347,6 +347,10 @@ func (app *App) Provision(ctx caddy.Context) error {
 		// route handler so that important security checks are done, etc.
 		primaryRoute := emptyHandler
 		if srv.Routes != nil {
+			if srv.Metrics != nil {
+				srv.Metrics.init = sync.Once{}
+				srv.Metrics.httpMetrics = &httpMetrics{}
+			}
 			err := srv.Routes.ProvisionHandlers(ctx, srv.Metrics)
 			if err != nil {
 				return fmt.Errorf("server %s: setting up route handlers: %v", srvName, err)

modules/caddyhttp/metrics.go

@@ -10,15 +10,23 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
 
+	"github.com/caddyserver/caddy/v2"
 	"github.com/caddyserver/caddy/v2/internal/metrics"
 )
 
 // Metrics configures metrics observations.
 // EXPERIMENTAL and subject to change or removal.
-type Metrics struct{}
+type Metrics struct {
+	// Enable per-host metrics. Enabling this option may
+	// incur high-memory consumption, depending on the number of hosts
+	// managed by Caddy.
+	PerHost bool `json:"per_host,omitempty"`
 
-var httpMetrics = struct {
+	init        sync.Once
-	init             sync.Once
+	httpMetrics *httpMetrics `json:"-"`
+}
+
+type httpMetrics struct {
 	requestInFlight  *prometheus.GaugeVec
 	requestCount     *prometheus.CounterVec
 	requestErrors    *prometheus.CounterVec
@@ -26,27 +34,28 @@ var httpMetrics = struct {
 	requestSize      *prometheus.HistogramVec
 	responseSize     *prometheus.HistogramVec
 	responseDuration *prometheus.HistogramVec
-}{
-	init: sync.Once{},
 }
 
-func initHTTPMetrics() {
+func initHTTPMetrics(ctx caddy.Context, metrics *Metrics) {
 	const ns, sub = "caddy", "http"
-
+	registry := ctx.GetMetricsRegistry()
 	basicLabels := []string{"server", "handler"}
-	httpMetrics.requestInFlight = promauto.NewGaugeVec(prometheus.GaugeOpts{
+	if metrics.PerHost {
+		basicLabels = append(basicLabels, "host")
+	}
+	metrics.httpMetrics.requestInFlight = promauto.With(registry).NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "requests_in_flight",
 		Help:      "Number of requests currently handled by this server.",
 	}, basicLabels)
-	httpMetrics.requestErrors = promauto.NewCounterVec(prometheus.CounterOpts{
+	metrics.httpMetrics.requestErrors = promauto.With(registry).NewCounterVec(prometheus.CounterOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "request_errors_total",
 		Help:      "Number of requests resulting in middleware errors.",
 	}, basicLabels)
-	httpMetrics.requestCount = promauto.NewCounterVec(prometheus.CounterOpts{
+	metrics.httpMetrics.requestCount = promauto.With(registry).NewCounterVec(prometheus.CounterOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "requests_total",
@@ -58,28 +67,31 @@ func initHTTPMetrics() {
 	sizeBuckets := prometheus.ExponentialBuckets(256, 4, 8)
 
 	httpLabels := []string{"server", "handler", "code", "method"}
-	httpMetrics.requestDuration = promauto.NewHistogramVec(prometheus.HistogramOpts{
+	if metrics.PerHost {
+		httpLabels = append(httpLabels, "host")
+	}
+	metrics.httpMetrics.requestDuration = promauto.With(registry).NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "request_duration_seconds",
 		Help:      "Histogram of round-trip request durations.",
 		Buckets:   durationBuckets,
 	}, httpLabels)
-	httpMetrics.requestSize = promauto.NewHistogramVec(prometheus.HistogramOpts{
+	metrics.httpMetrics.requestSize = promauto.With(registry).NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "request_size_bytes",
 		Help:      "Total size of the request. Includes body",
 		Buckets:   sizeBuckets,
 	}, httpLabels)
-	httpMetrics.responseSize = promauto.NewHistogramVec(prometheus.HistogramOpts{
+	metrics.httpMetrics.responseSize = promauto.With(registry).NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "response_size_bytes",
 		Help:      "Size of the returned response.",
 		Buckets:   sizeBuckets,
 	}, httpLabels)
-	httpMetrics.responseDuration = promauto.NewHistogramVec(prometheus.HistogramOpts{
+	metrics.httpMetrics.responseDuration = promauto.With(registry).NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: ns,
 		Subsystem: sub,
 		Name:      "response_duration_seconds",
@@ -101,14 +113,15 @@ func serverNameFromContext(ctx context.Context) string {
 type metricsInstrumentedHandler struct {
 	handler string
 	mh      MiddlewareHandler
+	metrics *Metrics
 }
 
