Drop dependency on github.com/cyberdelia/go-metrics-graphite

The go-metrics-graphite package has been unmaintained for 10+ years,
which is a packaging and supply-chain concern for downstream
distributors (see #1831). Nebula only used its Config struct and Once()
entrypoint, so inline just those (~65 lines) into a local graphite.go,
preserving the upstream BSD-2-Clause copyright notice, and remove the
dependency.

Fixes #1831

Co-Authored-By: Claude <svc-devxp-claude@slack-corp.com>

CHANGELOG.md

@@ -7,6 +7,88 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.11.0] - 2026-07-23
+
+See the [v1.11.0](https://github.com/slackhq/nebula/milestone/25?closed=1) milestone for a complete list of changes.
+
+### Breaking
+
+- Logging has switched from logrus to Go's structured `slog`. Log output changes: levels are upper case
+  (`level=INFO`), trace prints as `level=DEBUG-4`, timestamps are always RFC3339Nano and `logging.timestamp_format`
+  is ignored, and some messages were reworded. Review any log parsing before upgrading. This is also an API break
+  for embedders, as constructors now take a `*slog.Logger`. (#1672, #1734, #1621)
+- `firewall.inbound_action` and `firewall.outbound_action` (used to set reject vs. drop policy) were each being
+  applied to the opposite direction, that is now corrected. This only affects how blocked packets are answered, not
+  which packets the firewall allows or denies. If you set either of these you are getting the behavior of the other
+  one today and likely want to swap them before upgrading. (#1798)
+- On Windows, Nebula now installs WFP PERMIT filters for the nebula adapter and the listener port by default. WFP
+  sits below Windows Defender Firewall, so any WDF inbound rules you rely on for either will no longer apply. Set
+  `tun.windows_bypass_wdf` and `listen.windows_bypass_wdf` to false to leave WDF in charge. (#1710)
+- On Windows, the nebula device is now set to the `private` network category instead of whatever Windows decided,
+  which is usually `Public`. This makes the host firewall less restrictive on the overlay. Set
+  `tun.network_category` to `unset` to keep the old behavior. (#1710)
+- Reject packets for non-TCP now use ICMP code 13, communication administratively prohibited, instead of code 3,
+  port unreachable. Anything keying off the old code needs updating. (#1766, #1768)
+- The SSH debug server's profiling commands are now confined to `sshd.sandbox_dir`, which defaults to
+  `$TMP/nebula-debug`. Relative paths resolve inside it and absolute paths outside it are rejected, so anything
+  scripting `start-cpu-profile`, `save-heap-profile`, or `save-mutex-profile` with a path elsewhere needs the
+  directory set. The directory is not created for you. (#1622)
+
+### Added
+
+- Sign the Windows release binaries. (#1718)
+- Generate IPv6 reject packets, matching the existing IPv4 behavior. (#1766, #1767, #1768)
+- Accept `-` in `nebula-cert` to read from stdin or write to stdout. (#1714)
+- Search for both `config.yml` and `config.yaml` in service and command line modes. (#1717)
+- Add version labels to the Docker/OCI images. (#1772)
+- Rebind the listener and re-query lighthouses on macOS when the underlay network changes, so devices moving
+  between wifi and wired or between networks recover without waiting for dead tunnel detection. Controlled by
+  `listen.rebind_on_network_change` (default `true`, not reloadable). (#1816)
+
+### Changed
+
+- Reload the firewall when the unsafe networks in the certificate change. (#1719)
+- Reconfigure, start, and stop the stats listener on a config reload instead of requiring a restart. (#1670)
+- Update a static host's addresses when they change on reload. (#1713)
+- Don't require a port on ICMP firewall rules. (#1609)
+- Connection track ICMP traffic. (#1602)
+- Return `NODATA` instead of `NXDOMAIN` from the DNS server for a name that exists but has no record of the
+  requested type, so clients that query `AAAA` first (busybox/Alpine) fall through to `A`. (#1668)
+- Record the local host's details in the DNS server. (#1716)
+- Install Windows unsafe routes as link routes. (#1709)
+- Reduce relay handshake log spam, and only log a handshake send error at error level when the remote list
+  changes. (#1733, #1765, #1810)
+- Start, stop, and reload subsystems (DNS, stats, conntrack, ssh, punchy) cleanly without leaking goroutines. (#1640, #1654, #1661, #1667, #1669, #1708, #1806, #1815)
+- `Control` is now safe to stop and wait on from any lifecycle state, and a new `Control.Wait` blocks until nebula
+  has fully stopped and returns the first fatal reader error. Failed starts release the udp sockets and tun fd
+  instead of leaking them. (#1794)
+- Trigger an immediate lighthouse update when reconnecting to or adding a lighthouse instead of waiting for the next update tick. (#1645)
+- Bring the Darwin and OpenBSD tun implementations in line with the other BSDs. (#1703)
+- Update to build against go v1.26. (#1818)
+- Various dependency updates. (#1586, #1587, #1604, #1617, #1618, #1627, #1628, #1629, #1652, #1664, #1665, #1697, #1721, #1732, #1742, #1743, #1750, #1763, #1771, #1782, #1800, #1807)
+
+### Fixed
+
+- Fix a data race on a host's remote address that could send packets to the wrong address during a roam. (#1773)
+- Fix tunnels that could permanently escape connection manager monitoring. (#1752)
+- Fix a crash when reloading the SSH server's trusted keys. (#1787)
+- Fix hostmap corruption when a host has multiple overlay addresses. Each address now gets its own list instead of
+  a single shared chain, which also fixes two latent bugs on the add and makePrimary paths. (#1788, #1790)
+- Apply `remote_allow_list` IPv4 rules to 4-in-6 mapped addresses. (#1786)
+- Don't panic in the DNS server on a short or empty query name. (#1635)
+- Advance the replay window on relayed packets so a relay drops replayed frames instead of re-forwarding them. (#1751)
+- Fix a race in relay state handling. (#1753)
+- Lock replay window updates so concurrent readers can't corrupt it. (#1802)
+- Reject malformed handshakes more reliably, including invalid ed25519 key lengths. (#1601, #1756)
+- Properly handle `closetunnel` packets. (#1638)
+- Fix an IPv6 extension-header length overflow that could make the firewall parse the wrong protocol and ports. (#1789)
+- Fix relay re-establishment when a handshake arrives over a relay entry that a one-sided teardown left
+  `Disestablished`, which silently dropped every send until dead tunnel detection forced a re-handshake. (#1805)
+- Don't build new relay state on a tunnel that was just discarded. (#1796)
+- Don't delete the wrong pending hostinfo in the handshake manager. (#1811)
+- Don't call the packet reader after a UDP error on Darwin. (#1755)
+- Open the FreeBSD tun device non blocking. (#1666)
+
 ## [1.10.3] - 2026-02-06
 
