Instrument auth package

Add a histogram duration transport wrapper to the auth http client connection. * Refactor metrics namespace to a separate package to avoid recursive imports. Signed-off-by: SuperQ <[email protected]>
voc · Dec 25, 2024 · 954c410 · 954c410
1 parent cece1cf
commit 954c410
Show file tree

Hide file tree

Showing 4 changed files with 52 additions and 28 deletions.
diff --git a/api/metrics.go b/api/metrics.go
@@ -1,7 +1,7 @@
 package api
 
 import (
-	"github.com/voc/srtrelay/config"
+	"github.com/voc/srtrelay/internal/metrics"
 	"github.com/voc/srtrelay/srt"
 
 	"github.com/prometheus/client_golang/prometheus"
@@ -11,128 +11,128 @@ const srtSubsystem = "srt"
 
 var (
 	activeSocketsDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "active_sockets"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "active_sockets"),
 		"The number of active SRT sockets",
 		nil, nil,
 	)
 
 	// Metrics from: https://pkg.go.dev/github.com/haivision/srtgo#SrtStats
 	pktSentTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_packets_total"),
 		"total number of sent data packets, including retransmissions",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRecvTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_packets_total"),
 		"total number of received packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktSndLossTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_lost_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_lost_packets_total"),
 		"total number of lost packets (sender side)",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRcvLossTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_lost_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_lost_packets_total"),
 		"total number of lost packets (receive_side)",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRetransTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "retransmitted_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "retransmitted_packets_total"),
 		"total number of retransmitted packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktSentACKTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_ack_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_ack_packets_total"),
 		"total number of sent ACK packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRecvACKTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_ack_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_ack_packets_total"),
 		"total number of received ACK packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktSentNAKTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_nak_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_nak_packets_total"),
 		"total number of received NAK packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRecvNAKTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_nak_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_nak_packets_total"),
 		"total number of received NAK packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	sndDurationTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "udt_sent_duration_seconds_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "udt_sent_duration_seconds_total"),
 		"total time duration when UDT is sending data (idle time exclusive)",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktSndDropTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_dropped_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_dropped_packets_total"),
 		"number of too-late-to-send dropped packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRcvDropTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_dropped_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_dropped_packets_total"),
 		"number of too-late-to play missing packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	pktRcvUndecryptTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_undecrypted_packets_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_undecrypted_packets_total"),
 		"number of undecrypted packets",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteSentTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_bytes_total"),
 		"total number of sent data bytes, including retransmissions",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteRecvTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_bytes_total"),
 		"total number of received bytes",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteRcvLossTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_lost_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_lost_bytes_total"),
 		"total number of lost bytes",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteRetransTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "retransmitted_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "retransmitted_bytes_total"),
 		"total number of retransmitted bytes",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteSndDropTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "sent_dropped_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "sent_dropped_bytes_total"),
 		"number of too-late-to-send dropped bytes",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteRcvDropTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_dropped_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_dropped_bytes_total"),
 		"number of too-late-to play missing bytes (estimate based on average packet size)",
 		[]string{"address", "stream_id"}, nil,
 	)
 
 	byteRcvUndecryptTotalDesc = prometheus.NewDesc(
-		prometheus.BuildFQName(config.MetricsNamespace, srtSubsystem, "receive_undecrypted_bytes_total"),
+		prometheus.BuildFQName(metrics.Namespace, srtSubsystem, "receive_undecrypted_bytes_total"),
 		"number of undecrypted bytes",
 		[]string{"address", "stream_id"}, nil,
 	)

diff --git a/auth/http.go b/auth/http.go
@@ -6,7 +6,26 @@ import (
 	"net/url"
 	"time"
 
+	"github.com/voc/srtrelay/internal/metrics"
 	"github.com/voc/srtrelay/stream"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
+)
+
+var (
+	requestDurations = promauto.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace:                   metrics.Namespace,
+			Subsystem:                   "auth",
+			Name:                        "request_duration_seconds",
+			Help:                        "A histogram of auth http request latencies.",
+			Buckets:                     prometheus.DefBuckets,
+			NativeHistogramBucketFactor: 1.1,
+		},
+		[]string{"url", "application"},
+	)
 )
 
 type httpAuth struct {
@@ -33,11 +52,13 @@ type HTTPAuthConfig struct {
 }
 
 // NewHttpAuth creates an Authenticator with a HTTP backend
-func NewHTTPAuth(config HTTPAuthConfig) *httpAuth {
+func NewHTTPAuth(authConfig HTTPAuthConfig) *httpAuth {
+	m := requestDurations.MustCurryWith(prometheus.Labels{"url": authConfig.URL, "application": authConfig.Application})
 	return &httpAuth{
-		config: config,
+		config: authConfig,
 		client: &http.Client{
-			Timeout: time.Duration(config.Timeout),
+			Timeout:   time.Duration(authConfig.Timeout),
+			Transport: promhttp.InstrumentRoundTripperDuration(m, http.DefaultTransport),
 		},
 	}
 }

diff --git a/internal/metrics/metrics.go b/internal/metrics/metrics.go
@@ -0,0 +1,3 @@
+package metrics
+
+const Namespace = "srtrelay"
diff --git a/relay/channel.go b/relay/channel.go
@@ -6,7 +6,7 @@ import (
 	"sync/atomic"
 	"time"
 
-	"github.com/voc/srtrelay/config"
+	"github.com/voc/srtrelay/internal/metrics"
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
@@ -17,14 +17,14 @@ const relaySubsystem = "relay"
 var (
 	activeClients = promauto.NewGaugeVec(
 		prometheus.GaugeOpts{
-			Name: prometheus.BuildFQName(config.MetricsNamespace, relaySubsystem, "active_clients"),
+			Name: prometheus.BuildFQName(metrics.Namespace, relaySubsystem, "active_clients"),
 			Help: "The number of active clients per channel",
 		},
 		[]string{"channel_name"},
 	)
 	channelCreatedTimestamp = promauto.NewGaugeVec(
 		prometheus.GaugeOpts{
-			Name: prometheus.BuildFQName(config.MetricsNamespace, relaySubsystem, "created_timestamp_seconds"),
+			Name: prometheus.BuildFQName(metrics.Namespace, relaySubsystem, "created_timestamp_seconds"),
 			Help: "The UNIX timestamp when the channel was created",
 		},
 		[]string{"channel_name"},