mirror of
https://github.com/microsoft/agent-framework.git
synced 2026-06-16 21:04:09 +08:00
Fix off-thread RunStatus race where GetStatusAsync can return Running after ResumeAsync halts
This commit is contained in:
@@ -77,12 +77,13 @@ internal sealed class StreamingRunEventStream : IRunEventStream
|
||||
|
||||
try
|
||||
{
|
||||
// Wait for the first input before starting
|
||||
// The consumer will call EnqueueMessageAsync which signals the run loop
|
||||
// Wait for the first input before starting.
|
||||
// The consumer will call EnqueueMessageAsync which signals the run loop.
|
||||
// Note: AsyncRunHandle also signals here on checkpoint resume when there are
|
||||
// already pending requests, so the first iteration can emit a PendingRequests
|
||||
// halt signal even without unprocessed messages.
|
||||
await this._inputWaiter.WaitForInputAsync(cancellationToken: linkedSource.Token).ConfigureAwait(false);
|
||||
|
||||
this._runStatus = RunStatus.Running;
|
||||
|
||||
while (!linkedSource.Token.IsCancellationRequested)
|
||||
{
|
||||
// Start a new run-stage activity for this input→processing→halt cycle
|
||||
@@ -95,6 +96,13 @@ internal sealed class StreamingRunEventStream : IRunEventStream
|
||||
// Events are streamed out in real-time as they happen via the event handler
|
||||
if (this._stepRunner.HasUnprocessedMessages)
|
||||
{
|
||||
// Flip to Running only when there's actual work to process.
|
||||
// This is intentionally inside the HasUnprocessedMessages branch so
|
||||
// that stale input signals cannot transiently flip status back to
|
||||
// Running after a prior halt has already been observed by callers
|
||||
// (e.g. Run.ResumeAsync returning after reading an Idle halt signal).
|
||||
this._runStatus = RunStatus.Running;
|
||||
|
||||
// Emit WorkflowStartedEvent only when there's actual work to process
|
||||
// This avoids spurious events on timeout-only loop iterations
|
||||
await this._eventChannel.Writer.WriteAsync(new WorkflowStartedEvent(), linkedSource.Token).ConfigureAwait(false);
|
||||
@@ -129,9 +137,6 @@ internal sealed class StreamingRunEventStream : IRunEventStream
|
||||
// Wait for next input from the consumer
|
||||
// Works for both Idle (no work) and PendingRequests (waiting for responses)
|
||||
await this._inputWaiter.WaitForInputAsync(linkedSource.Token).ConfigureAwait(false);
|
||||
|
||||
// When signaled, resume running
|
||||
this._runStatus = RunStatus.Running;
|
||||
}
|
||||
}
|
||||
catch (OperationCanceledException)
|
||||
|
||||
@@ -269,6 +269,29 @@ public class SampleSmokeTest
|
||||
_ = await Step9EntryPoint.RunAsync(writer, environment.ToWorkflowExecutionEnvironment());
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Stress regression for the off-thread run-status race: after
|
||||
/// <c>Run.ResumeAsync</c> returns at a halt boundary,
|
||||
/// callers must observe a stable terminal status and never a transient
|
||||
/// <see cref="RunStatus.Running"/>. Step9 is the canonical multi-response resume
|
||||
/// sample; prior to the fix in <see cref="Execution.StreamingRunEventStream"/>,
|
||||
/// its `runStatus.Should().Be(RunStatus.Idle)` assertion failed intermittently
|
||||
/// on roughly 1-in-10 iterations under InProcess_OffThread.
|
||||
/// </summary>
|
||||
[Fact]
|
||||
internal async Task Test_RunSample_Step9_OffThread_MultiResponseResume_StatusIsStableAsync()
|
||||
{
|
||||
const int Iterations = 50;
|
||||
|
||||
for (int i = 0; i < Iterations; i++)
|
||||
{
|
||||
using StringWriter writer = new();
|
||||
_ = await Step9EntryPoint.RunAsync(
|
||||
writer,
|
||||
ExecutionEnvironment.InProcess_OffThread.ToWorkflowExecutionEnvironment());
|
||||
}
|
||||
}
|
||||
|
||||
[Theory]
|
||||
[InlineData(ExecutionEnvironment.InProcess_Lockstep)]
|
||||
[InlineData(ExecutionEnvironment.InProcess_OffThread)]
|
||||
|
||||
Reference in New Issue
Block a user