-func newMetricsInstrumentedHandler(handler string, mh MiddlewareHandler) *metricsInstrumentedHandler {
+func newMetricsInstrumentedHandler(ctx caddy.Context, handler string, mh MiddlewareHandler, metrics *Metrics) *metricsInstrumentedHandler {
-	httpMetrics.init.Do(func() {
+	metrics.init.Do(func() {
-		initHTTPMetrics()
+		initHTTPMetrics(ctx, metrics)
 	})
 
-	return &metricsInstrumentedHandler{handler, mh}
+	return &metricsInstrumentedHandler{handler, mh, metrics}
 }
 
 func (h *metricsInstrumentedHandler) ServeHTTP(w http.ResponseWriter, r *http.Request, next Handler) error {
@@ -119,7 +132,12 @@ func (h *metricsInstrumentedHandler) ServeHTTP(w http.ResponseWriter, r *http.Re
 	// of a panic
 	statusLabels := prometheus.Labels{"server": server, "handler": h.handler, "method": method, "code": ""}
 
-	inFlight := httpMetrics.requestInFlight.With(labels)
+	if h.metrics.PerHost {
+		labels["host"] = r.Host
+		statusLabels["host"] = r.Host
+	}
+
+	inFlight := h.metrics.httpMetrics.requestInFlight.With(labels)
 	inFlight.Inc()
 	defer inFlight.Dec()
 
@@ -131,13 +149,13 @@ func (h *metricsInstrumentedHandler) ServeHTTP(w http.ResponseWriter, r *http.Re
 	writeHeaderRecorder := ShouldBufferFunc(func(status int, header http.Header) bool {
 		statusLabels["code"] = metrics.SanitizeCode(status)
 		ttfb := time.Since(start).Seconds()
-		httpMetrics.responseDuration.With(statusLabels).Observe(ttfb)
+		h.metrics.httpMetrics.responseDuration.With(statusLabels).Observe(ttfb)
 		return false
 	})
 	wrec := NewResponseRecorder(w, nil, writeHeaderRecorder)
 	err := h.mh.ServeHTTP(wrec, r, next)
 	dur := time.Since(start).Seconds()
-	httpMetrics.requestCount.With(labels).Inc()
+	h.metrics.httpMetrics.requestCount.With(labels).Inc()
 
 	observeRequest := func(status int) {
 		// If the code hasn't been set yet, and we didn't encounter an error, we're
@@ -148,9 +166,9 @@ func (h *metricsInstrumentedHandler) ServeHTTP(w http.ResponseWriter, r *http.Re
 			statusLabels["code"] = metrics.SanitizeCode(status)
 		}
 
-		httpMetrics.requestDuration.With(statusLabels).Observe(dur)
+		h.metrics.httpMetrics.requestDuration.With(statusLabels).Observe(dur)
-		httpMetrics.requestSize.With(statusLabels).Observe(float64(computeApproximateRequestSize(r)))
+		h.metrics.httpMetrics.requestSize.With(statusLabels).Observe(float64(computeApproximateRequestSize(r)))
-		httpMetrics.responseSize.With(statusLabels).Observe(float64(wrec.Size()))
+		h.metrics.httpMetrics.responseSize.With(statusLabels).Observe(float64(wrec.Size()))
 	}
 
 	if err != nil {
@@ -159,7 +177,7 @@ func (h *metricsInstrumentedHandler) ServeHTTP(w http.ResponseWriter, r *http.Re
 			observeRequest(handlerErr.StatusCode)
 		}
 
-		httpMetrics.requestErrors.With(labels).Inc()
+		h.metrics.httpMetrics.requestErrors.With(labels).Inc()
 
 		return err
 	}

modules/caddyhttp/metrics_test.go

@@ -6,9 +6,10 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"strings"
+	"sync"
 	"testing"
 
-	"github.com/prometheus/client_golang/prometheus"
+	"github.com/caddyserver/caddy/v2"
 	"github.com/prometheus/client_golang/prometheus/testutil"
 )
 
