Using new otlpfile pipeline for self metrics

cmd/google_cloud_ops_agent_diagnostics/main_linux.go

@@ -24,20 +24,19 @@ import (
 	"syscall"
 
 	"github.com/GoogleCloudPlatform/ops-agent/cmd/google_cloud_ops_agent_diagnostics/utils"
-	"github.com/GoogleCloudPlatform/ops-agent/internal/self_metrics"
 )
 
 var (
 	config = flag.String("config", "/etc/google-cloud-ops-agent/config.yaml", "path to the user specified agent config")
 )
 
 func run(ctx context.Context) error {
-	userUc, mergedUc, err := utils.GetUserAndMergedConfigs(ctx, *config)
+	_, _, err := utils.GetUserAndMergedConfigs(ctx, *config)
 	if err != nil {
 		return err
 	}
 
-	ctx, cancel := context.WithCancel(ctx)
+	_, cancel := context.WithCancel(ctx)
 	defer cancel()
 
 	go func() {
@@ -54,10 +53,5 @@ func run(ctx context.Context) error {
 		}
 	}()
 
-	err = self_metrics.CollectOpsAgentSelfMetrics(ctx, userUc, mergedUc)
-	if err != nil {
-		return err
-	}
-
 	return nil
 }

cmd/google_cloud_ops_agent_diagnostics/main_windows.go

@@ -23,7 +23,6 @@ import (
 	"os"
 
 	"github.com/GoogleCloudPlatform/ops-agent/cmd/google_cloud_ops_agent_diagnostics/utils"
-	"github.com/GoogleCloudPlatform/ops-agent/internal/self_metrics"
 	"go.opentelemetry.io/otel"
 	"golang.org/x/sys/windows/svc"
 	"golang.org/x/sys/windows/svc/debug"
@@ -81,7 +80,7 @@ func (s *service) Execute(args []string, r <-chan svc.ChangeRequest, changes cha
 		return false, ERROR_INVALID_PARAMETER
 	}
 
-	userUc, mergedUc, err := utils.GetUserAndMergedConfigs(ctx, s.userConf)
+	_, _, err := utils.GetUserAndMergedConfigs(ctx, s.userConf)
 	if err != nil {
 		s.log.Error(DiagnosticsEventID, fmt.Sprintf("failed to obtain unified configuration: %v", err))
 		return false, ERROR_FILE_NOT_FOUND
@@ -113,12 +112,6 @@ func (s *service) Execute(args []string, r <-chan svc.ChangeRequest, changes cha
 	// Set otel error handler
 	otel.SetErrorHandler(s)
 
-	err = self_metrics.CollectOpsAgentSelfMetrics(ctx, userUc, mergedUc)
-	if err != nil {
-		s.log.Error(DiagnosticsEventID, fmt.Sprintf("failed to collect ops agent self metrics: %v", err))
-		return false, ERROR_INVALID_DATA
-	}
-
 	return false, ERROR_SUCCESS
 }
 

cmd/ops_agent_uap_plugin/service_linux.go

@@ -191,13 +191,6 @@ func runSubagents(ctx context.Context, cancel context.CancelFunc, pluginInstallD
 	})
 
 	var wg sync.WaitGroup
-	// Starting the diagnostics service
-	runDiagnosticsCmd := exec.CommandContext(ctx,
-		path.Join(pluginInstallDirectory, DiagnosticsBinary),
-		"-config", OpsAgentConfigLocationLinux,
-	)
-	wg.Add(1)
-	go runSubAgentCommand(ctx, cancel, runDiagnosticsCmd, runCommand, &wg)
 
 	// Starting Otel
 	runOtelCmd := exec.CommandContext(ctx,

cmd/ops_agent_uap_plugin/service_windows.go

@@ -29,13 +29,11 @@ import (
 	"unsafe"
 
