OTel experimentation

[baronfel]

Fixes #49668

This is entirely exploratory so far - we can't use AppInsights libraries if we do AOT in the SDK, so I'm exploring what an OTel-based way of doing telemetry might look like.

• All CLI calls make at least one Activity
• TrackEvent now creates Events on the ambient Activity with the same content as before

Q: telemetry filtering/hashing
A: for now we should keep everything as it is. As we make new activities, we should use Activity.IsAllDataRequested as a check before adding/computing expensive data.

Q: do we keep the store-out-of-band-then-rehydrate-and-send approach, or just do a best-case flush?
A: We can be very optimistic! We can have multiple collectors, each for different purposes. The default OTel collector can be for local use only (by default) and can be verbose, while the Azure Monitor collector can be for production only and be more restrained. Each Collector has its own implementation, and the Azure Monitor collector even handles durable message persistence, so we can delete all of our current implementation. We should still have it store its durable store to our owned directory, though.

Example Aspire Dashboard outputs for dotnet new classlib:

Telemetry events are visible in this view if the CLI is opted in to telemetry.

[KalleOlaviNiemitalo]

Overeager find & replace 😸

[baronfel]

Exactly right 😄 Part of me cutting out parts of this PR will involve minimizing those diffs

[baronfel]

Rebased on top of #49749

[Copilot]

Pull Request Overview

This PR explores replacing Application Insights telemetry with OpenTelemetry (OTel) for the .NET CLI. The change removes the AppInsights dependency and custom persistence channel implementation, switching to OTel's Azure Monitor exporter with built-in durable storage. This enables AOT compilation support while maintaining telemetry functionality.

Key Changes:

• Replaced Application Insights with OpenTelemetry SDK, adding Azure Monitor and OTLP exporters
• Removed custom persistence channel infrastructure (storage, transmission, sender classes)
• Converted telemetry events to OTel Activities with events
• Implemented activity context propagation to child processes via environment variables

Reviewed Changes

Copilot reviewed 45 out of 46 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
src/Cli/dotnet/dotnet.csproj	Added OpenTelemetry package references
src/Cli/dotnet/Program.cs	Major refactor: initialized OTel providers, removed AppInsights code, added activity tracking and signal handlers
src/Cli/dotnet/Telemetry/Telemetry.cs	Simplified telemetry to use Activities instead of AppInsights TelemetryClient
src/Cli/dotnet/Telemetry/TelemetryCommonProperties.cs	Updated return types to support nullable object values
src/Cli/dotnet/Telemetry/ITelemetry.cs	Removed Flush/Dispose methods, updated TrackEvent signature
src/Cli/dotnet/Telemetry/PersistenceChannel/*	Deleted entire persistence channel implementation
test/dotnet.Tests/TelemetryTests/*	Removed tests for deleted persistence channel
test/dotnet.Tests/TelemetryCommandTest.cs	Updated test calls removing startup time parameter
src/Cli/dotnet/Commands/MSBuild/MSBuildLogger.cs	Removed sentinel dependency, simplified telemetry initialization
src/Cli/dotnet/CommandFactory/CommandResolution/ActivityContext.cs	Added new class for propagating activity context to child processes

[Copilot]

Variable 'exitCode' is used but not declared in this scope. The exit code should be captured from the catch block above (line 127 returns hardcoded 1). Consider adding int exitCode = 1; before this line or removing this tag addition from the catch block.

[Copilot]

Variable 'exitCode' is used but not declared in this scope. The exit code should be captured from the catch block (line 136 returns hardcoded 1). Consider adding int exitCode = 1; before this line or removing this tag addition from the catch block.

[Copilot]

This test method is a duplicate of the test at lines 158-161. The method name and implementation are identical, which will cause test execution issues. One of these duplicate tests should be removed.

Suggested change

	[Fact]
	public void TelemetryCommonPropertiesShouldReturnIsLLMDetection()
	{
	var unitUnderTest = new TelemetryCommonProperties(getMACAddress: () => null, userLevelCacheWriter: new NothingCache());
	unitUnderTest.GetTelemetryCommonProperties("dummySessionId")["llm"].Should().BeOneOf("claude", null);
	}

[Copilot]

This is a new file with #nullable disable at the top, which is discouraged for new code. Consider enabling nullable reference types and properly annotating nullability throughout the file.

[Copilot]

The method returns FrozenDictionary<string, object?> but creates a Dictionary<string, object?>. The dictionary should be converted to a frozen dictionary by calling .ToFrozenDictionary() on the dictionary before returning to match the return type and provide the immutability/performance benefits of FrozenDictionary.

[Copilot]

The parameter 'eventName' is now non-nullable (removed the ?), but the method body still checks for null at line 18 with if (string.IsNullOrEmpty(eventName)). Either the parameter should remain nullable with string?, or the null check should be removed since non-nullable parameters cannot be null when nullable reference types are enabled.