Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,182 @@
using System;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;

using FluentAssertions;
using Xunit;

namespace DynamicData.Tests.Cache;

public static partial class SuspendNotificationsFixture
{
public sealed class IntegrationTests
: IntegrationTestFixtureBase
{
[Fact]

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Any particular reason to move this?

public void ResumeDeliversPendingChangesWithoutHoldingTheLock()
{
// On resume, the changes accumulated while suspended must be delivered to
// subscribers WITHOUT the cache lock held. A subscriber that blocks mid-delivery
// must therefore not stall an unrelated, lock-requiring operation on another
// thread. If the lock were held during delivery, the concurrent operation would
// block for the full duration of the blocked subscriber.
//
// Dedicated threads are used (rather than the thread pool) so the test is immune
// to pool starvation when run alongside other test collections.
using var cache = new SourceCache<int, int>(static x => x);
using var deliveryStarted = new ManualResetEventSlim(false);
using var releaseDelivery = new ManualResetEventSlim(false);
using var concurrentOpDone = new ManualResetEventSlim(false);

// An active subscriber (connected before suspension) that blocks on its first delivery.
var blockOnce = true;
using var slowSub = cache.Connect().Subscribe(_ =>
{
if (blockOnce)
{
blockOnce = false;
deliveryStarted.Set();
releaseDelivery.Wait(TimeSpan.FromSeconds(30));
}
});

var suspend = cache.SuspendNotifications();
cache.AddOrUpdate(Enumerable.Range(0, 10));

// Resume on a background thread: the accumulated changes are delivered to the
// slow subscriber, which blocks partway through.
var resumeThread = new Thread(suspend.Dispose) { IsBackground = true };
resumeThread.Start();
deliveryStarted.Wait(TimeSpan.FromSeconds(10)).Should().BeTrue("delivery of pending changes should have started");

// While delivery is blocked, a concurrent lock-requiring operation must complete
// promptly. It cannot if the delivery is happening under the cache lock. The
// distinguishing gap is large (milliseconds if the lock is free, ~30s if held),
// so a generous threshold stays reliable under cold-start JIT and heavy load.
var concurrentThread = new Thread(() =>
{
cache.SuspendNotifications().Dispose();
concurrentOpDone.Set();
}) { IsBackground = true };
concurrentThread.Start();

var completedWhileBlocked = concurrentOpDone.Wait(TimeSpan.FromSeconds(10));

releaseDelivery.Set();
resumeThread.Join(TimeSpan.FromSeconds(30)).Should().BeTrue("resume should complete");
concurrentThread.Join(TimeSpan.FromSeconds(30)).Should().BeTrue("concurrent operation should complete");

completedWhileBlocked.Should().BeTrue("a concurrent operation must not block while pending changes are delivered; the lock must not be held during delivery");
cache.Count.Should().Be(10, "all items should be present after resume");
}

[Fact]
public void StaleResumeSignalIsSuppressedByConcurrentReSuspend()
{
// Deterministic reproduction of the suspend/resume state-divergence race (#1131).
// Resume decrements the suspend count in one step and emits its resume signal in a
// later step. If a re-suspend slips in between, the resume signal must NOT fire:
// otherwise the suspended-notification subject would say "resumed" while the count
// says "suspended", and a connection made during the re-suspension would wrongly
// activate and receive data while notifications are suspended.
//
// The interleaving is forced deterministically: an active subscriber blocks the
// resume thread inside delivery, after the count has been decremented to zero but
// before the resume signal, giving the main thread a window to re-suspend and connect.
using var cache = new SourceCache<int, int>(static x => x);
var dataSet = Enumerable.Range(0, 50).ToList();

using var deliveryStarted = new ManualResetEventSlim(false);
using var releaseDelivery = new ManualResetEventSlim(false);

var blockOnce = true;
using var activeSub = cache.Connect().Subscribe(_ =>
{
if (blockOnce)
{
blockOnce = false;
deliveryStarted.Set();
releaseDelivery.Wait(TimeSpan.FromSeconds(30));
}
});

var suspend1 = cache.SuspendNotifications();
cache.AddOrUpdate(dataSet);

// Resume on a background thread: it decrements the count to zero and, while delivering
// the accumulated changes to the blocking subscriber, parks BEFORE the resume signal.
var resumeThread = new Thread(suspend1.Dispose) { IsBackground = true };
resumeThread.Start();
deliveryStarted.Wait(TimeSpan.FromSeconds(10)).Should().BeTrue("delivery should have started");

// The resume thread is parked after decrementing the count but before signalling
// resume. Re-suspend and connect a new subscriber while the count is transiently zero.
var suspend2 = cache.SuspendNotifications();
using var lateResults = cache.Connect().AsAggregator();

// Let the resume thread proceed to its now-stale resume signal.
releaseDelivery.Set();
resumeThread.Join(TimeSpan.FromSeconds(30)).Should().BeTrue("resume should complete");

// The late subscriber connected while re-suspended: it must NOT have activated,
// because notifications ARE suspended (suspend2 is active). The stale resume signal
// must be suppressed by re-checking the suspend count.
lateResults.Messages.Count.Should().Be(0, "a connection made during re-suspension must not activate on a stale resume signal");
lateResults.Data.Count.Should().Be(0, "no data should be delivered while suspended");

// Releasing the real suspension delivers the data normally.
suspend2.Dispose();
lateResults.Data.Count.Should().Be(dataSet.Count, "all data should arrive once truly resumed");
lateResults.Messages.Count.Should().Be(1, "a single changeset on the real resume");
}

[Fact]
public async Task ConcurrentSuspendDuringResumeDoesNotCorrupt()
{
// Stress test: races resume against re-suspend on two threads.
// Both orderings are correct (tested deterministically above).
// This test verifies no corruption, deadlocks, or data loss under contention.
const int iterations = 200;
var dataSet1 = Enumerable.Range(0, 100).ToList();
var dataSet2 = Enumerable.Range(1000, 100).ToList();
var allData = dataSet1.Concat(dataSet2).ToList();

for (var iter = 0; iter < iterations; iter++)
{
using var cache = new SourceCache<int, int>(static x => x);

var suspend1 = cache.SuspendNotifications();
cache.AddOrUpdate(dataSet1);
using var results = cache.Connect().AsAggregator();

using var barrier = new Barrier(2);
var resumeTask = Task.Run(() =>
{
barrier.SignalAndWait();
suspend1.Dispose();
});

var reSuspendTask = Task.Run(() =>
{
barrier.SignalAndWait();
return cache.SuspendNotifications();
});

await Task.WhenAll(resumeTask, reSuspendTask);
var suspend2 = await reSuspendTask;

cache.AddOrUpdate(dataSet2);
suspend2.Dispose();

results.Summary.Overall.Adds.Should().Be(allData.Count, $"iteration {iter}: exactly {allData.Count} adds");
results.Summary.Overall.Removes.Should().Be(0, $"iteration {iter}: no removes");
results.Summary.Overall.Updates.Should().Be(0, $"iteration {iter}: no updates because keys don't overlap");
results.Data.Count.Should().Be(allData.Count, $"iteration {iter}: {allData.Count} items in final state");
results.Data.Keys.OrderBy(k => k).Should().Equal(allData, $"iteration {iter}: all keys present in order");
results.Error.Should().BeNull($"iteration {iter}: no errors");
results.IsCompleted.Should().BeFalse($"iteration {iter}: not completed");
}
}
}
}
Loading
Loading