Use a random port for otel collector monitoring endpoint

go.mod

@@ -124,7 +124,6 @@ require (
 	go.opentelemetry.io/collector/processor/memorylimiterprocessor v0.135.0
 	go.opentelemetry.io/collector/receiver/nopreceiver v0.135.0
 	go.opentelemetry.io/collector/receiver/otlpreceiver v0.135.0
-	go.opentelemetry.io/collector/service v0.135.0
 	go.opentelemetry.io/ebpf-profiler v0.0.202536
 	go.uber.org/zap v1.27.0
 	go.yaml.in/yaml/v3 v3.0.4
@@ -722,6 +721,7 @@ require (
 	go.opentelemetry.io/collector/scraper v0.135.0 // indirect
 	go.opentelemetry.io/collector/scraper/scraperhelper v0.135.0 // indirect
 	go.opentelemetry.io/collector/semconv v0.128.1-0.20250610090210-188191247685 // indirect
+	go.opentelemetry.io/collector/service v0.135.0 // indirect
 	go.opentelemetry.io/collector/service/hostcapabilities v0.135.0 // indirect
 	go.opentelemetry.io/contrib/bridges/otelzap v0.12.0 // indirect
 	go.opentelemetry.io/contrib/detectors/gcp v1.36.0 // indirect

internal/pkg/agent/application/application.go

@@ -133,7 +133,13 @@ func New(
 	if err != nil {
 		return nil, nil, nil, fmt.Errorf("failed to create upgrader: %w", err)
 	}
-	monitor := componentmonitoring.New(isMonitoringSupported, cfg.Settings.DownloadConfig.OS(), cfg.Settings.MonitoringConfig, rawConfig.OTel, agentInfo, isOtelExecModeSubprocess)
+	monitor := componentmonitoring.New(
+		isMonitoringSupported,
+		cfg.Settings.DownloadConfig.OS(),
+		cfg.Settings.MonitoringConfig,
+		agentInfo,
+		isOtelExecModeSubprocess,
+	)
 
 	runtime, err := runtime.NewManager(
 		log,
@@ -245,7 +251,16 @@ func New(
 		return nil, nil, nil, errors.New(err, "failed to initialize composable controller")
 	}
 
-	otelManager, err := otelmanager.NewOTelManager(log.Named("otel_manager"), logLevel, baseLogger, otelExecMode, agentInfo, monitor.ComponentMonitoringConfig, cfg.Settings.ProcessConfig.StopTimeout)
+	otelManager, err := otelmanager.NewOTelManager(
+		log.Named("otel_manager"),
+		logLevel, baseLogger,
+		otelExecMode,
+		agentInfo,
+		0, // TODO: make this configurable in a follow-up
+		0, // TODO: make this configurable in a follow-up
+		monitor.ComponentMonitoringConfig,
+		cfg.Settings.ProcessConfig.StopTimeout,
+	)
 	if err != nil {
 		return nil, nil, nil, fmt.Errorf("failed to create otel manager: %w", err)
 	}

internal/pkg/agent/application/monitoring/component/v1_monitor.go

@@ -21,14 +21,11 @@ import (
 	"time"
 	"unicode"
 
-	"go.opentelemetry.io/collector/confmap"
-	"go.opentelemetry.io/collector/service"
+	koanfmaps "github.com/knadh/koanf/maps"
 
 	"github.com/elastic/elastic-agent/pkg/component"
 	"github.com/elastic/elastic-agent/pkg/utils"
 