 ### Security

e2e/recovery_timing_test.go

@@ -0,0 +1,136 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula"
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/e2e/router"
+	"github.com/slackhq/nebula/udp"
+)
+
+// TestRecoveryTiming measures how long a tunnel takes to come back after the peer stops accepting our traffic,
+// which is what a laptop waking on a new network looks like from the peer's side: its NAT has no state for where
+// we are now, so everything we send disappears.
+//
+// It is a measurement, not a pass/fail assertion. Recovery is timed to the moment the peer punches back at us,
+// since that is when its NAT opens and the tunnel is usable again.
+//
+//	go test -tags e2e_testing -v -run TestRecoveryTiming ./e2e/
+func TestRecoveryTiming(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		rebind bool
+	}{
+		{"no trigger", false},
+		{"rebind counter", true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			d, lost := measureRecovery(t, tc.rebind)
+			t.Logf("RESULT %-16s recovered in %-9v (%d packets lost)", tc.name, d.Round(time.Millisecond), lost)
+		})
+	}
+}
+
+// measureRecovery returns how long until the peer punched back, and how many of our packets died meanwhile. When
+// rebind is true we call RebindUDPServer once the tunnel goes dark, which is what the darwin network change
+// monitor does and what iOS has always done. When false, nothing tells nebula anything is wrong.
+func measureRecovery(t *testing.T, rebind bool) (time.Duration, int) {
+	t.Helper()
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	lhControl, lhVpnIpNet, lhUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "lh", "10.128.0.1/24", m{
+		"lighthouse": m{"am_lighthouse": true},
+	})
+
+	peerCfg := m{
+		"lighthouse": m{
+			"hosts":    []any{lhVpnIpNet[0].Addr().String()},
+			"interval": 600,
+			"local_allow_list": m{
+				"10.0.0.0/24": true,
+				"::/0":        false,
+			},
+		},
+		"static_host_map": m{
+			lhVpnIpNet[0].Addr().String(): []any{lhUdpAddr.String()},
+		},
+	}
+
+	myControl, myVpnIpNet, myUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.2/24", peerCfg)
+	theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.3/24", peerCfg)
+
+	r := router.NewR(t, lhControl, myControl, theirControl)
+	defer r.RenderFlow()
+	defer func() {
+		lhControl.Stop()
+		myControl.Stop()
+		theirControl.Stop()
+	}()
+
+	lhControl.Start()
+	myControl.Start()
+	theirControl.Start()
+	r.RouteFor(time.Millisecond * 500)
+
+	myControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
+	theirControl.InjectLightHouseAddr(myVpnIpNet[0].Addr(), myUdpAddr)
+
+	myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("establish")))
+	r.RouteFor(time.Second)
+	if myControl.GetHostInfoByVpnAddr(theirVpnIpNet[0].Addr(), false) == nil {
+		t.Fatal("failed to establish the tunnel we are measuring")
+	}
+	r.RouteFor(time.Millisecond * 500)
+
+	// From here the peer's NAT has no state for us, everything we send it disappears
+	start := time.Now()
+	blackholed := 0
+	var recovered time.Duration
+
+	if rebind {
+		myControl.RebindUDPServer()
+	}
+
+	// Keep the tun busy the way someone retrying a stalled connection would
+	stop := make(chan struct{})
+	defer close(stop)
+	go func() {
+		tick := time.NewTicker(time.Millisecond * 200)
+		defer tick.Stop()
+		for {
+			select {
+			case <-stop:
+				return
+			case <-tick.C:
+				myControl.InjectTunPacket(BuildTunUDPPacket(
+					theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("retry")))
+			}
+		}
+	}()
+
+	r.RouteForAllExitFuncOrTimeout(time.Second*30, func(p *udp.Packet, c *nebula.Control) router.ExitType {
+		if c == theirControl && p.From == myControl.GetUDPAddr() {
+			blackholed++
+			return router.Drop
+		}
+
+		// The peer reaching us directly is the moment its NAT opened, whether that is a punch or a handshake
+		if c == myControl && p.From == theirUdpAddr {
+			recovered = time.Since(start)
+			return router.RouteAndExit
+		}
+
+		return router.KeepRouting
+	})
+
+	if recovered == 0 {
+		t.Fatalf("no recovery within 30s (%d packets blackholed)", blackholed)
+	}
+	return recovered, blackholed
+}

