feat: expose rcmgr metrics when enabled (#8785)

* add metrics for the resource manager
* export protocol and service name in Prometheus metrics
* fix: expose rcmgr metrics only when enabled

Co-authored-by: Marcin Rataj <lidel@lidel.org>

core/node/groups.go

@@ -146,7 +146,6 @@ func LibP2P(bcfg *BuildCfg, cfg *config.Config) fx.Option {
 
 		// Services (resource management)
 		fx.Provide(libp2p.ResourceManager(cfg.Swarm.ResourceMgr)),
-
 		fx.Provide(libp2p.AddrFilters(cfg.Swarm.AddrFilters)),
 		fx.Provide(libp2p.AddrsFactory(cfg.Addresses.Announce, cfg.Addresses.AppendAnnounce, cfg.Addresses.NoAnnounce)),
 		fx.Provide(libp2p.SmuxTransport(cfg.Swarm.Transports)),

core/node/libp2p/rcmgr.go

@@ -6,15 +6,19 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"strconv"
 	"strings"
 
 	config "github.com/ipfs/go-ipfs/config"
 	"github.com/ipfs/go-ipfs/repo"
+
 	"github.com/libp2p/go-libp2p"
 	"github.com/libp2p/go-libp2p-core/network"
 	"github.com/libp2p/go-libp2p-core/peer"
 	"github.com/libp2p/go-libp2p-core/protocol"
 	rcmgr "github.com/libp2p/go-libp2p-resource-manager"
+
+	"github.com/prometheus/client_golang/prometheus"
 	"go.uber.org/fx"
 )
 
@@ -71,7 +75,8 @@ func ResourceManager(cfg config.ResourceMgr) func(fx.Lifecycle, repo.Repo) (netw
 
 			libp2p.SetDefaultServiceLimits(limiter)
 
