Monitor App Performance

Question 1

What is App Insights (AI)?

Answer 1

• Extension of Azure Monitor and provides app performance monitoring features
• Helps us to understand how app is performing and reactivity review app execution data to determine the cause of an incident
• Stores app trace logging data
• Free to sign up and if you choose basic pricing plan of AI there is no charge until your app has grown to have substantial usage

Question 2

What features does AI provide?

Answer 2

• live metrics
• availability checks via probes
• Github/DevOps integration
• Usage
• Smart Detection
• Application Map
• Distributed Tracing

Question 3

What does AI monitor?

Answer 3

• Request rates, response times and failure rates (find out most popular pages and at what times)
• dependency rates, response times and failure rates (find out if external services slowing us down)
• exceptions
• page views
• AJAX calls
• user and session counts
• host diagnostics
• custom events and metrics

Question 4

What type of metrics does AI provide? Brief Benefits of each?

Answer 4

• Log based = translated into kutso queries from stored events
• Standard = stored as preaggregated time series
• Standard have better performance at query time and a better choice for real-time alerting + dashboarding
• Log-based have more dimensions which makes them the superior option for data analysis and ad-hoc diagnostics
• In the UX standard are called “Standard metrics (preview)” and traditional metrics from events are renamed “Log-based metrics”

Question 5

In more detail what are log based metrics?

Answer 5

• devs can use the SDK to send events manually or rely on the auto collection of events from autoinstrumentation
• All events stored as logs
• Using logs to retain a complete set of events brings great analytical value e.g. can get an exact count of request to a URL with the number of distinct users who made the calls
• however cna result in large volume of telemetry which requires filtering to help reduce

Question 6

What is Automatic instrumentation?

Answer 6

• enables telemetry through config without touching the apps code
• tends to be less configurable and not available in all languages
• Easiest way to enable AI

Question 7

In more detail what are standard/pre-aggregated metrics?

Answer 7

• arent stored as individual events
• stored as preaggregated time series and only with key dimensions making them better at query time
• retrieving data happens faster and requires less power
• Newer SDKs preaggregate metrics during collection, reducing volume of data transmitted over wire
• older ones require AI to populate the new metrics by aggregating the events received by the AI collection endpoint, doesnt reduce volume but still has better performance in real-time than SDKs that dont preaggregate

Question 8

What is manual instrumentation?

Answer 8

• coding against AI or open Telemetry API
• in context of user refers to installing a language specific SDK in an app
• have to manage the updates to latest package version by yourself
• this in itself has 2 methods, AI SDKs and AM OpenTelemetry Distros

Question 9

How can we enable AI via SDKs?

Answer 9

• Only need to install AI SDK when you require custom events and metrics, require control over the flow of telemetry or auto-instrumentation isn’t available (due to lang or platform issues)
• Install a package in your app and then instrument the web app, any background components and JS within web pages
• App and components don’t have to be hosted within Azure
• Instrumentation monitors app and directs the telemetry data to an AI resource by using a token

Question 10

How do we enable AI via openTelemetry?

Answer 10

• Working with OpenCensus and OpenTracing MS helped to create a single project, OpenTelemetry
• Includes contributions from all major APM vendors and lives within the CNCF which MS is a platinum member of

Question 11

How can we setup recurring tests to monitor availability and responsivness?

Answer 11

• AI sends web requests to your app at regular intervals from points around the world and can alert you if your app is responding slowly, or not at all
• Can create 100 tests per app insights resource
• Require no code changes and work with HTTP/S publicly accessible endpoint

Question 12

What are the three types of availability test?

Answer 12

• Standard test = Checks the availability of a website by sending a single request, also includes the TLS/SSL certificate validity, proactive lifetime check, HTTP request verb (GET, POST etc), custom headers and custom data associated with your HTTP request
  Custom TrackAvailability test = If you decide to create custom app to run availability tests you can use the TrackAvailabilty() method to send the results to AI
• URL ping test (classic) = Deprecated from 2026, Single url ping endpoint

Question 13

What is an application map?

Answer 13

• Application Map helps you spot performance bottlenecks or failure hotspots across all components of your distributed app
• Each node on the map represents an app component or its dependencies and has health key performance indicator and alerts status
• Can select component for more details diagnostics
• Enables us to visualise complex topologies with hundreds of components

Question 14

What are components in application map?

Answer 14

• Independently deployable parts of your distributed/microservice application
• different from observed external dependencies such as SQL, Event hubs etc which your org may not have access too
• can run on any number of server/role/container instances
• can be separate AI instrumentation keys or different roles reporting to a single AI key

Question 15

How does application map find components?

Answer 15

• By following HTTP dependency calls made between servers with the AI SDK installed
• When you first load the app a set of queries are triggered to discover the components related to this component
• Selecting Update Map refreshes all the components discovered up to that point
• If all components are roles within a single AI resource then this discovery step isn’t required and the initial load has all the components