e2e/router/router.go

@@ -153,6 +153,9 @@ const (
 	ExitNow ExitType = 1
 	// RouteAndExit routes this packet and exits immediately afterwards
 	RouteAndExit ExitType = 2
+	// Drop discards this packet without delivering it and keeps routing. Use it to simulate a blackhole, such as
+	// a restrictive NAT refusing traffic from an address it has not seen.
+	Drop ExitType = 3
 )
 
 type ExitFunc func(packet *udp.Packet, receiver *nebula.Control) ExitType
@@ -163,7 +166,9 @@ type ExitFunc func(packet *udp.Packet, receiver *nebula.Control) ExitType
 func NewR(t testing.TB, controls ...*nebula.Control) *R {
 	ctx, cancel := context.WithCancel(context.Background())
 
-	if err := os.MkdirAll("mermaid", 0755); err != nil {
+	// t.Name() contains a slash for subtests, so the flow log can land in a nested directory
+	fn := filepath.Join("mermaid", fmt.Sprintf("%s.md", t.Name()))
+	if err := os.MkdirAll(filepath.Dir(fn), 0755); err != nil {
 		panic(err)
 	}
 
@@ -174,7 +179,7 @@ func NewR(t testing.TB, controls ...*nebula.Control) *R {
 		outNat:       make(map[outNatKey]netip.AddrPort),
 		flow:         []flowEntry{},
 		ignoreFlows:  []ignoreFlow{},
-		fn:           filepath.Join("mermaid", fmt.Sprintf("%s.md", t.Name())),
+		fn:           fn,
 		t:            t,
 		cancelRender: cancel,
 	}
@@ -687,6 +692,10 @@ func (r *R) RouteExitFunc(sender *nebula.Control, whatDo ExitFunc) {
 			p.Release()
 			return
 
+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(sender, receiver, p, false)
+
 		case KeepRouting:
 			fp := r.unlockedInjectFlow(sender, receiver, p, false)
 			receiver.InjectUDPPacket(p)
@@ -779,6 +788,10 @@ func (r *R) RouteForAllExitFuncOrTimeout(timeout time.Duration, whatDo ExitFunc)
 			p.Release()
 			return true
 
+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(cm[x], receiver, p, false)
+
 		case KeepRouting:
 			fp := r.unlockedInjectFlow(cm[x], receiver, p, false)
 			receiver.InjectUDPPacket(p)
@@ -884,6 +897,10 @@ func (r *R) RouteForAllExitFunc(whatDo ExitFunc) {
 			p.Release()
 			return
 