 	"github.com/GoogleCloudPlatform/ops-agent/apps"
-	"github.com/GoogleCloudPlatform/ops-agent/cmd/google_cloud_ops_agent_diagnostics/utils"
 	"github.com/GoogleCloudPlatform/ops-agent/confgenerator"
 	"github.com/GoogleCloudPlatform/ops-agent/internal/healthchecks"
 	"github.com/GoogleCloudPlatform/ops-agent/internal/logs"
 	"github.com/GoogleCloudPlatform/ops-agent/internal/self_metrics"
 	"github.com/kardianos/osext"
-	"go.opentelemetry.io/otel"
 	"golang.org/x/sys/windows"
 	"golang.org/x/sys/windows/svc/debug"
 	"golang.org/x/sys/windows/svc/eventlog"
@@ -168,8 +166,7 @@ func (ps *OpsAgentPluginServer) Start(ctx context.Context, msg *pb.StartRequest)
 		windowsEventLogger.Close()
 	}
 
-	otelErrorHandler := &otelErrorHandler{windowsEventLogger: windowsEventLogger, windowsEventId: OpsAgentUAPPluginEventID}
-	go runSubagents(pContext, cancelFunc, pluginInstallDir, pluginStateDir, runSubAgentCommand, ps.runCommand, otelErrorHandler)
+	go runSubagents(pContext, cancelFunc, pluginInstallDir, pluginStateDir, runSubAgentCommand, ps.runCommand)
 
 	return &pb.StartResponse{}, nil
 }
@@ -386,14 +383,9 @@ func createWindowsJobHandle() (windows.Handle, error) {
 //
 // cancel: the cancel function for the parent context. By calling this function, the parent context is canceled,
 // and GetStatus() returns a non-healthy status, signaling UAP to re-trigger Start().
-//
-// otelErrorHandler: an implementation of otel.ErrorHandler that is used in the diagnostics service to log otel errors to the Windows event log.
-func runSubagents(ctx context.Context, cancel context.CancelFunc, pluginInstallDirectory string, pluginStateDirectory string, runSubAgentCommand RunSubAgentCommandFunc, runCommand RunCommandFunc, otelErrorHandler otel.ErrorHandler) {
+func runSubagents(ctx context.Context, cancel context.CancelFunc, pluginInstallDirectory string, pluginStateDirectory string, runSubAgentCommand RunSubAgentCommandFunc, runCommand RunCommandFunc) {
 
 	var wg sync.WaitGroup
-	// Starting the diagnostics service
-	wg.Add(1)
-	go runDiagnosticsService(ctx, cancel, otelErrorHandler, &wg)
 
 	// Starting Otel
 	runOtelCmd := exec.CommandContext(ctx,
@@ -419,27 +411,6 @@ func runSubagents(ctx context.Context, cancel context.CancelFunc, pluginInstallD
 	wg.Wait()
 }
 
-func runDiagnosticsService(ctx context.Context, cancel context.CancelFunc, otelErrorHandler otel.ErrorHandler, wg *sync.WaitGroup) {
-	defer wg.Done()
-
-	userUc, mergedUc, err := utils.GetUserAndMergedConfigs(ctx, OpsAgentConfigLocationWindows)
-	if err != nil {
-		log.Printf("Failed to run the diagnostics service: %v", err)
-		cancel()
-		return
-	}
-
-	// Set otel error handler
-	otel.SetErrorHandler(otelErrorHandler)
-
-	err = self_metrics.CollectOpsAgentSelfMetrics(ctx, userUc, mergedUc)
-	if err != nil {
-		log.Printf("Failed to run the diagnostics service: %v", err)
-		cancel()
-		return
-	}
-}
-
 func runCommand(cmd *exec.Cmd) (string, error) {
 	if cmd == nil {
 		return "", nil

cmd/ops_agent_windows/main_windows.go

@@ -146,14 +146,6 @@ func initServices() error {
 				"--storage_path", fluentbitStoragePath,
 			},
 		},
-		{
-			fmt.Sprintf("%s-diagnostics", serviceName),
-			fmt.Sprintf("%s - Diagnostics", serviceDisplayName),
-			filepath.Join(base, fmt.Sprintf("%s-diagnostics.exe", serviceName)),
-			[]string{
-				"-config", filepath.Join(base, "../config/config.yaml"),
-			},
-		},
 	}
 	return nil
 }

confgenerator/agentmetrics.go

@@ -15,7 +15,10 @@
 package confgenerator
 
 import (
+	"context"
 	"fmt"
+	"path/filepath"
+	"time"
 
