router: add metric to track service instance count (#3918)

Add two metrics to track the number of service instances known to the
data plane.

    # HELP router_service_instance_changes_total Number of total service instance changes. Both addition and removal of a service instance is accumulated
    # TYPE router_service_instance_changes_total counter
    router_service_instance_changes_total{isd_as="1-ff00:0:110",service="CS"} 11
    router_service_instance_changes_total{isd_as="1-ff00:0:110",service="DS"} 11
    # HELP router_service_instance_count Number of service instances known by the data plane.
    # TYPE router_service_instance_count gauge
    router_service_instance_count{isd_as="1-ff00:0:110",service="CS"} 3
    router_service_instance_count{isd_as="1-ff00:0:110",service="DS"} 3

go/pkg/router/dataplane.go

@@ -329,6 +329,11 @@ func (d *DataPlane) AddSvc(svc addr.HostSVC, a *net.UDPAddr) error {
 		d.svc = newServices()
 	}
 	d.svc.AddSvc(svc, a)
+	if d.Metrics != nil {
+		labels := serviceMetricLabels(d.localIA, svc)
+		d.Metrics.ServiceInstanceChanges.With(labels).Add(1)
+		d.Metrics.ServiceInstanceCount.With(labels).Add(1)
+	}
 	return nil
 }
 
@@ -343,6 +348,11 @@ func (d *DataPlane) DelSvc(svc addr.HostSVC, a *net.UDPAddr) error {
 		return nil
 	}
 	d.svc.DelSvc(svc, a)
+	if d.Metrics != nil {
+		labels := serviceMetricLabels(d.localIA, svc)
+		d.Metrics.ServiceInstanceChanges.With(labels).Add(1)
+		d.Metrics.ServiceInstanceCount.With(labels).Add(-1)
+	}
 	return nil
 }
 
@@ -1436,3 +1446,10 @@ func interfaceToMetricLabels(id uint16, localIA addr.IA,
 		"neighbor_isd_as": neighbors[id].String(),
 	}
 }
+
+func serviceMetricLabels(localIA addr.IA, svc addr.HostSVC) prometheus.Labels {
+	return prometheus.Labels{
+		"isd_as":  localIA.String(),
+		"service": svc.BaseString(),
+	}
+}

go/pkg/router/metrics.go

@@ -30,6 +30,8 @@ type Metrics struct {
 	BFDInterfaceStateChanges  *prometheus.CounterVec
 	BFDPacketsSent            *prometheus.CounterVec
 	BFDPacketsReceived        *prometheus.CounterVec
+	ServiceInstanceCount      *prometheus.GaugeVec
+	ServiceInstanceChanges    *prometheus.CounterVec
 	SiblingReachable          *prometheus.GaugeVec
 	SiblingBFDPacketsSent     *prometheus.CounterVec
 	SiblingBFDPacketsReceived *prometheus.CounterVec
@@ -104,6 +106,21 @@ func NewMetrics() *Metrics {
 			},
 			[]string{"interface", "isd_as", "neighbor_isd_as"},
 		),
+		ServiceInstanceCount: promauto.NewGaugeVec(
+			prometheus.GaugeOpts{
+				Name: "router_service_instance_count",
+				Help: "Number of service instances known by the data plane.",
+			},
+			[]string{"service", "isd_as"},
+		),
+		ServiceInstanceChanges: promauto.NewCounterVec(
+			prometheus.CounterOpts{
+				Name: "router_service_instance_changes_total",
+				Help: "Number of total service instance changes. Both addition and removal of a " +
+					"service instance is accumulated.",
+			},
+			[]string{"service", "isd_as"},
+		),
 		SiblingReachable: promauto.NewGaugeVec(
 			prometheus.GaugeOpts{
 				Name: "router_sibling_reachable",

go/posix-router/main.go

@@ -30,7 +30,7 @@ import (
 	"github.com/scionproto/scion/go/lib/serrors"
 	"github.com/scionproto/scion/go/pkg/command"
 	"github.com/scionproto/scion/go/pkg/router"
-	brconf "github.com/scionproto/scion/go/pkg/router/config"
+	"github.com/scionproto/scion/go/pkg/router/config"
 	"github.com/scionproto/scion/go/pkg/router/control"
 	"github.com/scionproto/scion/go/pkg/service"
 )
@@ -52,7 +52,7 @@ func main() {
 	}
 	cmd.AddCommand(
 		command.NewCompletion(cmd),
-		command.NewSample(cmd, command.NewSampleConfig(&brconf.Config{})),
+		command.NewSample(cmd, command.NewSampleConfig(&config.Config{})),
 		command.NewVersion(cmd),
 	)
 	cmd.Flags().StringVar(&flags.config, "config", "", "Configuration file (required)")
@@ -72,7 +72,7 @@ func run(file string, metrics *router.Metrics) error {
 	defer log.Flush()
 	defer env.LogAppStopped("BR", fileConfig.General.ID)
 	defer log.HandlePanic()
-	if err := validateCfg(fileConfig); err != nil {
+	if err := validateConfig(fileConfig); err != nil {
 		return err
 	}
 	controlConfig, err := loadControlConfig(fileConfig)
@@ -115,38 +115,38 @@ func run(file string, metrics *router.Metrics) error {
 	}
 }
 
-func setupBasic(file string) (brconf.Config, error) {
-	var cfg brconf.Config
+func setupBasic(file string) (config.Config, error) {
+	var cfg config.Config
 	if err := libconfig.LoadFile(file, &cfg); err != nil {
-		return brconf.Config{}, serrors.WrapStr("loading config from file", err, "file", file)
+		return config.Config{}, serrors.WrapStr("loading config from file", err, "file", file)
 	}
 	cfg.InitDefaults()
 	if err := log.Setup(cfg.Logging); err != nil {
-		return brconf.Config{}, serrors.WrapStr("initialize logging", err)
+		return config.Config{}, serrors.WrapStr("initialize logging", err)
 	}
 	prom.ExportElementID(cfg.General.ID)
 	if err := env.LogAppStarted("BR", cfg.General.ID); err != nil {
-		return brconf.Config{}, err
+		return config.Config{}, err
 	}
 	return cfg, nil
 }
 
-func validateCfg(cfg brconf.Config) error {
+func validateConfig(cfg config.Config) error {
 	if err := cfg.Validate(); err != nil {
 		return serrors.WrapStr("validating config", err)
 	}
 	return nil
 }
 
-func loadControlConfig(cfg brconf.Config) (*control.Config, error) {
+func loadControlConfig(cfg config.Config) (*control.Config, error) {
 	newConf, err := control.LoadConfig(cfg.General.ID, cfg.General.ConfigDir)
 	if err != nil {
 		return nil, serrors.WrapStr("loading topology", err)
 	}
 	return newConf, nil
 }
 
-func setupHTTPHandlers(cfg brconf.Config) error {
+func setupHTTPHandlers(cfg config.Config) error {
 	statusPages := service.StatusPages{
 		"info":      service.NewInfoHandler(),
 		"config":    service.NewConfigHandler(cfg),