-			var ropts []rcmgr.Option
+			ropts := []rcmgr.Option{rcmgr.WithMetrics(createRcmgrMetrics())}
+
 			if os.Getenv("LIBP2P_DEBUG_RCMGR") != "" {
 				traceFilePath := filepath.Join(repoPath, NetLimitTraceFilename)
 				ropts = append(ropts, rcmgr.WithTrace(traceFilePath))
@@ -373,3 +378,230 @@ func NetSetLimit(mgr network.ResourceManager, scope string, limit config.Resourc
 		return fmt.Errorf("invalid scope %q", scope)
 	}
 }
+
+func createRcmgrMetrics() rcmgr.MetricsReporter {
+	const (
+		direction = "direction"
+		usesFD    = "usesFD"
+		protocol  = "protocol"
+		service   = "service"
+	)
+
+	connAllowed := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_conns_allowed_total",
+			Help: "allowed connections",
+		},
+		[]string{direction, usesFD},
+	)
+	prometheus.MustRegister(connAllowed)
+
+	connBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_conns_blocked_total",
+			Help: "blocked connections",
+		},
+		[]string{direction, usesFD},
+	)
+	prometheus.MustRegister(connBlocked)
+
+	streamAllowed := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_streams_allowed_total",
+			Help: "allowed streams",
+		},
+		[]string{direction},
+	)
+	prometheus.MustRegister(streamAllowed)
+
+	streamBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_streams_blocked_total",
+			Help: "blocked streams",
+		},
+		[]string{direction},
+	)
+	prometheus.MustRegister(streamBlocked)
+
+	peerAllowed := prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "libp2p_rcmgr_peers_allowed_total",
+		Help: "allowed peers",
+	})
+	prometheus.MustRegister(peerAllowed)
+
+	peerBlocked := prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "libp2p_rcmgr_peer_blocked_total",
+		Help: "blocked peers",
+	})
+	prometheus.MustRegister(peerBlocked)
+
+	protocolAllowed := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_protocols_allowed_total",
+			Help: "allowed streams attached to a protocol",
+		},
+		[]string{protocol},
+	)
+	prometheus.MustRegister(protocolAllowed)
+
+	protocolBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_protocols_blocked_total",
+			Help: "blocked streams attached to a protocol",
+		},
+		[]string{protocol},
+	)
+	prometheus.MustRegister(protocolBlocked)
+
+	protocolPeerBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_protocols_for_peer_blocked_total",
+			Help: "blocked streams attached to a protocol for a specific peer",
+		},
+		[]string{protocol},
+	)
+	prometheus.MustRegister(protocolPeerBlocked)
+
+	serviceAllowed := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_services_allowed_total",
+			Help: "allowed streams attached to a service",
+		},
+		[]string{service},
+	)
+	prometheus.MustRegister(serviceAllowed)
+
+	serviceBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_services_blocked_total",
+			Help: "blocked streams attached to a service",
+		},
+		[]string{service},
+	)
+	prometheus.MustRegister(serviceBlocked)
+
+	servicePeerBlocked := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "libp2p_rcmgr_service_for_peer_blocked_total",
+			Help: "blocked streams attached to a service for a specific peer",
+		},
+		[]string{service},
+	)
+	prometheus.MustRegister(servicePeerBlocked)
+
+	memoryAllowed := prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "libp2p_rcmgr_memory_allocations_allowed_total",
+		Help: "allowed memory allocations",
+	})
+	prometheus.MustRegister(memoryAllowed)
+
+	memoryBlocked := prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "libp2p_rcmgr_memory_allocations_blocked_total",
+		Help: "blocked memory allocations",
+	})
+	prometheus.MustRegister(memoryBlocked)
+
+	return rcmgrMetrics{
+		connAllowed,
+		connBlocked,
+		streamAllowed,
+		streamBlocked,
+		peerAllowed,
+		peerBlocked,
+		protocolAllowed,
+		protocolBlocked,
+		protocolPeerBlocked,
+		serviceAllowed,
+		serviceBlocked,
+		servicePeerBlocked,
+		memoryAllowed,
+		memoryBlocked,
+	}
+}
+
+// Failsafe to ensure interface from go-libp2p-resource-manager is implemented
+var _ rcmgr.MetricsReporter = rcmgrMetrics{}
+
+type rcmgrMetrics struct {
+	connAllowed         *prometheus.CounterVec
+	connBlocked         *prometheus.CounterVec
+	streamAllowed       *prometheus.CounterVec
+	streamBlocked       *prometheus.CounterVec
+	peerAllowed         prometheus.Counter
+	peerBlocked         prometheus.Counter
+	protocolAllowed     *prometheus.CounterVec
+	protocolBlocked     *prometheus.CounterVec
+	protocolPeerBlocked *prometheus.CounterVec
+	serviceAllowed      *prometheus.CounterVec
+	serviceBlocked      *prometheus.CounterVec
+	servicePeerBlocked  *prometheus.CounterVec
+	memoryAllowed       prometheus.Counter
+	memoryBlocked       prometheus.Counter
+}
+
+func getDirection(d network.Direction) string {
+	switch d {
+	default:
+		return ""
+	case network.DirInbound:
+		return "inbound"
+	case network.DirOutbound:
+		return "outbound"
+	}
+}
+
+func (r rcmgrMetrics) AllowConn(dir network.Direction, usefd bool) {
+	r.connAllowed.WithLabelValues(getDirection(dir), strconv.FormatBool(usefd)).Inc()
+}
+
+func (r rcmgrMetrics) BlockConn(dir network.Direction, usefd bool) {
+	r.connBlocked.WithLabelValues(getDirection(dir), strconv.FormatBool(usefd)).Inc()
+}
+
+func (r rcmgrMetrics) AllowStream(_ peer.ID, dir network.Direction) {
+	r.streamAllowed.WithLabelValues(getDirection(dir)).Inc()
+}
+
+func (r rcmgrMetrics) BlockStream(_ peer.ID, dir network.Direction) {
+	r.streamBlocked.WithLabelValues(getDirection(dir)).Inc()
+}
+
+func (r rcmgrMetrics) AllowPeer(_ peer.ID) {
+	r.peerAllowed.Inc()
+}
+
+func (r rcmgrMetrics) BlockPeer(_ peer.ID) {
+	r.peerBlocked.Inc()
+}
+
+func (r rcmgrMetrics) AllowProtocol(proto protocol.ID) {
+	r.protocolAllowed.WithLabelValues(string(proto)).Inc()
+}
+
+func (r rcmgrMetrics) BlockProtocol(proto protocol.ID) {
+	r.protocolBlocked.WithLabelValues(string(proto)).Inc()
+}
+
+func (r rcmgrMetrics) BlockProtocolPeer(proto protocol.ID, _ peer.ID) {
+	r.protocolPeerBlocked.WithLabelValues(string(proto)).Inc()
+}
+
+func (r rcmgrMetrics) AllowService(svc string) {
+	r.serviceAllowed.WithLabelValues(svc).Inc()
+}
+
+func (r rcmgrMetrics) BlockService(svc string) {
+	r.serviceBlocked.WithLabelValues(svc).Inc()
+}
+
+func (r rcmgrMetrics) BlockServicePeer(svc string, _ peer.ID) {
+	r.servicePeerBlocked.WithLabelValues(svc).Inc()
+}
+
+func (r rcmgrMetrics) AllowMemory(_ int) {
+	r.memoryAllowed.Inc()
+}
+
+func (r rcmgrMetrics) BlockMemory(_ int) {
+	r.memoryBlocked.Inc()
+}

go.mod

@@ -105,9 +105,6 @@ require (
 	github.com/whyrusleeping/go-sysinfo v0.0.0-20190219211824-4a357d4b90b1
 	github.com/whyrusleeping/multiaddr-filter v0.0.0-20160516205228-e903e4adabd7
 	go.opencensus.io v0.23.0
-	go.uber.org/dig v1.14.0
-	go.uber.org/fx v1.16.0
-	go.uber.org/zap v1.21.0
 	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.27.0
 	go.opentelemetry.io/otel v1.2.0
 	go.opentelemetry.io/otel/exporters/jaeger v1.2.0
@@ -116,6 +113,9 @@ require (
 	go.opentelemetry.io/otel/exporters/stdout/stdouttrace v1.2.0
 	go.opentelemetry.io/otel/sdk v1.2.0
 	go.opentelemetry.io/otel/trace v1.2.0
+	go.uber.org/dig v1.14.0
+	go.uber.org/fx v1.16.0
+	go.uber.org/zap v1.21.0
 	golang.org/x/crypto v0.0.0-20210921155107-089bfa567519
 	golang.org/x/sync v0.0.0-20210220032951-036812b2e83c
 	golang.org/x/sys v0.0.0-20211025112917-711f33c9992c