[supervisor] add ssh tunnel metrics

gitpod-io · Sep 25, 2023 · f18f7ca · f18f7ca
1 parent f690f87
commit f18f7ca
Show file tree

Hide file tree

Showing 4 changed files with 40 additions and 2 deletions.
diff --git a/components/supervisor/pkg/metrics/metrics.go b/components/supervisor/pkg/metrics/metrics.go
@@ -13,6 +13,8 @@ import (
 type SupervisorMetrics struct {
 	IDEReadyDurationTotal *prometheus.HistogramVec
 	InitializerHistogram  *prometheus.HistogramVec
+	SSHTunnelOpenedTotal  *prometheus.CounterVec
+	SSHTunnelClosedTotal  *prometheus.CounterVec
 }
 
 func NewMetrics() *SupervisorMetrics {
@@ -27,6 +29,14 @@ func NewMetrics() *SupervisorMetrics {
 			Help:    "initializer speed in bytes per second",
 			Buckets: prometheus.ExponentialBuckets(1024*1024, 2, 12),
 		}, []string{"kind"}),
+		SSHTunnelOpenedTotal: prometheus.NewCounterVec(prometheus.CounterOpts{
+			Name: "supervisor_ssh_tunnel_opened_total",
+			Help: "Total number of SSH tunnels opened by the supervisor",
+		}, []string{}),
+		SSHTunnelClosedTotal: prometheus.NewCounterVec(prometheus.CounterOpts{
+			Name: "supervisor_ssh_tunnel_closed_total",
+			Help: "Total number of SSH tunnels closed by the supervisor",
+		}, []string{"code"}),
 	}
 }
 

diff --git a/components/supervisor/pkg/metrics/reporter.go b/components/supervisor/pkg/metrics/reporter.go
@@ -43,6 +43,8 @@ func NewGrpcMetricsReporter(gitpodHost string) *GrpcMetricsReporter {
 			"supervisor_initializer_bytes_second": true,
 			"supervisor_client_handled_total":     true,
 			"supervisor_client_handling_seconds":  true,
+			"supervisor_ssh_tunnel_opened_total":  true,
+			"supervisor_ssh_tunnel_closed_total":  true,
 		},
 		values: make(map[string]float64),
 		addCounter: func(name string, labels map[string]string, value uint64) {

diff --git a/components/supervisor/pkg/supervisor/supervisor.go b/components/supervisor/pkg/supervisor/supervisor.go
@@ -418,7 +418,7 @@ func Run(options ...RunOption) {
 	}
 
 	wg.Add(1)
-	go startAPIEndpoint(ctx, cfg, &wg, apiServices, tunneledPortsService, metricsReporter, apiEndpointOpts...)
+	go startAPIEndpoint(ctx, cfg, &wg, apiServices, tunneledPortsService, metricsReporter, supervisorMetrics, apiEndpointOpts...)
 
 	wg.Add(1)
 	go startSSHServer(ctx, cfg, &wg)
@@ -1187,7 +1187,7 @@ func isBlacklistedEnvvar(name string) bool {
 	return false
 }
 
-func startAPIEndpoint(ctx context.Context, cfg *Config, wg *sync.WaitGroup, services []RegisterableService, tunneled *ports.TunneledPortsService, metricsReporter *metrics.GrpcMetricsReporter, opts ...grpc.ServerOption) {
+func startAPIEndpoint(ctx context.Context, cfg *Config, wg *sync.WaitGroup, services []RegisterableService, tunneled *ports.TunneledPortsService, metricsReporter *metrics.GrpcMetricsReporter, supervisorMetrics *metrics.SupervisorMetrics, opts ...grpc.ServerOption) {
 	defer wg.Done()
 	defer log.Debug("startAPIEndpoint shutdown")
 
@@ -1308,6 +1308,16 @@ func startAPIEndpoint(ctx context.Context, cfg *Config, wg *sync.WaitGroup, serv
 		tunnelOverWebSocket(tunneled, conn)
 	}))
 	routes.Handle("/_supervisor/tunnel/ssh", http.HandlerFunc(func(rw http.ResponseWriter, r *http.Request) {
+		supervisorMetrics.SSHTunnelOpenedTotal.WithLabelValues().Inc()
+		defer func() {
+			code := "unknown"
+			if err != nil {
+				if e, ok := err.(*websocket.CloseError); ok {
+					code = strconv.Itoa(e.Code)
+				}
+			}
+			supervisorMetrics.SSHTunnelClosedTotal.WithLabelValues(code).Inc()
+		}()
 		wsConn, err := upgrader.Upgrade(rw, r, nil)
 		if err != nil {
 			log.WithError(err).Error("tunnel ssh: upgrade to the WebSocket protocol failed")

diff --git a/install/installer/pkg/components/ide-metrics/configmap.go b/install/installer/pkg/components/ide-metrics/configmap.go
@@ -336,6 +336,22 @@ func configmap(ctx *common.RenderContext) ([]runtime.Object, error) {
 				},
 			},
 		},
+		{
+			Name:   "supervisor_ssh_tunnel_opened_total",
+			Help:   "Total number of SSH tunnels opened by the supervisor",
+			Labels: []config.LabelAllowList{},
+		},
+		{
+			Name: "supervisor_ssh_tunnel_closed_total",
+			Help: "Total number of SSH tunnels closed by the supervisor",
+			Labels: []config.LabelAllowList{
+				{
+					Name:         "code",
+					AllowValues:  []string{"*"},
+					DefaultValue: "unknown",
+				},
+			},
+		},
 	}
 
 	histogramMetrics := []config.HistogramMetricsConfiguration{