 	"github.com/GoogleCloudPlatform/ops-agent/confgenerator/otel"
 	"github.com/GoogleCloudPlatform/ops-agent/confgenerator/otel/ottl"
@@ -166,4 +169,28 @@ func (r AgentSelfMetrics) LoggingSubmodulePipeline() otel.ReceiverPipeline {
 	}
 }
 
+func OpsAgentSelfMetricsPipeline(ctx context.Context, outDir string) otel.ReceiverPipeline {
+	receiver_config := map[string]any{
+		"include": []string{
+			filepath.Join(outDir, "enabled_receivers_otlp.json"),
+			filepath.Join(outDir, "feature_tracking_otlp.json")},
+		"replay_file":   true,
+		"poll_interval": time.Duration(60 * time.Second).String(),
+	}
+	return otel.ReceiverPipeline{
+		Receiver: otel.Component{
+			Type:   "otlpjsonfile",
+			Config: receiver_config,
+		},
+		ExporterTypes: map[string]otel.ExporterType{
+			"metrics": otel.System,
+		},
+		Processors: map[string][]otel.Component{
+			"metrics": {
+				otel.Transform("metric", "datapoint", []ottl.Statement{"set(time, Now())"}),
+			},
+		},
+	}
+}
+
 // intentionally not registered as a component because this is not created by users

confgenerator/confgenerator.go

@@ -80,7 +80,7 @@ func (uc *UnifiedConfig) getOTelLogLevel() string {
 	return logLevel
 }
 
-func (uc *UnifiedConfig) GenerateOtelConfig(ctx context.Context) (string, error) {
+func (uc *UnifiedConfig) GenerateOtelConfig(ctx context.Context, outDir string) (string, error) {
 	p := platform.FromContext(ctx)
 	userAgent, _ := p.UserAgent("Google-Cloud-Ops-Agent-Metrics")
 	metricVersionLabel, _ := p.VersionLabel("google-cloud-ops-agent-metrics")
@@ -100,6 +100,12 @@ func (uc *UnifiedConfig) GenerateOtelConfig(ctx context.Context) (string, error)
 		ReceiverPipelineName: "otel",
 	}
 
+	receiverPipelines["ops_agent"] = OpsAgentSelfMetricsPipeline(ctx, outDir)
+	pipelines["ops_agent"] = otel.Pipeline{
+		Type:                 "metrics",
+		ReceiverPipelineName: "ops_agent",
+	}
+
 	receiverPipelines["fluentbit"] = AgentSelfMetrics{
 		Version: loggingVersionLabel,
 		Port:    fluentbit.MetricsPort,

confgenerator/confgenerator_test.go

@@ -279,7 +279,7 @@ func generateConfigs(pc platformConfig, testDir string) (got map[string]string,
 	}
 
 	// Otel configs
-	otelGeneratedConfig, err := mergedUc.GenerateOtelConfig(ctx)
+	otelGeneratedConfig, err := mergedUc.GenerateOtelConfig(ctx, "")
 	if err != nil {
 		return
 	}

confgenerator/config.go

@@ -1154,7 +1154,7 @@ func (uc *UnifiedConfig) OTelLoggingSupported(ctx context.Context) bool {
 		ucLoggingCopy.Logging.Service = &LoggingService{}
 	}
 	ucLoggingCopy.Logging.Service.OTelLogging = true
-	_, err = ucLoggingCopy.GenerateOtelConfig(ctx)
+	_, err = ucLoggingCopy.GenerateOtelConfig(ctx, "")
 	return err == nil
 }
 

confgenerator/files.go

@@ -59,7 +59,7 @@ func (uc *UnifiedConfig) GenerateFilesFromConfig(ctx context.Context, service, l
 			}
 		}
 	case "otel":
-		otelConfig, err := uc.GenerateOtelConfig(ctx)
+		otelConfig, err := uc.GenerateOtelConfig(ctx, outDir)
 		if err != nil {
 			return fmt.Errorf("can't parse configuration: %w", err)
 		}

confgenerator/testdata/goldens/all-backward_compatible_with_explicit_exporters/golden/linux-gpu/otel.yaml