+		case Drop:
+			// Record it so the flow log shows the attempt, but never hand it to the receiver
+			r.unlockedInjectFlow(cm[x], receiver, p, false)
+
 		case KeepRouting:
 			fp := r.unlockedInjectFlow(cm[x], receiver, p, false)
 			receiver.InjectUDPPacket(p)

go.mod

@@ -7,7 +7,6 @@ require (
 	filippo.io/bigmod v0.1.0
 	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be
 	github.com/armon/go-radix v1.0.0
-	github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432
 	github.com/flynn/noise v1.1.0
 	github.com/gaissmai/bart v0.28.0
 	github.com/gogo/protobuf v1.3.2

go.sum

@@ -19,8 +19,6 @@ github.com/beorn7/perks v1.0.1/go.mod h1:G2ZrVWU2WbWT9wwq4/hrbKbnv/1ERSJQ0ibhJ6r
 github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UFvs=
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
-github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432 h1:M5QgkYacWj0Xs8MhpIK/5uwU02icXpEoSo9sM2aRCps=
-github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432/go.mod h1:xwIwAxMvYnVrGJPe2FKx5prTrnAjGOD8zvDOnxnrrkM=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=

graphite.go

@@ -0,0 +1,117 @@
+package nebula
+
+// This file is a trimmed, inlined copy of the graphite exporter from
+// github.com/cyberdelia/go-metrics-graphite, retaining only the Config type and
+// the Once entrypoint that Nebula uses. The upstream package has been
+// unmaintained for 10+ years, so it was vendored here to drop the dependency.
+// See https://github.com/slackhq/nebula/issues/1831.
+//
+// Copyright 2015 Timothée Peignier. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+//  1. Redistributions of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+import (
+	"bufio"
+	"fmt"
+	"net"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/rcrowley/go-metrics"
+)
+
+// graphiteConfigExport provides a container with configuration parameters for
+// the Graphite exporter.
+type graphiteConfigExport struct {
+	Addr          *net.TCPAddr     // Network address to connect to
+	Registry      metrics.Registry // Registry to be exported
+	FlushInterval time.Duration    // Flush interval
+	DurationUnit  time.Duration    // Time conversion unit for durations
+	Prefix        string           // Prefix to be prepended to metric names
+	Percentiles   []float64        // Percentiles to export from timers and histograms
+}
+
+// graphiteOnce performs a single submission to Graphite, returning a non-nil
+// error on failed connections.
+func graphiteOnce(c graphiteConfigExport) error {
+	now := time.Now().Unix()
+	du := float64(c.DurationUnit)
+	flushSeconds := float64(c.FlushInterval) / float64(time.Second)
+	conn, err := net.DialTCP("tcp", nil, c.Addr)
+	if err != nil {
+		return err
+	}
+	defer conn.Close()
+	w := bufio.NewWriter(conn)
+	c.Registry.Each(func(name string, i any) {
+		switch metric := i.(type) {
+		case metrics.Counter:
+			count := metric.Count()
+			fmt.Fprintf(w, "%s.%s.count %d %d\n", c.Prefix, name, count, now)
+			fmt.Fprintf(w, "%s.%s.count_ps %.2f %d\n", c.Prefix, name, float64(count)/flushSeconds, now)
+		case metrics.Gauge:
+			fmt.Fprintf(w, "%s.%s.value %d %d\n", c.Prefix, name, metric.Value(), now)
+		case metrics.GaugeFloat64:
+			fmt.Fprintf(w, "%s.%s.value %f %d\n", c.Prefix, name, metric.Value(), now)
+		case metrics.Histogram:
+			h := metric.Snapshot()
+			ps := h.Percentiles(c.Percentiles)
+			fmt.Fprintf(w, "%s.%s.count %d %d\n", c.Prefix, name, h.Count(), now)
+			fmt.Fprintf(w, "%s.%s.min %d %d\n", c.Prefix, name, h.Min(), now)
+			fmt.Fprintf(w, "%s.%s.max %d %d\n", c.Prefix, name, h.Max(), now)
+			fmt.Fprintf(w, "%s.%s.mean %.2f %d\n", c.Prefix, name, h.Mean(), now)
+			fmt.Fprintf(w, "%s.%s.std-dev %.2f %d\n", c.Prefix, name, h.StdDev(), now)
+			for psIdx, psKey := range c.Percentiles {
+				key := strings.Replace(strconv.FormatFloat(psKey*100.0, 'f', -1, 64), ".", "", 1)
+				fmt.Fprintf(w, "%s.%s.%s-percentile %.2f %d\n", c.Prefix, name, key, ps[psIdx], now)
+			}
+		case metrics.Meter:
+			m := metric.Snapshot()
+			fmt.Fprintf(w, "%s.%s.count %d %d\n", c.Prefix, name, m.Count(), now)
+			fmt.Fprintf(w, "%s.%s.one-minute %.2f %d\n", c.Prefix, name, m.Rate1(), now)
+			fmt.Fprintf(w, "%s.%s.five-minute %.2f %d\n", c.Prefix, name, m.Rate5(), now)
+			fmt.Fprintf(w, "%s.%s.fifteen-minute %.2f %d\n", c.Prefix, name, m.Rate15(), now)
+			fmt.Fprintf(w, "%s.%s.mean %.2f %d\n", c.Prefix, name, m.RateMean(), now)
+		case metrics.Timer:
+			t := metric.Snapshot()
+			ps := t.Percentiles(c.Percentiles)
+			count := t.Count()
+			fmt.Fprintf(w, "%s.%s.count %d %d\n", c.Prefix, name, count, now)
+			fmt.Fprintf(w, "%s.%s.count_ps %.2f %d\n", c.Prefix, name, float64(count)/flushSeconds, now)
+			fmt.Fprintf(w, "%s.%s.min %d %d\n", c.Prefix, name, t.Min()/int64(du), now)
+			fmt.Fprintf(w, "%s.%s.max %d %d\n", c.Prefix, name, t.Max()/int64(du), now)
+			fmt.Fprintf(w, "%s.%s.mean %.2f %d\n", c.Prefix, name, t.Mean()/du, now)
+			fmt.Fprintf(w, "%s.%s.std-dev %.2f %d\n", c.Prefix, name, t.StdDev()/du, now)
+			for psIdx, psKey := range c.Percentiles {
+				key := strings.Replace(strconv.FormatFloat(psKey*100.0, 'f', -1, 64), ".", "", 1)
+				fmt.Fprintf(w, "%s.%s.%s-percentile %.2f %d\n", c.Prefix, name, key, ps[psIdx]/du, now)
+			}
+			fmt.Fprintf(w, "%s.%s.one-minute %.2f %d\n", c.Prefix, name, t.Rate1(), now)
+			fmt.Fprintf(w, "%s.%s.five-minute %.2f %d\n", c.Prefix, name, t.Rate5(), now)
+			fmt.Fprintf(w, "%s.%s.fifteen-minute %.2f %d\n", c.Prefix, name, t.Rate15(), now)
+			fmt.Fprintf(w, "%s.%s.mean-rate %.2f %d\n", c.Prefix, name, t.RateMean(), now)
+		}
+		w.Flush()
+	})
+	return nil
+}