@@ -27,10 +28,15 @@ func TestServerNameFromContext(t *testing.T) {
 }
 
 func TestMetricsInstrumentedHandler(t *testing.T) {
+	ctx, _ := caddy.NewContext(caddy.Context{Context: context.Background()})
+	metrics := &Metrics{
+		init:        sync.Once{},
+		httpMetrics: &httpMetrics{},
+	}
 	handlerErr := errors.New("oh noes")
 	response := []byte("hello world!")
 	h := HandlerFunc(func(w http.ResponseWriter, r *http.Request) error {
-		if actual := testutil.ToFloat64(httpMetrics.requestInFlight); actual != 1.0 {
+		if actual := testutil.ToFloat64(metrics.httpMetrics.requestInFlight); actual != 1.0 {
 			t.Errorf("Not same: expected %#v, but got %#v", 1.0, actual)
 		}
 		if handlerErr == nil {
@@ -43,7 +49,7 @@ func TestMetricsInstrumentedHandler(t *testing.T) {
 		return h.ServeHTTP(w, r)
 	})
 
-	ih := newMetricsInstrumentedHandler("bar", mh)
+	ih := newMetricsInstrumentedHandler(ctx, "bar", mh, metrics)
 
 	r := httptest.NewRequest("GET", "/", nil)
 	w := httptest.NewRecorder()
@@ -51,7 +57,7 @@ func TestMetricsInstrumentedHandler(t *testing.T) {
 	if actual := ih.ServeHTTP(w, r, h); actual != handlerErr {
 		t.Errorf("Not same: expected %#v, but got %#v", handlerErr, actual)
 	}
-	if actual := testutil.ToFloat64(httpMetrics.requestInFlight); actual != 0.0 {
+	if actual := testutil.ToFloat64(metrics.httpMetrics.requestInFlight); actual != 0.0 {
 		t.Errorf("Not same: expected %#v, but got %#v", 0.0, actual)
 	}
 
@@ -64,7 +70,7 @@ func TestMetricsInstrumentedHandler(t *testing.T) {
 	mh = middlewareHandlerFunc(func(w http.ResponseWriter, r *http.Request, h Handler) error {
 		return nil
 	})
-	ih = newMetricsInstrumentedHandler("empty", mh)
+	ih = newMetricsInstrumentedHandler(ctx, "empty", mh, metrics)
 	r = httptest.NewRequest("GET", "/", nil)
 	w = httptest.NewRecorder()
 
@@ -83,7 +89,7 @@ func TestMetricsInstrumentedHandler(t *testing.T) {
 		return Error(http.StatusTooManyRequests, nil)
 	})
 
-	ih = newMetricsInstrumentedHandler("foo", mh)
+	ih = newMetricsInstrumentedHandler(ctx, "foo", mh, metrics)
 
