Adds connection state metrics

Signed-off-by: Elena Kolevska <elena@kolevska.com>

Author: elena-kolevska

extra/redisotel-native/metrics.go

@@ -21,8 +21,9 @@ const (
 // metricsRecorder implements the otel.Recorder interface
 type metricsRecorder struct {
 	operationDuration metric.Float64Histogram
+	connectionCount   metric.Int64UpDownCounter
 
-	// Client configuration for attributes
+	// Client configuration for attributes (used for operation metrics only)
 	serverAddr string
 	serverPort string
 	dbIndex    string
@@ -267,3 +268,59 @@ func formatDBIndex(db int) string {
 	}
 	return strconv.Itoa(db)
 }
+
+// RecordConnectionStateChange records a change in connection state
+// This is called from the pool when connections transition between states
+func (r *metricsRecorder) RecordConnectionStateChange(
+	ctx context.Context,
+	cn redis.ConnInfo,
+	fromState, toState string,
+) {
+	if r.connectionCount == nil {
+		return
+	}
+
+	// Extract server address from connection
+	serverAddr, serverPort := extractServerInfo(cn)
+
+	// Build base attributes
+	attrs := []attribute.KeyValue{
+		attribute.String("db.system", "redis"),
+		attribute.String("server.address", serverAddr),
+	}
+
+	// Add server.port if not default
+	if serverPort != "" && serverPort != "6379" {
+		attrs = append(attrs, attribute.String("server.port", serverPort))
+	}
+
+	// Decrement old state (if not empty)
+	if fromState != "" {
+		fromAttrs := append([]attribute.KeyValue{}, attrs...)
+		fromAttrs = append(fromAttrs, attribute.String("state", fromState))
+		r.connectionCount.Add(ctx, -1, metric.WithAttributes(fromAttrs...))
+	}
+
+	// Increment new state
+	if toState != "" {
+		toAttrs := append([]attribute.KeyValue{}, attrs...)
+		toAttrs = append(toAttrs, attribute.String("state", toState))
+		r.connectionCount.Add(ctx, 1, metric.WithAttributes(toAttrs...))
+	}
+}
+
+// extractServerInfo extracts server address and port from connection info
+func extractServerInfo(cn redis.ConnInfo) (addr, port string) {
+	if cn == nil {
+		return "", ""
+	}
+
+	remoteAddr := cn.RemoteAddr()
+	if remoteAddr == nil {
+		return "", ""
+	}
+
+	addrStr := remoteAddr.String()
+	host, portStr := parseAddr(addrStr)
+	return host, portStr
+}

extra/redisotel-native/redisotel.go

@@ -125,9 +125,20 @@ func initOnce(client redis.UniversalClient, opts ...Option) error {
 		return fmt.Errorf("failed to create operation duration histogram: %w", err)
 	}
 
+	// Create synchronous UpDownCounter for connection count
+	connectionCount, err := meter.Int64UpDownCounter(
+		"db.client.connection.count",
+		metric.WithDescription("The number of connections that are currently in state described by the state attribute"),
+		metric.WithUnit("{connection}"),
+	)
+	if err != nil {
+		return fmt.Errorf("failed to create connection count metric: %w", err)
+	}
+
 	// Create recorder
 	recorder := &metricsRecorder{
 		operationDuration: operationDuration,
+		connectionCount:   connectionCount,
 		serverAddr:        serverAddr,
 		serverPort:        serverPort,
 		dbIndex:           dbIndex,

internal/otel/metrics.go

@@ -21,6 +21,9 @@ type Cmder interface {
 type Recorder interface {
 	// RecordOperationDuration records the total operation duration (including all retries)
 	RecordOperationDuration(ctx context.Context, duration time.Duration, cmd Cmder, attempts int, cn *pool.Conn)
+
+	// RecordConnectionStateChange records when a connection changes state
+	RecordConnectionStateChange(ctx context.Context, cn *pool.Conn, fromState, toState string)
 }
 
 // Global recorder instance (initialized by extra/redisotel-native)
@@ -30,9 +33,16 @@ var globalRecorder Recorder = noopRecorder{}
 func SetGlobalRecorder(r Recorder) {
 	if r == nil {
 		globalRecorder = noopRecorder{}
+		// Unregister pool callback
+		pool.SetConnectionStateChangeCallback(nil)
 		return
 	}
 	globalRecorder = r
+
+	// Register pool callback to forward state changes to recorder
+	pool.SetConnectionStateChangeCallback(func(ctx context.Context, cn *pool.Conn, fromState, toState string) {
+		globalRecorder.RecordConnectionStateChange(ctx, cn, fromState, toState)
+	})
 }
 
 // RecordOperationDuration records the total operation duration.
@@ -41,7 +51,14 @@ func RecordOperationDuration(ctx context.Context, duration time.Duration, cmd Cm
 	globalRecorder.RecordOperationDuration(ctx, duration, cmd, attempts, cn)
 }
 