stats.go

@@ -13,7 +13,6 @@ import (
 	"sync/atomic"
 	"time"
 
-	graphite "github.com/cyberdelia/go-metrics-graphite"
 	mp "github.com/nbrownus/go-metrics-prometheus"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -253,7 +252,7 @@ func (s *statsServer) buildRuntime(cfg statsConfig) ([]func(), *http.Server) {
 		// loadStatsConfig already resolved and validated the address; re-parse
 		// the resolved form (no DNS lookup) to get a *net.TCPAddr.
 		addr, _ := net.ResolveTCPAddr(cfg.graphite.protocol, cfg.graphite.resolvedAddr)
-		gcfg := graphite.Config{
+		gcfg := graphiteConfigExport{
 			Addr:          addr,
 			Registry:      metrics.DefaultRegistry,
 			FlushInterval: cfg.interval,
@@ -262,7 +261,7 @@ func (s *statsServer) buildRuntime(cfg statsConfig) ([]func(), *http.Server) {
 			Percentiles:   []float64{0.5, 0.75, 0.95, 0.99, 0.999},
 		}
 		captureFns = append(captureFns, func() {
-			if err := graphite.Once(gcfg); err != nil {
+			if err := graphiteOnce(gcfg); err != nil {
 				s.l.Error("Graphite export failed", "error", err)
 			}
 		})

stats_test.go

@@ -371,7 +371,7 @@ func waitForListening(t *testing.T, addr string) {
 	})
 }
 
-// graphiteSink is a minimal TCP accept-and-discard server so graphite.Once
+// graphiteSink is a minimal TCP accept-and-discard server so graphiteOnce
 // calls in tests don't spam error logs or wedge on connection refused.
 type graphiteSink struct {
 	ln net.Listener