Auto activate checks required by some sysprobe features

cmd/agent/dist/conf.d/oom_kill.d/conf.yaml.example → cmd/agent/dist/conf.d/oom_kill.d/conf.yaml.default

@@ -1,3 +1,6 @@
+ad_identifiers:
+  - _oom_kill
+
 init_config:
 
 instances:

cmd/agent/dist/conf.d/tcp_queue_length.d/conf.yaml.example → cmd/agent/dist/conf.d/tcp_queue_length.d/conf.yaml.default

@@ -1,3 +1,6 @@
+ad_identifiers:
+  - _tcp_queue_length
+
 init_config:
 
 instances:

comp/core/autodiscovery/impl/autoconfig.go

@@ -41,6 +41,7 @@ import (
 	logComp "github.com/DataDog/datadog-agent/comp/core/log/def"
 	secrets "github.com/DataDog/datadog-agent/comp/core/secrets/def"
 	"github.com/DataDog/datadog-agent/comp/core/status"
+	sysprobeconfig "github.com/DataDog/datadog-agent/comp/core/sysprobeconfig/def"
 	tagger "github.com/DataDog/datadog-agent/comp/core/tagger/def"
 	telemetry "github.com/DataDog/datadog-agent/comp/core/telemetry/def"
 	workloadfilter "github.com/DataDog/datadog-agent/comp/core/workloadfilter/def"
@@ -69,6 +70,7 @@ type Requires struct {
 	TaggerComp     tagger.Component
 	Secrets        secrets.Component
 	WMeta          option.Option[workloadmeta.Component]
+	SysProbeConfig option.Option[sysprobeconfig.Component] `optional:"true"`
 	FilterStore    workloadfilter.Component
 	Telemetry      telemetry.Component
 	HealthPlatform option.Option[healthplatformdef.Component]
@@ -103,6 +105,7 @@ type AutoConfig struct {
 	healthPlatform           option.Option[healthplatformdef.Component]
 	staticConfigIndex        *listeners.StaticConfigIndex
 	serviceTracker           adtypes.ServiceTracker
+	sysProbeConfig           option.Option[sysprobeconfig.Component]
 
 	// m covers the `configPollers`, `listenerCandidates`, `listeners`, and `listenerRetryStop`, but
 	// not the values they point to.
@@ -181,6 +184,7 @@ func newAutoConfig(deps Requires) autodiscoverydef.Component {
 	}()
 
 	ac := createNewAutoConfig(schController, deps.Secrets, deps.WMeta, deps.TaggerComp, deps.Log, deps.Telemetry, deps.FilterStore, deps.HealthPlatform, deps.ServiceTracker)
+	ac.sysProbeConfig = deps.SysProbeConfig
 	deps.Lc.Append(compdef.Hook{
 		OnStart: func(_ context.Context) error {
 			ac.start()
@@ -600,6 +604,7 @@ func (ac *AutoConfig) addListenerCandidates(listenerConfigs []pkgconfigsetup.Lis
 			Filter:            ac.filterStore,
 			Tagger:            ac.taggerComp,
 			Wmeta:             ac.wmeta,
+			SysProbeConfig:    ac.sysProbeConfig,
 			StaticConfigIndex: ac.staticConfigIndex,
 			ServiceTracker:    ac.serviceTracker,
 		}

comp/core/autodiscovery/listeners/environment.go

@@ -7,15 +7,27 @@ package listeners
 
 import (
 	"github.com/DataDog/datadog-agent/comp/core/autodiscovery/integration"
+	sysprobeconfig "github.com/DataDog/datadog-agent/comp/core/sysprobeconfig/def"
 	workloadfilter "github.com/DataDog/datadog-agent/comp/core/workloadfilter/def"
 	workloadmeta "github.com/DataDog/datadog-agent/comp/core/workloadmeta/def"
 	"github.com/DataDog/datadog-agent/pkg/config/env"
 	"github.com/DataDog/datadog-agent/pkg/util/log"
+	"github.com/DataDog/datadog-agent/pkg/util/option"
 )
 
+// Checks activated from configuration state (avoid double work to activate it for users)
+var sysProbeConfigChecks = []struct {
+	adIdentifier string
+	configKey    string
+}{
+	{adIdentifier: "_oom_kill", configKey: "system_probe_config.enable_oom_kill"},
+	{adIdentifier: "_tcp_queue_length", configKey: "system_probe_config.enable_tcp_queue_length"},
+}
+
 // EnvironmentListener implements a ServiceListener based on current environment
 type EnvironmentListener struct {
-	newService chan<- Service
+	newService     chan<- Service
+	sysProbeConfig option.Option[sysprobeconfig.Component]
 }
 
 // EnvironmentService represents services generated from EnvironmentListener
@@ -27,8 +39,8 @@ type EnvironmentService struct {
 var _ Service = &EnvironmentService{}
 
 // NewEnvironmentListener creates an EnvironmentListener
-func NewEnvironmentListener(ServiceListernerDeps) (ServiceListener, error) {
-	return &EnvironmentListener{}, nil
+func NewEnvironmentListener(deps ServiceListernerDeps) (ServiceListener, error) {
+	return &EnvironmentListener{sysProbeConfig: deps.SysProbeConfig}, nil
 }
 
