Add metric labels (#612)

cmd/run.go

@@ -696,6 +696,8 @@ var runCmd = &cobra.Command{
 				// Add clients to the pool.
 				for range currentPoolSize {
 					clientConfig := clients[configGroupName][configBlockName]
+					clientConfig.GroupName = configGroupName
+					clientConfig.BlockName = configBlockName
 					client := network.NewClient(
 						runCtx, clientConfig, logger,
 						network.NewRetry(
@@ -716,6 +718,7 @@ var runCmd = &cobra.Command{
 					if client != nil {
 						eventOptions := trace.WithAttributes(
 							attribute.String("name", configBlockName),
+							attribute.String("group", configGroupName),
 							attribute.String("network", client.Network),
 							attribute.String("address", client.Address),
 							attribute.Int("receiveChunkSize", client.ReceiveChunkSize),
@@ -746,6 +749,8 @@ var runCmd = &cobra.Command{
 
 						clientCfg := map[string]interface{}{
 							"id":                 client.ID,
+							"name":               configBlockName,
+							"group":              configGroupName,
 							"network":            client.Network,
 							"address":            client.Address,
 							"receiveChunkSize":   client.ReceiveChunkSize,
@@ -851,7 +856,8 @@ var runCmd = &cobra.Command{
 				proxies[configGroupName][configBlockName] = network.NewProxy(
 					runCtx,
 					network.Proxy{
-						Name:                 configBlockName,
+						GroupName:            configGroupName,
+						BlockName:            configBlockName,
 						AvailableConnections: pools[configGroupName][configBlockName],
 						PluginRegistry:       pluginRegistry,
 						HealthCheckPeriod:    cfg.HealthCheckPeriod,
@@ -899,8 +905,9 @@ var runCmd = &cobra.Command{
 			servers[name] = network.NewServer(
 				runCtx,
 				network.Server{
-					Network: cfg.Network,
-					Address: cfg.Address,
+					GroupName: name,
+					Network:   cfg.Network,
+					Address:   cfg.Address,
 					TickInterval: config.If(
 						cfg.TickInterval > 0,
 						cfg.TickInterval,

config/types.go

@@ -44,6 +44,9 @@ type ActionRedisConfig struct {
 }
 
 type Client struct {
+	BlockName string `json:"-"`
+	GroupName string `json:"-"`
+
 	Network            string        `json:"network" jsonschema:"enum=tcp,enum=udp,enum=unix" yaml:"network"`
 	Address            string        `json:"address" yaml:"address"`
 	TCPKeepAlive       bool          `json:"tcpKeepAlive" yaml:"tcpKeepAlive"`

gatewayd.yaml

@@ -85,7 +85,7 @@ servers:
       # Load balancer strategies can be found in config/constants.go
       strategy: ROUND_ROBIN # ROUND_ROBIN, RANDOM, WEIGHTED_ROUND_ROBIN
       consistentHash:
-        useSourceIp: true
+        useSourceIp: true # Set to false for using the RANDOM strategy
       # Optional configuration for strategies that support rules (e.g., WEIGHTED_ROUND_ROBIN)
       # loadBalancingRules:
       #   - condition: "DEFAULT" # Currently, only the "DEFAULT" condition is supported

metrics/builtins.go

@@ -10,51 +10,51 @@ const (
 )
 