 	r = httptest.NewRequest("GET", "/", nil)
 	w = httptest.NewRecorder()
@@ -183,7 +189,183 @@ func TestMetricsInstrumentedHandler(t *testing.T) {
 	caddy_http_request_errors_total{handler="bar",server="UNKNOWN"} 1
 	caddy_http_request_errors_total{handler="foo",server="UNKNOWN"} 1
 	`
-	if err := testutil.GatherAndCompare(prometheus.DefaultGatherer, strings.NewReader(expected),
+	if err := testutil.GatherAndCompare(ctx.GetMetricsRegistry(), strings.NewReader(expected),
+		"caddy_http_request_size_bytes",
+		"caddy_http_response_size_bytes",
+		// caddy_http_request_duration_seconds_sum will vary based on how long the test took to run,
+		// so we check just the _bucket and _count metrics
+		"caddy_http_request_duration_seconds_bucket",
+		"caddy_http_request_duration_seconds_count",
+		"caddy_http_request_errors_total",
+	); err != nil {
+		t.Errorf("received unexpected error: %s", err)
+	}
+}
+
+func TestMetricsInstrumentedHandlerPerHost(t *testing.T) {
+	ctx, _ := caddy.NewContext(caddy.Context{Context: context.Background()})
+	metrics := &Metrics{
+		PerHost:     true,
+		init:        sync.Once{},
+		httpMetrics: &httpMetrics{},
+	}
+	handlerErr := errors.New("oh noes")
+	response := []byte("hello world!")
+	h := HandlerFunc(func(w http.ResponseWriter, r *http.Request) error {
+		if actual := testutil.ToFloat64(metrics.httpMetrics.requestInFlight); actual != 1.0 {
+			t.Errorf("Not same: expected %#v, but got %#v", 1.0, actual)
+		}
+		if handlerErr == nil {
+			w.Write(response)
+		}
+		return handlerErr
+	})
+
+	mh := middlewareHandlerFunc(func(w http.ResponseWriter, r *http.Request, h Handler) error {
+		return h.ServeHTTP(w, r)
+	})
+
+	ih := newMetricsInstrumentedHandler(ctx, "bar", mh, metrics)
+
+	r := httptest.NewRequest("GET", "/", nil)
+	w := httptest.NewRecorder()
+
+	if actual := ih.ServeHTTP(w, r, h); actual != handlerErr {
+		t.Errorf("Not same: expected %#v, but got %#v", handlerErr, actual)
+	}
+	if actual := testutil.ToFloat64(metrics.httpMetrics.requestInFlight); actual != 0.0 {
+		t.Errorf("Not same: expected %#v, but got %#v", 0.0, actual)
+	}
+
+	handlerErr = nil
+	if err := ih.ServeHTTP(w, r, h); err != nil {
+		t.Errorf("Received unexpected error: %v", err)
+	}
+
+	// an empty handler - no errors, no header written
+	mh = middlewareHandlerFunc(func(w http.ResponseWriter, r *http.Request, h Handler) error {
+		return nil
+	})
+	ih = newMetricsInstrumentedHandler(ctx, "empty", mh, metrics)
+	r = httptest.NewRequest("GET", "/", nil)
+	w = httptest.NewRecorder()
+
+	if err := ih.ServeHTTP(w, r, h); err != nil {
+		t.Errorf("Received unexpected error: %v", err)
+	}
+	if actual := w.Result().StatusCode; actual != 200 {
+		t.Errorf("Not same: expected status code %#v, but got %#v", 200, actual)
+	}
+	if actual := w.Result().Header; len(actual) != 0 {
+		t.Errorf("Not empty: expected headers to be empty, but got %#v", actual)
+	}
+
+	// handler returning an error with an HTTP status
+	mh = middlewareHandlerFunc(func(w http.ResponseWriter, r *http.Request, h Handler) error {
+		return Error(http.StatusTooManyRequests, nil)
+	})
+
+	ih = newMetricsInstrumentedHandler(ctx, "foo", mh, metrics)
+
+	r = httptest.NewRequest("GET", "/", nil)
+	w = httptest.NewRecorder()
+
+	if err := ih.ServeHTTP(w, r, nil); err == nil {
+		t.Errorf("expected error to be propagated")
+	}
+
+	expected := `
+	# HELP caddy_http_request_duration_seconds Histogram of round-trip request durations.
+	# TYPE caddy_http_request_duration_seconds histogram
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.005"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.01"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.025"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.05"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.1"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.25"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="0.5"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="1"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="2.5"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="5"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="10"} 1
+	caddy_http_request_duration_seconds_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+	caddy_http_request_duration_seconds_count{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN"} 1
+	# HELP caddy_http_request_size_bytes Total size of the request. Includes body
+	# TYPE caddy_http_request_size_bytes histogram
+	caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+    caddy_http_request_size_bytes_sum{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN"} 23
+    caddy_http_request_size_bytes_count{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+    caddy_http_request_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+    caddy_http_request_size_bytes_sum{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN"} 23
+    caddy_http_request_size_bytes_count{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+	caddy_http_request_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+	caddy_http_request_size_bytes_sum{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN"} 23
+	caddy_http_request_size_bytes_count{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN"} 1
+	# HELP caddy_http_response_size_bytes Size of the returned response.
+	# TYPE caddy_http_response_size_bytes histogram
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+	caddy_http_response_size_bytes_sum{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN"} 12
+	caddy_http_response_size_bytes_count{code="200",handler="bar",host="example.com",method="GET",server="UNKNOWN"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+	caddy_http_response_size_bytes_sum{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN"} 0
+	caddy_http_response_size_bytes_count{code="200",handler="empty",host="example.com",method="GET",server="UNKNOWN"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="256"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="1024"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="4096"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="16384"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="65536"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="262144"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="1.048576e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="4.194304e+06"} 1
+	caddy_http_response_size_bytes_bucket{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN",le="+Inf"} 1
+	caddy_http_response_size_bytes_sum{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN"} 0
+	caddy_http_response_size_bytes_count{code="429",handler="foo",host="example.com",method="GET",server="UNKNOWN"} 1
+	# HELP caddy_http_request_errors_total Number of requests resulting in middleware errors.
+	# TYPE caddy_http_request_errors_total counter
+	caddy_http_request_errors_total{handler="bar",host="example.com",server="UNKNOWN"} 1
+	caddy_http_request_errors_total{handler="foo",host="example.com",server="UNKNOWN"} 1
+	`
+	if err := testutil.GatherAndCompare(ctx.GetMetricsRegistry(), strings.NewReader(expected),
 		"caddy_http_request_size_bytes",
 		"caddy_http_response_size_bytes",
 		// caddy_http_request_duration_seconds_sum will vary based on how long the test took to run,

