refactor: Simplify telemetry metrics by removing site_id and enhancing tunnel_id usage

internal/telemetry/metrics.go

@@ -13,8 +13,8 @@ import (
 // low-cardinality label guidance from the issue description.
 //
 // Counters end with _total, durations are in seconds, sizes in bytes.
-// Only low-cardinality stable labels are supported: site_id, tunnel_id,
-// transport, direction, result, reason, error_type, region.
+// Only low-cardinality stable labels are supported: tunnel_id,
+// transport, direction, result, reason, error_type.
 var (
 	initOnce sync.Once
 
@@ -147,9 +147,9 @@ var (
 // Example inside your code (where you have access to current state):
 //
 //	telemetry.SetObservableCallback(func(ctx context.Context, o metric.Observer) error {
-//	    o.ObserveInt64(mSiteOnline, 1, attribute.String("site_id", siteID))
-//	    o.ObserveFloat64(mSiteLastHeartbeat, time.Since(lastHB).Seconds(), attribute.String("site_id", siteID))
-//	    o.ObserveInt64(mTunnelSessions, int64(len(activeSessions)), attribute.String("site_id", siteID))
+//	    o.ObserveInt64(mSiteOnline, 1)
+//	    o.ObserveFloat64(mSiteLastHeartbeat, time.Since(lastHB).Seconds())
+//	    o.ObserveInt64(mTunnelSessions, int64(len(activeSessions)))
 //	    return nil
 //	})
 func SetObservableCallback(cb func(context.Context, metric.Observer) error) {
@@ -174,20 +174,15 @@ func IncConfigReload(ctx context.Context, result string) {
 
 // Helpers for counters/histograms
 
-func IncSiteRegistration(ctx context.Context, siteID, region, result string) {
+func IncSiteRegistration(ctx context.Context, result string) {
 	attrs := []attribute.KeyValue{
-		attribute.String("site_id", siteID),
 		attribute.String("result", result),
 	}
-	if region != "" {
-		attrs = append(attrs, attribute.String("region", region))
-	}
 	mSiteRegistrations.Add(ctx, 1, metric.WithAttributes(attrs...))
 }
 
-func AddTunnelBytes(ctx context.Context, siteID, tunnelID, direction string, n int64) {
+func AddTunnelBytes(ctx context.Context, tunnelID, direction string, n int64) {
 	mTunnelBytes.Add(ctx, n, metric.WithAttributes(
-		attribute.String("site_id", siteID),
 		attribute.String("tunnel_id", tunnelID),
 		attribute.String("direction", direction),
 	))
@@ -198,33 +193,29 @@ func AddTunnelBytesSet(ctx context.Context, n int64, attrs attribute.Set) {
 	mTunnelBytes.Add(ctx, n, metric.WithAttributeSet(attrs))
 }
 
-func ObserveTunnelLatency(ctx context.Context, siteID, tunnelID, transport string, seconds float64) {
+func ObserveTunnelLatency(ctx context.Context, tunnelID, transport string, seconds float64) {
 	mTunnelLatency.Record(ctx, seconds, metric.WithAttributes(
-		attribute.String("site_id", siteID),
 		attribute.String("tunnel_id", tunnelID),
 		attribute.String("transport", transport),
 	))
 }
 
-func IncReconnect(ctx context.Context, siteID, tunnelID, reason string) {
+func IncReconnect(ctx context.Context, tunnelID, reason string) {
 	mReconnects.Add(ctx, 1, metric.WithAttributes(
-		attribute.String("site_id", siteID),
 		attribute.String("tunnel_id", tunnelID),
 		attribute.String("reason", reason),
 	))
 }
 
-func IncConnAttempt(ctx context.Context, siteID, transport, result string) {
+func IncConnAttempt(ctx context.Context, transport, result string) {
 	mConnAttempts.Add(ctx, 1, metric.WithAttributes(
-		attribute.String("site_id", siteID),
 		attribute.String("transport", transport),
 		attribute.String("result", result),
 	))
 }
 
