[perf-improver] perf: cache distinct processors in AsynchronousMessageBus to eliminate per-drain allocation

[Evangelink]

🤖 This is an automated contribution from Perf Improver.

Goal and Rationale

AsynchronousMessageBus.DrainDataAsync() allocated a fresh Dictionary<IAsyncConsumerDataProcessor, long> on every call solely to track how many items each processor had received between drain rounds. Since the set of processors is fixed after InitAsync, this dictionary can be replaced with a cached array pair (IAsyncConsumerDataProcessor[] + long[]) allocated once.

Additionally, DisableAsync and Dispose iterated _dataTypeConsumers.Values with a nested loop — because a consumer subscribed to N data types appears in N separate lists, each processor was being completed/disposed N times. Using the deduplicated _distinctProcessors array fixes this.

Approach

1. Add two new fields initialized at end of InitAsync:

   • _distinctProcessors = [.. _consumerProcessor.Values] — the unique set of processors
   • _drainLastReceived = new long[_distinctProcessors.Length] — reusable count snapshot

2. DrainDataAsync: replace new Dictionary<IAsyncConsumerDataProcessor, long>() + hash-table lookups with direct array indexing over _distinctProcessors.

3. DisableAsync / Dispose: replace double nested foreach over _dataTypeConsumers.Values with a single foreach over _distinctProcessors.

Performance Evidence

Metric	Before	After
Dictionary<IAsyncConsumerDataProcessor, long> allocated per drain	1	0
Processor visits in DisableAsync (N consumers × M types)	N×M	N
Processor visits in Dispose (N consumers × M types)	N×M	N
Inner-loop access pattern in drain	hash-table lookup	array index (cache-friendly)

DrainDataAsync is called once at the end of every test run. The dictionary allocation and hash operations are eliminated; the reused long[] snapshot costs only a stack read per iteration.

Methodology: code inspection + allocation diff (directly verifiable from the diff).

Trade-offs

• Two new fields per AsynchronousMessageBus instance (one pointer + one pointer — negligible).
• No behavioral change for DrainDataAsync.
• DisableAsync / Dispose now call CompleteAddingAsync / Dispose exactly once per processor (vs. once per (processor, data-type) pair previously) — this is strictly more correct.

Test Status

• Microsoft.Testing.Platform.UnitTests (net8.0): 992 passed, 0 failed, 3 skipped ✅

Reproducibility

./build.sh
artifacts/bin/Microsoft.Testing.Platform.UnitTests/Debug/net8.0/Microsoft.Testing.Platform.UnitTests

Generated by Perf Improver · sonnet46

Generated by Perf Improver · sonnet46 3.1M · ◷

Add this agentic workflows to your repo

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/perf-improver.md@main

[Copilot]

Pull request overview

This PR optimizes AsynchronousMessageBus by caching the distinct consumer processors after initialization and reusing that cache for draining, disabling, and disposal.

Changes:

• Adds cached _distinctProcessors and _drainLastReceived arrays initialized in InitAsync.
• Replaces per-drain dictionary allocation with indexed array snapshots.
• Updates DisableAsync and Dispose to iterate distinct processors once.

Show a summary per file

File	Description
src/Platform/Microsoft.Testing.Platform/Messages/AsynchronousMessageBus.cs	Caches distinct processors and reuses arrays for drain/disable/dispose paths.

Copilot's findings

• Files reviewed: 1/1 changed files
• Comments generated: 2

[Evangelink]

@copilot address review comments

[Copilot]

@copilot address review comments

Both review comments addressed in commit fix: clear _distinctProcessors in Dispose and add multi-type consumer regression test:

1. Dispose retains references / double-dispose: _distinctProcessors = [] and _drainLastReceived = [] are now cleared alongside the existing dictionary clears.

2. Missing multi-type consumer test: Added DisableAsync_WithConsumerSubscribedToMultipleDataTypes_ShouldCompleteProcessorOnce — a consumer subscribed to DataTypeA and DataTypeB publishes one of each, drains, asserts both were received, then calls DisableAsync. All 5 AsynchronousMessageBus tests pass.

[Evangelink]

/review

[github-actions]

⚠️ Security scanning failed for Expert Code Review (command). Review the logs for details.

Expert-reviewer launched in background for PR #8704. It will post the review directly.