modules/caddyhttp/routes.go

@@ -314,11 +314,11 @@ func wrapRoute(route Route) Middleware {
 // we need to pull this particular MiddlewareHandler
 // pointer into its own stack frame to preserve it so it
 // won't be overwritten in future loop iterations.
-func wrapMiddleware(_ caddy.Context, mh MiddlewareHandler, metrics *Metrics) Middleware {
+func wrapMiddleware(ctx caddy.Context, mh MiddlewareHandler, metrics *Metrics) Middleware {
 	handlerToUse := mh
 	if metrics != nil {
 		// wrap the middleware with metrics instrumentation
-		handlerToUse = newMetricsInstrumentedHandler(caddy.GetModuleName(mh), mh)
+		handlerToUse = newMetricsInstrumentedHandler(ctx, caddy.GetModuleName(mh), mh, metrics)
 	}
 
 	return func(next Handler) Handler {

modules/metrics/adminmetrics.go

@@ -15,8 +15,11 @@
 package metrics
 
 import (
+	"errors"
 	"net/http"
 
+	"github.com/prometheus/client_golang/prometheus"
+
 	"github.com/caddyserver/caddy/v2"
 )
 
@@ -29,7 +32,11 @@ func init() {
 // is permanently mounted to the admin API endpoint at "/metrics".
 // See the Metrics module for a configurable endpoint that is usable if the
 // Admin API is disabled.
-type AdminMetrics struct{}
+type AdminMetrics struct {
+	registry *prometheus.Registry
+
+	metricsHandler http.Handler
+}
 
 // CaddyModule returns the Caddy module information.
 func (AdminMetrics) CaddyModule() caddy.ModuleInfo {
@@ -39,17 +46,28 @@ func (AdminMetrics) CaddyModule() caddy.ModuleInfo {
 	}
 }
 
+// Provision -
+func (m *AdminMetrics) Provision(ctx caddy.Context) error {
+	m.registry = ctx.GetMetricsRegistry()
+	if m.registry == nil {
+		return errors.New("no metrics registry found")
+	}
+	m.metricsHandler = createMetricsHandler(nil, false, m.registry)
+	return nil
+}
+
 // Routes returns a route for the /metrics endpoint.
 func (m *AdminMetrics) Routes() []caddy.AdminRoute {
-	metricsHandler := createMetricsHandler(nil, false)
+	return []caddy.AdminRoute{{Pattern: "/metrics", Handler: caddy.AdminHandlerFunc(m.serveHTTP)}}
-	h := caddy.AdminHandlerFunc(func(w http.ResponseWriter, r *http.Request) error {
+}
-		metricsHandler.ServeHTTP(w, r)
+
-		return nil
+func (m *AdminMetrics) serveHTTP(w http.ResponseWriter, r *http.Request) error {
-	})
+	m.metricsHandler.ServeHTTP(w, r)
-	return []caddy.AdminRoute{{Pattern: "/metrics", Handler: h}}
+	return nil
 }
 
 // Interface guards
 var (
+	_ caddy.Provisioner = (*AdminMetrics)(nil)
 	_ caddy.AdminRouter = (*AdminMetrics)(nil)
 )

modules/metrics/metrics.go

@@ -15,6 +15,7 @@
 package metrics
 
 import (
+	"errors"
 	"net/http"
 
 	"github.com/prometheus/client_golang/prometheus"
@@ -62,7 +63,11 @@ func (l *zapLogger) Println(v ...any) {
 // Provision sets up m.
 func (m *Metrics) Provision(ctx caddy.Context) error {
 	log := ctx.Logger()
-	m.metricsHandler = createMetricsHandler(&zapLogger{log}, !m.DisableOpenMetrics)
+	registry := ctx.GetMetricsRegistry()
+	if registry == nil {
+		return errors.New("no metrics registry found")
+	}
+	m.metricsHandler = createMetricsHandler(&zapLogger{log}, !m.DisableOpenMetrics, registry)
 	return nil
 }
 
@@ -107,9 +112,9 @@ var (
 	_ caddyfile.Unmarshaler       = (*Metrics)(nil)
 )
 
-func createMetricsHandler(logger promhttp.Logger, enableOpenMetrics bool) http.Handler {
+func createMetricsHandler(logger promhttp.Logger, enableOpenMetrics bool, registry *prometheus.Registry) http.Handler {
-	return promhttp.InstrumentMetricHandler(prometheus.DefaultRegisterer,
+	return promhttp.InstrumentMetricHandler(registry,
-		promhttp.HandlerFor(prometheus.DefaultGatherer, promhttp.HandlerOpts{
+		promhttp.HandlerFor(registry, promhttp.HandlerOpts{
 			// will only log errors if logger is non-nil
 			ErrorLog: logger,