-	koanfmaps "github.com/knadh/koanf/maps"
-
 	"github.com/elastic/elastic-agent/internal/pkg/agent/application/info"
 	"github.com/elastic/elastic-agent/internal/pkg/agent/application/paths"
 	"github.com/elastic/elastic-agent/internal/pkg/agent/errors"
@@ -38,6 +35,12 @@ import (
 )
 
 const (
+	// OtelCollectorMetricsPortEnvVarName is the name of the environment variable used to pass the collector metrics
+	// port to the managed EDOT collector. It exists because by default we use a random port for this, and we want to
+	// determine it as late as possible. However, the monitoring manager is instantiated early in the application
+	// startup process, so instead we rely on this variable. The OTel manager is required to set it whenever it starts
+	// a collector.
+	OtelCollectorMetricsPortEnvVarName = "EDOT_COLLECTOR_METRICS_PORT"
 	// args: data path, pipeline name, application name
 	logFileFormat = "%s/logs/%s"
 	// args: data path, install path, pipeline name, application name
@@ -93,7 +96,6 @@ var (
 type BeatsMonitor struct {
 	enabled                 bool // feature flag disabling whole v1 monitoring story
 	config                  *monitoringConfig
-	otelConfig              *confmap.Conf
 	operatingSystem         string
 	agentInfo               info.Agent
 	isOtelRuntimeSubprocess bool
@@ -115,13 +117,12 @@ type monitoringConfig struct {
 }
 
 // New creates a new BeatsMonitor instance.
-func New(enabled bool, operatingSystem string, cfg *monitoringCfg.MonitoringConfig, otelCfg *confmap.Conf, agentInfo info.Agent, isOtelRuntimeSubprocess bool) *BeatsMonitor {
+func New(enabled bool, operatingSystem string, cfg *monitoringCfg.MonitoringConfig, agentInfo info.Agent, isOtelRuntimeSubprocess bool) *BeatsMonitor {
 	return &BeatsMonitor{
 		enabled: enabled,
 		config: &monitoringConfig{
 			C: cfg,
 		},
-		otelConfig:              otelCfg,
 		operatingSystem:         operatingSystem,
 		agentInfo:               agentInfo,
 		isOtelRuntimeSubprocess: isOtelRuntimeSubprocess,
@@ -149,7 +150,6 @@ func (b *BeatsMonitor) Reload(rawConfig *config.Config) error {
 	}
 
 	b.config = &newConfig
-	b.otelConfig = rawConfig.OTel
 	return nil
 }
 
@@ -519,34 +519,9 @@ func (b *BeatsMonitor) monitoringNamespace() string {
 }
 
 func (b *BeatsMonitor) getCollectorTelemetryEndpoint() string {
-	if b.otelConfig != nil {
-		if serviceConfig, err := b.otelConfig.Sub("service"); err == nil {
-			var service service.Config
-			if serviceConfig.Unmarshal(&service, confmap.WithIgnoreUnused()) == nil {
-				for _, reader := range service.Telemetry.Metrics.Readers {
-					if reader.Pull == nil || reader.Pull.Exporter.Prometheus == nil {
-						continue
-					}
-					prometheus := *reader.Pull.Exporter.Prometheus
-					host := "localhost"
-					port := 8888
-
-					if prometheus.Host != nil {
-						host = *prometheus.Host
-					}
-					if prometheus.Port != nil {
-						port = *prometheus.Port
-					}
-					if prometheus.Host != nil || prometheus.Port != nil {
-						return host + ":" + strconv.Itoa(port)
-					}
-				}
-			}
-		}
-	}
-
-	// If there is no explicit configuration, the collector publishes its telemetry on port 8888.
-	return "localhost:8888"
+	// The OTel manager is required to set the environment variable. See comment at the constant definition for more
+	// information.
+	return fmt.Sprintf("localhost:${env:%s}", OtelCollectorMetricsPortEnvVarName)
 }
 
 // injectMetricsInput injects monitoring config for agent monitoring to the `cfg` object.

internal/pkg/agent/application/monitoring/component/v1_monitor_test.go

@@ -1102,7 +1102,7 @@ func TestMonitorReload(t *testing.T) {
 	monitorcfg.MonitorLogs = false
 	monitorcfg.MonitorMetrics = false
 