 // Listen starts the goroutine to detect checks based on environment
@@ -69,6 +81,16 @@ func (l *EnvironmentListener) createServices() {
 		log.Infof("Listener created container service from environment")
 		l.newService <- &EnvironmentService{adIdentifier: "_container"}
 	}
+
+	// Handle checks auto-activated from system-probe configuration state.
+	if sysProbeConfig, ok := l.sysProbeConfig.Get(); ok {
+		for _, check := range sysProbeConfigChecks {
+			if sysProbeConfig.GetBool(check.configKey) {
+				log.Infof("Listener created %s service from system-probe configuration", check.adIdentifier)
+				l.newService <- &EnvironmentService{adIdentifier: check.adIdentifier}
+			}
+		}
+	}
 }
 
 // Equal returns whether the two EnvironmentService are equal

comp/core/autodiscovery/listeners/environment_test.go

@@ -0,0 +1,93 @@
+// Unless explicitly stated otherwise all files in this repository are licensed
+// under the Apache License Version 2.0.
+// This product includes software developed at Datadog (https://www.datadoghq.com/).
+// Copyright 2017-present Datadog, Inc.
+
+package listeners
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+
+	sysprobeconfig "github.com/DataDog/datadog-agent/comp/core/sysprobeconfig/def"
+	sysprobeconfigmock "github.com/DataDog/datadog-agent/comp/core/sysprobeconfig/mock"
+	"github.com/DataDog/datadog-agent/pkg/util/option"
+)
+
+// collectEnvironmentServices runs createServices synchronously and returns the AD
+// identifiers of the emitted services.
+func collectEnvironmentServices(l *EnvironmentListener) []string {
+	ch := make(chan Service, 16)
+	l.newService = ch
+	l.createServices()
+	close(ch)
+
+	var ids []string
+	for svc := range ch {
+		ids = append(ids, svc.GetServiceID())
+	}
+	return ids
+}
+
+func TestEnvironmentListenerSysProbeChecks(t *testing.T) {
+	tests := []struct {
+		name      string
+		overrides map[string]any
+		expected  []string
+		absent    []string
+	}{
+		{
+			name:      "nothing enabled",
+			overrides: map[string]any{},
+			absent:    []string{"_oom_kill", "_tcp_queue_length"},
+		},
+		{
+			name:      "oom_kill enabled",
+			overrides: map[string]any{"system_probe_config.enable_oom_kill": true},
+			expected:  []string{"_oom_kill"},
+			absent:    []string{"_tcp_queue_length"},
+		},
+		{
+			name:      "tcp_queue_length enabled",
+			overrides: map[string]any{"system_probe_config.enable_tcp_queue_length": true},
+			expected:  []string{"_tcp_queue_length"},
+			absent:    []string{"_oom_kill"},
+		},
+		{
+			name: "both enabled",
+			overrides: map[string]any{
+				"system_probe_config.enable_oom_kill":         true,
+				"system_probe_config.enable_tcp_queue_length": true,
+			},
+			expected: []string{"_oom_kill", "_tcp_queue_length"},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			sysProbeCfg := sysprobeconfigmock.NewMockWithOverrides(t, tc.overrides)
+			l := &EnvironmentListener{sysProbeConfig: option.New(sysProbeCfg)}
+
+			ids := collectEnvironmentServices(l)
+
+			for _, id := range tc.expected {
+				assert.Contains(t, ids, id)
+			}
+			for _, id := range tc.absent {
+				assert.NotContains(t, ids, id)
+			}
+		})
+	}
+}
+
+// TestEnvironmentListenerNoSysProbeConfig ensures the listener does not panic and
+// emits no system-probe based service when the sysprobeconfig component is absent.
+func TestEnvironmentListenerNoSysProbeConfig(t *testing.T) {
+	l := &EnvironmentListener{sysProbeConfig: option.None[sysprobeconfig.Component]()}
+
+	ids := collectEnvironmentServices(l)
+
+	assert.NotContains(t, ids, "_oom_kill")
+	assert.NotContains(t, ids, "_tcp_queue_length")
+}

comp/core/autodiscovery/listeners/types.go

@@ -11,6 +11,7 @@ import (
 	"github.com/DataDog/datadog-agent/comp/core/autodiscovery/common/types"
 	"github.com/DataDog/datadog-agent/comp/core/autodiscovery/integration"
 	"github.com/DataDog/datadog-agent/comp/core/autodiscovery/telemetry"
+	sysprobeconfig "github.com/DataDog/datadog-agent/comp/core/sysprobeconfig/def"
 	tagger "github.com/DataDog/datadog-agent/comp/core/tagger/def"
 	workloadfilter "github.com/DataDog/datadog-agent/comp/core/workloadfilter/def"
 	workloadmeta "github.com/DataDog/datadog-agent/comp/core/workloadmeta/def"
@@ -76,6 +77,7 @@ type ServiceListernerDeps struct {
 	Tagger            tagger.Component
 	Filter            workloadfilter.Component
 	Wmeta             option.Option[workloadmeta.Component]
+	SysProbeConfig    option.Option[sysprobeconfig.Component]
 	StaticConfigIndex *StaticConfigIndex
 	ServiceTracker    types.ServiceTracker
 }