grpc/grpc-go · error

failed to instrument OpenCensus: %v

Error message

failed to instrument OpenCensus: %v

What it means

A wrapper returned by Start() when startOpenCensus(config) fails. The inner error (interpolated via %v) is usually one of: Stackdriver exporter creation failure, view registration failure, or exporter type-assertion failure. Because Start() defers End() on error, partial resources are cleaned up, but no telemetry instrumentation is applied to subsequently created clients/servers.

Source

Thrown at gcp/observability/observability.go:72

		return fmt.Errorf("no ObservabilityConfig found")
	}

	// Set the project ID if it isn't configured manually.
	if err = ensureProjectIDInObservabilityConfig(ctx, config); err != nil {
		return err
	}

	// Cleanup any created resources this function created in case this function
	// errors.
	defer func() {
		if err != nil {
			End()
		}
	}()

	// Enabling tracing and metrics via OpenCensus
	if err = startOpenCensus(config); err != nil {
		return fmt.Errorf("failed to instrument OpenCensus: %v", err)
	}

	if err = startLogging(ctx, config); err != nil {
		return fmt.Errorf("failed to start logging: %v", err)
	}

	// Logging is controlled by the config at methods level.
	return nil
}

// End is the clean-up API for gRPC Observability plugin. It is expected to be
// invoked in the main function of the application. The suggested usage is
// "defer observability.End()". This function also flushes data to upstream, and
// cleanup resources.
//
// Note: this method should only be invoked once.
func End() {
	stopLogging()

View on GitHub (pinned to 03255a9237)

Solutions

1. Inspect the wrapped error string after the colon; it identifies the underlying cause (project ID, auth, duplicate view registration).
2. Ensure Application Default Credentials are available: run `gcloud auth application-default login` or set GOOGLE_APPLICATION_CREDENTIALS.
3. Confirm config.ProjectID is valid and the account has monitoring/trace write scopes.
4. Call Start() exactly once per process; do not re-invoke after End().

Example fix

// before
if err := observability.Start(ctx); err != nil {
    log.Fatal(err) // "failed to instrument OpenCensus: ..."
}
// after
if err := observability.Start(ctx); err != nil {
    log.Fatalf("observability disabled, continuing without telemetry: %v", err)
}
// and ensure ADC is present: gcloud auth application-default login

if err := observability.Start(ctx); err != nil {
    if strings.Contains(err.Error(), "failed to instrument OpenCensus") {
        log.Printf("running without tracing/metrics: %v", err)
    } else {
        return err
    }
}

When it happens

Trigger: Invoking Start() with a config that enables cloud_trace or cloud_monitoring but whose ProjectID cannot be resolved, or whose Stackdriver exporter cannot authenticate. Also when default views have already been registered by a prior Start() call in the same process.

Common situations: Running on a machine without Application Default Credentials (no GOOGLE_APPLICATION_CREDENTIALS and no metadata server); specifying a non-existent GCP project ID; calling Start() twice (e.g. in a test binary and in the app).

Related errors

• failed to start logging: %v
• unable to create CloudLogging exporter: %v
• no ObservabilityConfig found
• failed to create Stackdriver exporter: %v
• invalid method string: %v, err: %v

AI-assisted analysis of grpc/grpc-go@03255a9237 (2026-08-07). Data as JSON: /api/errors/b8c114056a7e1aeb. Report an issue: GitHub.