Fix thread safety issues in WorkerNodeTelemetryData

[AR-May]

Related to #12867

Context

Fixing two bugs in telemetry infrastructure when using /m /mt (in-proc multithreaded) mode:

Thread-safety crash: all in-proc nodes share a single TelemetryForwarderProvider singleton. Multiple RequestBuilder instances run on dedicated threads and call AddTask/AddTarget concurrently on the same WorkerNodeTelemetryData dictionary fields, causing race conditions and dictionary corruption.

Nx telemetry duplication: In /m /mt mode, N BuildRequestEngine instances share one TelemetryForwarderProvider singleton. Each engine calls FinalizeProcessing on shutdown, sending the entire accumulated data each time. The InternalTelemetryConsumingLogger merges all N copies, inflating every counter N times.

Reproduction: 20+ non-SDK .NET Framework library projects + 1 exe referencing all of them, built with MSBuild.exe Repro.sln /m /mt.

Changes Made

Fix

Batch-then-merge in RequestBuilder: Each RequestBuilder now accumulates task/target telemetry into a local WorkerNodeTelemetryData instance (zero contention), then merges once into the shared state via elemetryForwarder.MergeWorkerData().

Thread-safe TelemetryForwarder: Added an internal lock protecting both MergeWorkerData and FinalizeProcessing. The forwarder is a singleton shared across BuildRequestEngine instances in /m /mt mode, so concurrent access is expected.

Swap-and-send in FinalizeProcessing: Instead of sending the same accumulated data on every call, FinalizeProcessing atomically swaps the internal data with a fresh empty instance under the lock, then sends only if non-empty. This ensures:

• First engine to finalize sends all data accumulated so far
• Subsequent engines find empty data and skip sending (no duplication)
• Late merges from other engines go into the new instance and are sent by the next FinalizeProcessing call (no
  data loss)

Testing

Locally tested that the issue is gone on a repro project.
Unit tests

[Copilot]

Pull request overview

This PR addresses a concurrency bug in MSBuild’s internal worker-node telemetry aggregation when running in in-proc multi-threaded mode (/m /mt), where multiple RequestBuilder instances can concurrently mutate shared telemetry dictionaries and corrupt their state.

Changes:

• Switch telemetry reporting in RequestBuilder.UpdateStatisticsPostBuild from per-target/per-task updates to batching into a local WorkerNodeTelemetryData and merging once.
• Replace ITelemetryForwarder.AddTask/AddTarget with a single MergeWorkerData API and update the provider implementations accordingly.
• Add explicit locking around aggregation in the internal telemetry-consuming logger.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/Framework/Telemetry/WorkerNodeTelemetryData.cs	Adds method documentation and minor refactor while preserving merge/aggregation behavior.
src/Build/TelemetryInfra/TelemetryForwarderProvider.cs	Replaces per-item update APIs with MergeWorkerData and exposes key creation helper for batching.
src/Build/TelemetryInfra/InternalTelemetryConsumingLogger.cs	Adds a lock around worker telemetry aggregation.
src/Build/TelemetryInfra/ITelemetryForwarder.cs	Updates forwarder contract to support batched merging.
src/Build/BackEnd/Components/RequestBuilder/RequestBuilder.cs	Implements local accumulation + single merge under a lock to prevent concurrent dictionary writes.

[AR-May]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[Copilot]

Pull request overview

Fixes telemetry thread-safety and counter inflation in in-proc multithreaded (/m /mt) builds by batching telemetry per RequestBuilder, merging into a shared forwarder under a lock, and preventing repeated “send the whole buffer” behavior during engine shutdown.

Changes:

• Accumulate task/target telemetry in a per-RequestBuilder WorkerNodeTelemetryData and merge once into the shared forwarder.
• Make TelemetryForwarder thread-safe and change finalization to “swap-and-send” to avoid Nx duplication across multiple BuildRequestEngine finalizers.
• Add unit tests for WorkerNodeTelemetryData.IsEmpty and forwarder reset behavior after FinalizeProcessing.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/Framework/Telemetry/WorkerNodeTelemetryData.cs	Adds IsEmpty and improves merge/add method clarity used by the forwarder swap-and-send logic.
src/Framework/Telemetry/TaskOrTargetTelemetryKey.cs	Introduces a helper factory (Create) to centralize key construction used by RequestBuilder.
src/Build/TelemetryInfra/TelemetryForwarderProvider.cs	Adds locking, batch merge entrypoint, and swap-and-send finalization to prevent races and duplicate sends.
src/Build/TelemetryInfra/ITelemetryForwarder.cs	Replaces per-item APIs with a batch merge API (MergeWorkerData).
src/Build/BackEnd/Components/RequestBuilder/RequestBuilder.cs	Switches to batch-then-merge telemetry collection to remove dictionary contention.
src/Build.UnitTests/Telemetry/Telemetry_Tests.cs	Adds tests for the new reset/empty behavior and forwarder finalization semantics.