+// RecordConnectionStateChange records when a connection changes state.
+// This is called from pool.go when connections transition between states.
+func RecordConnectionStateChange(ctx context.Context, cn *pool.Conn, fromState, toState string) {
+	globalRecorder.RecordConnectionStateChange(ctx, cn, fromState, toState)
+}
+
 // noopRecorder is a no-op implementation (zero overhead when metrics disabled)
 type noopRecorder struct{}
 
 func (noopRecorder) RecordOperationDuration(context.Context, time.Duration, Cmder, int, *pool.Conn) {}
+func (noopRecorder) RecordConnectionStateChange(context.Context, *pool.Conn, string, string)       {}

internal/pool/pool.go

@@ -24,6 +24,9 @@ var (
 	// ErrPoolTimeout timed out waiting to get a connection from the connection pool.
 	ErrPoolTimeout = errors.New("redis: connection pool timeout")
 
+	// Global callback for connection state changes (set by otel package)
+	connectionStateChangeCallback func(ctx context.Context, cn *Conn, fromState, toState string)
+
 	// popAttempts is the maximum number of attempts to find a usable connection
 	// when popping from the idle connection pool. This handles cases where connections
 	// are temporarily marked as unusable (e.g., during maintenanceNotifications upgrades or network issues).
@@ -42,6 +45,12 @@ var (
 	noExpiration = maxTime
 )
 
+// SetConnectionStateChangeCallback sets the global callback for connection state changes.
+// This is called by the otel package to register metrics recording.
+func SetConnectionStateChangeCallback(fn func(ctx context.Context, cn *Conn, fromState, toState string)) {
+	connectionStateChangeCallback = fn
+}
+
 var timers = sync.Pool{
 	New: func() interface{} {
 		t := time.NewTimer(time.Hour)
@@ -468,6 +477,12 @@ func (p *ConnPool) getConn(ctx context.Context) (*Conn, error) {
 		}
 
 		atomic.AddUint32(&p.stats.Hits, 1)
+
+		// Notify metrics: connection moved from idle to used
+		if connectionStateChangeCallback != nil {
+			connectionStateChangeCallback(ctx, cn, "idle", "used")
+		}
+
 		return cn, nil
 	}
 
@@ -492,6 +507,12 @@ func (p *ConnPool) getConn(ctx context.Context) (*Conn, error) {
 			return nil, err
 		}
 	}
+
+	// Notify metrics: new connection is created and used
+	if connectionStateChangeCallback != nil {
+		connectionStateChangeCallback(ctx, newcn, "", "used")
+	}
+
 	return newcn, nil
 }
 
@@ -659,9 +680,19 @@ func (p *ConnPool) Put(ctx context.Context, cn *Conn) {
 			p.connsMu.Unlock()
 		}
 		p.idleConnsLen.Add(1)
+
+		// Notify metrics: connection moved from used to idle
+		if connectionStateChangeCallback != nil {
+			connectionStateChangeCallback(ctx, cn, "used", "idle")
+		}
 	} else {
 		p.removeConnWithLock(cn)
 		shouldCloseConn = true
+
+		// Notify metrics: connection removed (used -> nothing)
+		if connectionStateChangeCallback != nil {
+			connectionStateChangeCallback(ctx, cn, "used", "")
+		}
 	}
 
 	p.freeTurn()
@@ -671,11 +702,16 @@ func (p *ConnPool) Put(ctx context.Context, cn *Conn) {
 	}
 }
 
-func (p *ConnPool) Remove(_ context.Context, cn *Conn, reason error) {
+func (p *ConnPool) Remove(ctx context.Context, cn *Conn, reason error) {
 	p.removeConnWithLock(cn)
 
 	p.freeTurn()
 
+	// Notify metrics: connection removed (assume from used state)
+	if connectionStateChangeCallback != nil {
+		connectionStateChangeCallback(ctx, cn, "used", "")
+	}
+
 	_ = p.closeConn(cn)
 
 	// Check if we need to create new idle connections to maintain MinIdleConns

otel.go

@@ -24,6 +24,9 @@ type ConnInfo interface {
 type OTelRecorder interface {
 	// RecordOperationDuration records the total operation duration (including all retries)
 	RecordOperationDuration(ctx context.Context, duration time.Duration, cmd Cmder, attempts int, cn ConnInfo)
+
+	// RecordConnectionStateChange records when a connection changes state (e.g., idle -> used)
+	RecordConnectionStateChange(ctx context.Context, cn ConnInfo, fromState, toState string)
 }
 
 // SetOTelRecorder sets the global OpenTelemetry recorder.
@@ -55,3 +58,12 @@ func (a *otelRecorderAdapter) RecordOperationDuration(ctx context.Context, durat
 	}
 }
 
+func (a *otelRecorderAdapter) RecordConnectionStateChange(ctx context.Context, cn *pool.Conn, fromState, toState string) {
+	// Convert internal pool.Conn to public ConnInfo
+	var connInfo ConnInfo
+	if cn != nil {
+		connInfo = cn
+	}
+	a.recorder.RecordConnectionStateChange(ctx, connInfo, fromState, toState)
+}
+