@@ -448,6 +448,12 @@ processors:
   resourcedetection/_global_0:
     detectors:
     - gcp
+  transform/ops_agent_0:
+    error_mode: ignore
+    metric_statements:
+    - context: datapoint
+      statements:
+      - set(time, Now())
   transform/otel_1:
     error_mode: ignore
     metric_statements:
@@ -471,6 +477,12 @@ receivers:
       processes: {}
   nvml/hostmetrics_1:
     collection_interval: 60s
+  otlpjsonfile/ops_agent:
+    include:
+    - enabled_receivers_otlp.json
+    - feature_tracking_otlp.json
+    poll_interval: 1m0s
+    replay_file: true
   prometheus/fluentbit:
     config:
       scrape_configs:
@@ -517,6 +529,14 @@ service:
       - resourcedetection/_global_0
       receivers:
       - prometheus/fluentbit
+    metrics/ops_agent:
+      exporters:
+      - googlecloud
+      processors:
+      - transform/ops_agent_0
+      - resourcedetection/_global_0
+      receivers:
+      - otlpjsonfile/ops_agent
     metrics/otel:
       exporters:
       - googlecloud

confgenerator/testdata/goldens/all-backward_compatible_with_explicit_exporters/golden/linux/otel.yaml

@@ -424,6 +424,12 @@ processors:
   resourcedetection/_global_0:
     detectors:
     - gcp
+  transform/ops_agent_0:
+    error_mode: ignore
+    metric_statements:
+    - context: datapoint
+      statements:
+      - set(time, Now())
   transform/otel_1:
     error_mode: ignore
     metric_statements:
@@ -445,6 +451,12 @@ receivers:
         mute_process_exe_error: true
         mute_process_name_error: true
       processes: {}
+  otlpjsonfile/ops_agent:
+    include:
+    - enabled_receivers_otlp.json
+    - feature_tracking_otlp.json
+    poll_interval: 1m0s
+    replay_file: true
   prometheus/fluentbit:
     config:
       scrape_configs:
@@ -483,6 +495,14 @@ service:
       - resourcedetection/_global_0
       receivers:
       - prometheus/fluentbit
+    metrics/ops_agent:
+      exporters:
+      - googlecloud
+      processors:
+      - transform/ops_agent_0
+      - resourcedetection/_global_0
+      receivers:
+      - otlpjsonfile/ops_agent
     metrics/otel:
       exporters:
       - googlecloud

confgenerator/testdata/goldens/all-backward_compatible_with_explicit_exporters/golden/windows-2012/otel.yaml

@@ -431,6 +431,12 @@ processors:
   resourcedetection/_global_0:
     detectors:
     - gcp
+  transform/ops_agent_0:
+    error_mode: ignore
+    metric_statements:
+    - context: datapoint
+      statements:
+      - set(time, Now())
   transform/otel_1:
     error_mode: ignore
     metric_statements:
@@ -455,6 +461,12 @@ receivers:
         mute_process_exe_error: true
         mute_process_name_error: true
       processes: {}
+  otlpjsonfile/ops_agent:
+    include:
+    - enabled_receivers_otlp.json
+    - feature_tracking_otlp.json
+    poll_interval: 1m0s
+    replay_file: true
   prometheus/fluentbit:
     config:
       scrape_configs:
@@ -493,6 +505,14 @@ service:
       - resourcedetection/_global_0
       receivers:
       - prometheus/fluentbit
+    metrics/ops_agent:
+      exporters:
+      - googlecloud
+      processors:
+      - transform/ops_agent_0
+      - resourcedetection/_global_0
+      receivers:
+      - otlpjsonfile/ops_agent
     metrics/otel:
       exporters:
       - googlecloud