Azure Application Insights #

What? #

This is a Dart library that integrates with Azure's Application Insights service. It allows you to push telemetry items into your Application Insights instance.

Why? #

Application Insights is a powerful tool for understanding how your software system is operating. In the context of Flutter applications, it allows you to deeply understand user interactions, performance, and faults in your application.

Where? #

Firstly, install the azure_application_insights package, then you can import it into your Dart code as follows:

import 'package:azure_application_insights/azure_application_insights.dart';

How? #

The only piece of information you need to integrate with Application Insights is your connection string, which you can find inside the Azure portal. Open your Application Insights resource and look in the Overview tab.

Once you have your connection string, you can construct a TelemetryClient as follows:

final processor = BufferedProcessor(
  next: TransmissionProcessor(
    connectionString: connectionString,
    httpClient: client,
    timeout: const Duration(seconds: 10),
  ),
);

final telemetryClient = TelemetryClient(
  processor: processor,
);

This is a typical setup where telemetry items are buffered before being transmitted. Depending on your processing needs, you may have a need for more than one TelemetryClient in your application. For example, you might have one TelemetryClient that buffers telemetry items and is used for all telemetry other than errors, and a second that does not buffer and is used only to submit errors as promptly as possible. Please review the example code and API docs for alternative configurations.

Once you have a TelemetryClient, you can simply invoke the various methods to capture telemetry items:

telemetryClient.trackTrace(
  severity: Severity.information,
  message: 'Hello from Dart!',
);

You may also want to configure additional properties to be submitted with telemetry items. These can be configured in two places:

On the TelemetryContext associated with your TelemetryClient. Properties in this context object will be attached to every telemetry item submitted. Moreover, you can share a TelemetryContext between multiple TelemetryClient instances if desired.
Every method on TelemetryClient allows you to specify additionalProperties that will be captured only for that telemetry item. As the name suggests, these properties are in addition to those within the context.

Here are examples of both:

final telemetryClient = ...;

// 1. Properties associated with the TelemetryClient will be attached to
// all telemetry items submitted via that client.
telemetryClient.context
    // These built-in helpers simply set the pre-defined properties
    // Application Insights provides.
    ..applicationVersion = 'my version'
    ..device.type = 'Android';
    // But you can also set whatever property name you like.
    ..properties['environmentName'] = 'dev';

// 2. Additional properties can be bundled with individual telemetry items.
telemetryClient.traceTrace(
    severity: Severity.information,
    message: 'An example',
    additionalProperties: <String, Object>{
        'answer': 42,
    },
);

Of course, you can leverage whatever data sources and third party libraries make sense in order to populate properties. Typically you would use a package like device_info_plus to obtain information on the device and fill in the appropriate properties on the context.

Flutter Integration #

To submit crashes in Flutter applications as telemetry, follow the following recipe:

void main() {
    // You probably don't want to always run with crash reporting because it interferes with the normal
    // debug/development experience. Here we use kReleaseMode to only enable crash reporting for release builds, but
    // you can use whatever criteria and mechanism you like.
    if (kReleaseMode) {
        runWithCrashReporting(codeToExecute: run);
    } else {
        run();
    }
}

void run() => runApp(MyApp());

Future<void> runWithCrashReporting({
  required VoidCallback codeToExecute,
}) async {
  // Hook into Flutter error handling.
  FlutterError.onError = (details) => submitErrorAsTelemetry(
        isFatal: true,
        error: details.exception,
        trace: details.stack,
      );

  // Run the code to execute in a zone and handle all errors within.
  runZonedGuarded(
    codeToExecute,
    (error, trace) => submitErrorAsTelemetry(
      isFatal: true,
      error: error,
      trace: trace,
    ),
  );
}

Future<void> submitErrorAsTelemetry({
  required bool isFatal,
  required Object error,
  required StackTrace trace,
}) async {
  debugPrint('reporting ${isFatal ? 'fatal' : 'non-fatal'} error: $error');
  debugPrint('$trace');

  try {
    // Get your TelemetryClient instance here, perhaps by DI or some other mechanism.
    final telemetryClient = ...;

    // Get any additional properties for the crash report here, such as device information.
    final errorProperties = ...;

    // Write an error telemetry item.
    telemetryClient.trackError(
      error: error,
      stackTrace: trace,
      severity: isFatal ? Severity.critical : Severity.error,
      additionalProperties: errorProperties,
    );

    if (isFatal) {
      await telemetryClient.flush();
    }
  } on Object catch (e, t) {
    // We print synchronously here to ensure the output is written in the case we force exit.
    debugPrintSynchronously('Sending error telemetry failed: $e\r\n$t');
    debugPrintSynchronously('Original error: $error');
  } finally {
    if (isFatal && kReleaseMode) {
      debugPrintSynchronously('Forcing exit');
    }
  }
}

HTTP Middleware #

Whilst you can certainly use TelemetryClient.trackDependency to track HTTP requests yourself, you can save time by using the TelemetryHttpClient HTTP client. This class will automatically time requests and forward on the details of completed requests to a given TelemetryClient as dependency telemetry items.

Typical usage looks like this:

final telemetryHttpClient = TelemetryHttpClient(
  telemetryClient: telemetryClient,
  inner: Client(),
);

// This request will be captured as a dependency telemetry item.
await telemetryHttpClient.get('https://kent-boogaart.com/');

The TelemetryHttpClient could be one piece in a long chain of HTTP middleware - it is composable.