-func IncConnError(ctx context.Context, siteID, transport, typ string) {
+func IncConnError(ctx context.Context, transport, typ string) {
 	mConnErrors.Add(ctx, 1, metric.WithAttributes(
-		attribute.String("site_id", siteID),
 		attribute.String("transport", transport),
 		attribute.String("error_type", typ),
 	))

internal/telemetry/state_view.go

@@ -42,16 +42,19 @@ func RegisterStateView(v StateView) {
 							if online {
 								val = 1
 							}
-							o.ObserveInt64(mSiteOnline, val, metric.WithAttributes(attribute.String("site_id", siteID)))
+							o.ObserveInt64(mSiteOnline, val)
 						}
 						if t, ok := sv.LastHeartbeat(siteID); ok {
 							secs := time.Since(t).Seconds()
-							o.ObserveFloat64(mSiteLastHeartbeat, secs, metric.WithAttributes(attribute.String("site_id", siteID)))
+							o.ObserveFloat64(mSiteLastHeartbeat, secs)
 						}
 						// If the view supports per-tunnel sessions, report them labeled by tunnel_id.
 						if tm, ok := any.(interface{ SessionsByTunnel() map[string]int64 }); ok {
 							for tid, n := range tm.SessionsByTunnel() {
-								o.ObserveInt64(mTunnelSessions, n, metric.WithAttributes(attribute.String("tunnel_id", tid)))
+								o.ObserveInt64(mTunnelSessions, n, metric.WithAttributes(
+									attribute.String("tunnel_id", tid),
+									attribute.String("transport", "tcp"),
+								))
 							}
 						}
 					}

internal/telemetry/telemetry.go

@@ -93,16 +93,21 @@ type Setup struct {
 // installs recommended histogram views for *_latency_seconds, and returns a Setup with
 // a Shutdown method to flush exporters.
 func Init(ctx context.Context, cfg Config) (*Setup, error) {
+	// Build resource with required attributes and only include optional ones when non-empty
+	attrs := []attribute.KeyValue{
+		semconv.ServiceName(cfg.ServiceName),
+		semconv.ServiceVersion(cfg.ServiceVersion),
+	}
+	if cfg.SiteID != "" {
+		attrs = append(attrs, attribute.String("site_id", cfg.SiteID))
+	}
+	if cfg.Region != "" {
+		attrs = append(attrs, attribute.String("region", cfg.Region))
+	}
 	res, _ := resource.New(ctx,
 		resource.WithFromEnv(),
 		resource.WithHost(),
-		resource.WithAttributes(
-			semconv.ServiceName(cfg.ServiceName),
-			semconv.ServiceVersion(cfg.ServiceVersion),
-			// Optional resource attributes
-			attribute.String("site_id", cfg.SiteID),
-			attribute.String("region", cfg.Region),
-		),
+		resource.WithAttributes(attrs...),
 	)
 
 	s := &Setup{}
@@ -168,7 +173,7 @@ func Init(ctx context.Context, cfg Config) (*Setup, error) {
 			AttributeFilter: func(kv attribute.KeyValue) bool {
 				k := string(kv.Key)
 				switch k {
-				case "tunnel_id", "transport", "direction", "protocol", "result", "reason", "error_type":
+				case "tunnel_id", "transport", "direction", "protocol", "result", "reason", "error_type", "version", "commit":
 					return true
 				default:
 					return false

internal/telemetry/telemetry_golden_test.go

@@ -25,7 +25,7 @@ cfg := Config{ServiceName: "newt", PromEnabled: true, AdminAddr: "127.0.0.1:0",
 	defer ts.Close()
 
 	// Trigger a counter
-	IncConnAttempt(ctx, "ignored", "websocket", "success")
+	IncConnAttempt(ctx, "websocket", "success")
 	time.Sleep(100 * time.Millisecond)
 
 	resp, err := http.Get(ts.URL)

internal/telemetry/telemetry_smoke_test.go

@@ -36,7 +36,7 @@ func TestMetricsSmoke(t *testing.T) {
 	defer ts.Close()
 
 	// Record a simple metric and then fetch /metrics
-	IncConnAttempt(ctx, "site-1", "websocket", "success")
+	IncConnAttempt(ctx, "websocket", "success")
 	// Give the exporter a tick to collect
 	time.Sleep(100 * time.Millisecond)