-	beatsMonitor := New(true, "", monitorcfg, nil, nil, false)
+	beatsMonitor := New(true, "", monitorcfg, nil, false)
 	assert.Equal(t, beatsMonitor.config.C.MonitorLogs, false)
 	assert.Equal(t, beatsMonitor.config.C.MonitorLogs, false)
 

internal/pkg/agent/cmd/inspect.go

@@ -414,7 +414,7 @@ func getMonitoringFn(ctx context.Context, logger *logger.Logger, cfg map[string]
 	}
 	otelExecMode := otelconfig.GetExecutionModeFromConfig(logger, config)
 	isOtelExecModeSubprocess := otelExecMode == manager.SubprocessExecutionMode
-	monitor := componentmonitoring.New(agentCfg.Settings.V1MonitoringEnabled, agentCfg.Settings.DownloadConfig.OS(), agentCfg.Settings.MonitoringConfig, otelCfg, agentInfo, isOtelExecModeSubprocess)
+	monitor := componentmonitoring.New(agentCfg.Settings.V1MonitoringEnabled, agentCfg.Settings.DownloadConfig.OS(), agentCfg.Settings.MonitoringConfig, agentInfo, isOtelExecModeSubprocess)
 	return monitor.MonitoringConfig, nil
 }
 

internal/pkg/otel/manager/common.go

@@ -6,6 +6,7 @@ package manager
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net"
 
@@ -52,21 +53,30 @@ func reportCollectorStatus(ctx context.Context, statusCh chan *status.AggregateS
 	}
 }
 
-// findRandomTCPPort finds a random available TCP port on the localhost interface.
-func findRandomTCPPort() (int, error) {
-	l, err := netListen("tcp", "localhost:0")
-	if err != nil {
-		return 0, err
-	}
+// findRandomTCPPorts finds count random available TCP ports on the localhost interface.
+func findRandomTCPPorts(count int) (ports []int, err error) {
+	ports = make([]int, 0, count)
+	listeners := make([]net.Listener, 0, count)
+	defer func() {
+		for _, listener := range listeners {
+			if closeErr := listener.Close(); closeErr != nil {
+				err = errors.Join(err, fmt.Errorf("error closing listener: %w", closeErr))
+			}
+		}
+	}()
+	for range count {
+		l, err := netListen("tcp", "localhost:0")
+		if err != nil {
+			return nil, err
+		}
+		listeners = append(listeners, l)
 
-	port := l.Addr().(*net.TCPAddr).Port
-	err = l.Close()
-	if err != nil {
-		return 0, err
-	}
-	if port == 0 {
-		return 0, fmt.Errorf("failed to find random port")
+		port := l.Addr().(*net.TCPAddr).Port
+		if port == 0 {
+			return nil, fmt.Errorf("failed to find random port")
+		}
+		ports = append(ports, port)
 	}
 
-	return port, nil
+	return ports, err
 }

internal/pkg/otel/manager/common_test.go

@@ -8,8 +8,11 @@ import (
 	"errors"
 	"net"
 	"path/filepath"
+	"slices"
 	"testing"
 
+	"github.com/stretchr/testify/assert"
+
 	"github.com/elastic/elastic-agent-client/v7/pkg/client"
 	"github.com/elastic/elastic-agent/internal/pkg/agent/application/paths"
 	"github.com/elastic/elastic-agent/pkg/component"
@@ -18,9 +21,15 @@ import (
 )
 
 func TestFindRandomPort(t *testing.T) {
-	port, err := findRandomTCPPort()
+	portCount := 2
+	ports, err := findRandomTCPPorts(portCount)
 	require.NoError(t, err)
-	require.NotEqual(t, 0, port)
+	require.Len(t, ports, portCount)
+	for _, port := range ports {
+		assert.NotEqual(t, 0, port)
+	}
+	slices.Sort(ports)
+	require.Len(t, slices.Compact(ports), portCount, "returned ports should be unique")
 
 	defer func() {
 		netListen = net.Listen
@@ -29,8 +38,8 @@ func TestFindRandomPort(t *testing.T) {
 	netListen = func(string, string) (net.Listener, error) {
 		return nil, errors.New("some error")
 	}
-	_, err = findRandomTCPPort()
-	require.Error(t, err, "failed to find random port")
+	_, err = findRandomTCPPorts(portCount)
+	assert.Error(t, err, "failed to find random port")
 }
 
 func testComponent(componentId string) component.Component {

internal/pkg/otel/manager/execution_embedded.go

@@ -6,6 +6,8 @@ package manager
 
 import (
 	"context"
+	"os"
+	"strconv"
 	"time"
 