 var (
-	ClientConnections = promauto.NewGauge(prometheus.GaugeOpts{
+	ClientConnections = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: Namespace,
 		Name:      "client_connections",
 		Help:      "Number of client connections",
-	})
-	ServerConnections = promauto.NewGauge(prometheus.GaugeOpts{
+	}, []string{"group", "block"})
+	ServerConnections = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: Namespace,
 		Name:      "server_connections",
 		Help:      "Number of server connections",
-	})
-	TLSConnections = promauto.NewGauge(prometheus.GaugeOpts{
+	}, []string{"group", "block"})
+	TLSConnections = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: Namespace,
 		Name:      "tls_connections",
 		Help:      "Number of TLS connections",
-	})
+	}, []string{"group", "block"})
 	ServerTicksFired = promauto.NewCounter(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "server_ticks_fired_total",
 		Help:      "Total number of server ticks fired",
 	})
-	BytesReceivedFromClient = promauto.NewSummary(prometheus.SummaryOpts{
+	BytesReceivedFromClient = promauto.NewSummaryVec(prometheus.SummaryOpts{
 		Namespace: Namespace,
 		Name:      "bytes_received_from_client",
 		Help:      "Number of bytes received from client",
-	})
-	BytesSentToServer = promauto.NewSummary(prometheus.SummaryOpts{
+	}, []string{"group", "block"})
+	BytesSentToServer = promauto.NewSummaryVec(prometheus.SummaryOpts{
 		Namespace: Namespace,
 		Name:      "bytes_sent_to_server",
 		Help:      "Number of bytes sent to server",
-	})
-	BytesReceivedFromServer = promauto.NewSummary(prometheus.SummaryOpts{
+	}, []string{"group", "block"})
+	BytesReceivedFromServer = promauto.NewSummaryVec(prometheus.SummaryOpts{
 		Namespace: Namespace,
 		Name:      "bytes_received_from_server",
 		Help:      "Number of bytes received from server",
-	})
-	BytesSentToClient = promauto.NewSummary(prometheus.SummaryOpts{
+	}, []string{"group", "block"})
+	BytesSentToClient = promauto.NewSummaryVec(prometheus.SummaryOpts{
 		Namespace: Namespace,
 		Name:      "bytes_sent_to_client",
 		Help:      "Number of bytes sent to client",
-	})
-	TotalTrafficBytes = promauto.NewSummary(prometheus.SummaryOpts{
+	}, []string{"group", "block"})
+	TotalTrafficBytes = promauto.NewSummaryVec(prometheus.SummaryOpts{
 		Namespace: Namespace,
 		Name:      "traffic_bytes",
 		Help:      "Number of total bytes passed through GatewayD via client or server",
-	})
+	}, []string{"group", "block"})
 	PluginsLoaded = promauto.NewCounter(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "plugins_loaded_total",
@@ -70,31 +70,31 @@ var (
 		Name:      "plugin_hooks_executed_total",
 		Help:      "Number of plugin hooks executed",
 	})
-	ProxyHealthChecks = promauto.NewCounter(prometheus.CounterOpts{
+	ProxyHealthChecks = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "proxy_health_checks_total",
 		Help:      "Number of proxy health checks",
-	})
-	ProxiedConnections = promauto.NewGauge(prometheus.GaugeOpts{
+	}, []string{"group", "block"})
+	ProxiedConnections = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: Namespace,
 		Name:      "proxied_connections",
 		Help:      "Number of proxy connects",
-	})
-	ProxyPassThroughsToClient = promauto.NewCounter(prometheus.CounterOpts{
+	}, []string{"group", "block"})
+	ProxyPassThroughsToClient = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "proxy_passthroughs_to_client_total",
 		Help:      "Number of successful proxy passthroughs from server to client",
-	})
-	ProxyPassThroughsToServer = promauto.NewCounter(prometheus.CounterOpts{
+	}, []string{"group", "block"})
+	ProxyPassThroughsToServer = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "proxy_passthroughs_to_server_total",
 		Help:      "Number of successful proxy passthroughs from client to server",
-	})
-	ProxyPassThroughTerminations = promauto.NewCounter(prometheus.CounterOpts{
+	}, []string{"group", "block"})
+	ProxyPassThroughTerminations = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "proxy_passthrough_terminations_total",
 		Help:      "Number of proxy passthrough terminations by plugins",
-	})
+	}, []string{"group", "block"})
 	APIRequests = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: Namespace,
 		Name:      "api_requests_total",

network/client.go

@@ -35,6 +35,9 @@ type Client struct {
 	mu        sync.Mutex
 	retry     IRetry
 
+	GroupName string
+	BlockName string
+
 	TCPKeepAlive       bool
 	TCPKeepAlivePeriod time.Duration
 	ReceiveChunkSize   int
@@ -168,7 +171,7 @@ func NewClient(
 		logger,
 	)
 
-	metrics.ServerConnections.Inc()
+	metrics.ServerConnections.WithLabelValues(clientConfig.GroupName, clientConfig.BlockName).Inc()
 
 	return &client
 }
@@ -267,7 +270,7 @@ func (c *Client) Reconnect() error {
 	if c.conn != nil {
 		c.Close()
 	} else {
-		metrics.ServerConnections.Dec()
+		metrics.ServerConnections.WithLabelValues(c.GroupName, c.BlockName).Dec()
 	}
 	c.connected.Store(false)
 
@@ -306,7 +309,7 @@ func (c *Client) Reconnect() error {
 	)
 	c.connected.Store(true)
 	c.logger.Debug().Str("address", c.Address).Msg("Reconnected to server")
-	metrics.ServerConnections.Inc()
+	metrics.ServerConnections.WithLabelValues(c.GroupName, c.BlockName).Inc()
 	span.AddEvent("Reconnected to server")
 
 	return nil
@@ -343,7 +346,7 @@ func (c *Client) Close() {
 	c.Address = ""
 	c.Network = ""
 
-	metrics.ServerConnections.Dec()
+	metrics.ServerConnections.WithLabelValues(c.GroupName, c.BlockName).Dec()
 
 	span.AddEvent("Closed connection to server")
 }

network/network_helpers_test.go

@@ -290,11 +290,15 @@ func (m MockProxy) BusyConnectionsString() []string {
 	return nil
 }
 
-// GetName returns the name of the MockProxy.
-func (m MockProxy) GetName() string {
+// GetBlockName returns the name of the MockProxy.
+func (m MockProxy) GetBlockName() string {
 	return m.name
 }
 
+func (m MockProxy) GetGroupName() string {
+	return "default"
+}
+
 // Mock implementation of IConnWrapper.
 type MockConnWrapper struct {
 	mock.Mock