[glen-84]

Summary

• The saga IntegrationTests slept for a fixed time (Task.Delay) between steps and before asserting on store state, which flaked under CI load: the Main Branch Coverage run had Saga_Should_TimeoutWithCustomResponse fail with Expected: 0, Actual: 1.
• A saga dispatches a transition's messages before its state is committed, and a correlated message that arrives before that commit finds no instance and is discarded, so no fixed delay makes these tests reliable. Each test now waits until the saga state its next message depends on is persisted before sending that message, then polls for the expected state or for the instance to be deleted, bounded by the existing 10 s timeout.
• Saga_Should_ExecuteThroughSteps waits for Started to be persisted before the first TriggerEvent, and for the saga to finish consuming that event before sending the second. The in-memory saga store exposes a transition's new state before it commits, so the end of the consume, which comes after the commit, is the reliable signal. The timeout tests wait until the saga is persisted before sending SagaTimedOutEvent.
• Saga_Should_SupportRequestResponse uses a new GatedTriggerRequestHandler that holds the reply, parking the saga in AwaitingResponse until its persisted state is observed; releasing the reply then drives finalization and deletion.

Test plan

• IntegrationTests: 10/10 passing on net8.0, net9.0, net10.0, and net11.0.
• Full Mocha.Sagas.Tests suite: 612 passed, 4 skipped, 0 failed across the four target frameworks.
• With a temporary local saga store whose commit takes 2 s, the four tests still pass. With one that holds back a transition's commit until the next transition has committed, Saga_Should_ExecuteThroughSteps passes.

[Copilot]

Pull request overview

Stabilizes Mocha saga integration tests by replacing fixed Task.Delay waits with deterministic synchronization against persisted saga state, reducing CI timing flakes.

Changes:

• Replaced fixed delays with bounded polling that waits for specific saga persistence/deletion conditions.
• Ensured timeout tests only send SagaTimedOutEvent after the saga instance is persisted.
• Added a gated request handler to hold replies so the AwaitingResponse persisted state can be observed deterministically.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-code-quality]

Code Coverage Overview

Languages: C#

C# / code-coverage/dotnet

The overall coverage in the gai/stabilize-saga-i... branch remains at 49%, unchanged from the main branch.

Show a code coverage summary of the most impacted files.

File	main 443ea28	gai/stabilize-saga-i... b9841b3	+/-
/home/runner/wo...ageProcessor.cs	100%	89%	-11%
/home/runner/wo...toResetEvent.cs	73%	68%	-5%
/home/runner/wo...eiveEndpoint.cs	72%	69%	-3%
/home/runner/wo...DefaultTopic.cs	89%	90%	+1%
/home/runner/wo...efaultPubSub.cs	89%	91%	+2%
/home/runner/wo...ueryExecutor.cs	65%	67%	+2%
/home/runner/wo...tor/Mediator.cs	89%	91%	+2%
/home/runner/wo...PromiseCache.cs	78%	81%	+3%
/home/runner/wo...entsListener.cs	62%	66%	+4%
/home/runner/wo...heMiddleware.cs	86%	92%	+6%

---

_{Code Coverage is in Public Preview. Learn more and provide us with your feedback.}

[Copilot]

🟡 Changes recommended

A persistence poll can expire before an event is sent, and a failed assertion can leave the gated handler blocked.

2 open findings

• Assert Started state before publishing the first TriggerEvent · New
• Release request gate in finally when test assertions fail · New

3 resolved since last review

• Unnecessary cast to MessagingRuntime: IMessagingRuntime already exposes Naming, so prefer resolving…
• Unnecessary cast to MessagingRuntime: IMessagingRuntime already provides the Naming property, so…
• Unnecessary cast to MessagingRuntime here: IMessagingRuntime already exposes Naming via…

🧠 Review effort: Lite

[Copilot]

🔵 Needs a closer look

The step-through test can still observe a state change before its transaction commits, leaving a timing race unresolved.

0 open findings

2 resolved since last review

• Release request gate in finally when test assertions fail
• Assert Started state before publishing the first TriggerEvent

Previously missed (1)

In code that hasn't changed since last review

Wait for first transition commit before publishing second event

src/​Mocha/​test/​Mocha.Sagas.Tests/​IntegrationTests.cs:71

Wait for the first transition's commit, not just State == "Triggered". The in-memory store loads the object already held by InMemorySagaStateStorage, and Saga.HandleEvent changes its State before the consumer commits. This poll can therefore pass while the first save is still in flight; the second event may finish and delete the saga before that save commits, leaving a state behind and making this test flaky. Use a commit-visible signal or snapshot before publishing the second event.

🧠 Review effort: Lite

[Copilot]

🟡 Changes recommended

One wait ignores its timeout result, so the test can proceed without confirming the first transition committed.

1 open finding

• Assert first consume completed before sending the second event · New

🧠 Review effort: Lite

[Copilot]

🟡 Changes recommended

The step-through test exceeds the repository’s hard limit of five assertions per test.

1 open finding

• Test exceeds five-assertion limit · New

1 resolved since last review

• Assert first consume completed before sending the second event

🧠 Review effort: Lite