 	"github.com/open-telemetry/opentelemetry-collector-contrib/pkg/status"
@@ -14,6 +16,8 @@ import (
 	"go.uber.org/zap"
 	"go.uber.org/zap/zapcore"
 
+	componentmonitoring "github.com/elastic/elastic-agent/internal/pkg/agent/application/monitoring/component"
+
 	"github.com/elastic/elastic-agent/internal/pkg/agent/application/paths"
 	"github.com/elastic/elastic-agent/internal/pkg/otel"
 	"github.com/elastic/elastic-agent/internal/pkg/otel/agentprovider"
@@ -22,11 +26,14 @@ import (
 	"github.com/elastic/elastic-agent/pkg/core/logger"
 )
 
-func newExecutionEmbedded() *embeddedExecution {
-	return &embeddedExecution{}
+// newExecutionEmbedded creates a new execution which runs the otel collector in a goroutine. A metricsPort of 0 will
+// result in a random port being used.
+func newExecutionEmbedded(metricsPort int) *embeddedExecution {
+	return &embeddedExecution{collectorMetricsPort: metricsPort}
 }
 
 type embeddedExecution struct {
+	collectorMetricsPort int
 }
 
 // startCollector starts the collector in a new goroutine.
@@ -41,6 +48,10 @@ func (r *embeddedExecution) startCollector(ctx context.Context, logger *logger.L
 	extConf := map[string]any{
 		"endpoint": paths.DiagnosticsExtensionSocket(),
 	}
+	collectorMetricsPort, err := r.getCollectorMetricsPort()
+	if err != nil {
+		return nil, err
+	}
 	// NewForceExtensionConverterFactory is used to ensure that the agent_status extension is always enabled.
 	// It is required for the Elastic Agent to extract the status out of the OTel collector.
 	settings := otel.NewSettings(
@@ -59,13 +70,41 @@ func (r *embeddedExecution) startCollector(ctx context.Context, logger *logger.L
 		return nil, err
 	}
 	go func() {
+		// Set the environment variable for the collector metrics port. See comment at the constant definition for more information.
+		setErr := os.Setenv(componentmonitoring.OtelCollectorMetricsPortEnvVarName, strconv.Itoa(collectorMetricsPort))
+		defer func() {
+			unsetErr := os.Unsetenv(componentmonitoring.OtelCollectorMetricsPortEnvVarName)
+			if unsetErr != nil {
+				logger.Errorf("couldn't unset environment variable %s: %v", componentmonitoring.OtelCollectorMetricsPortEnvVarName, unsetErr)
+			}
+		}()
+		if setErr != nil {
+			reportErr(ctx, errCh, setErr)
+			return
+		}
 		runErr := svc.Run(collectorCtx)
 		close(ctl.collectorDoneCh)
 		reportErr(ctx, errCh, runErr)
 	}()
 	return ctl, nil
 }
 
+// getCollectorPorts returns the metrics port used by the OTel collector. If the port set in the execution struct is 0,
+// a random port is returned instead.
+func (r *embeddedExecution) getCollectorMetricsPort() (metricsPort int, err error) {
+	// if the port is defined (non-zero), use it
+	if r.collectorMetricsPort > 0 {
+		return r.collectorMetricsPort, nil
+	}
+
+	// get a random port
+	ports, err := findRandomTCPPorts(1)
+	if err != nil {
+		return 0, err
+	}
+	return ports[0], nil
+}
+
 type ctxHandle struct {
 	collectorDoneCh chan struct{}
 	cancel          context.CancelFunc