From b87cab2d9a8bb06aa0c8e5795a11d4aa2789c9ec Mon Sep 17 00:00:00 2001
From: Mitch Denny <midenn@microsoft.com>
Date: Thu, 17 Oct 2024 05:50:32 +1100
Subject: [PATCH] Move SlowDownMonitoringAsync. (#6309)

* Move SlowDownMonitoringAsync.

* Add tests.
---
 .../Health/ResourceHealthCheckService.cs      |  9 +-
 .../Health/ResourceHealthCheckServiceTests.cs | 93 +++++++++++++++++++
 2 files changed, 95 insertions(+), 7 deletions(-)

diff --git a/src/Aspire.Hosting/Health/ResourceHealthCheckService.cs b/src/Aspire.Hosting/Health/ResourceHealthCheckService.cs
index ab72e756de..297fbe4dc3 100644
--- a/src/Aspire.Hosting/Health/ResourceHealthCheckService.cs
+++ b/src/Aspire.Hosting/Health/ResourceHealthCheckService.cs
@@ -83,10 +83,9 @@ await eventing.PublishAsync(
                         cancellationToken).ConfigureAwait(false);
                 }
 
-                if (_latestEvents[resource.Name].Snapshot.HealthStatus == report.Status)
+                if (_latestEvents[resource.Name] is { } latestEvent && latestEvent.Snapshot.HealthStatus == report.Status)
                 {
-                    // If the last health status is the same as this health status then we don't need
-                    // to publish anything as it just creates noise.
+                    await SlowDownMonitoringAsync(latestEvent, cancellationToken).ConfigureAwait(false);
                     continue;
                 }
 
@@ -103,10 +102,6 @@ await resourceNotificationService.PublishUpdateAsync(resource, s =>
                         HealthReports = healthReports
                     };
                 }).ConfigureAwait(false);
-
-                var lastEvent = _latestEvents[resource.Name];
-                await SlowDownMonitoringAsync(lastEvent, cancellationToken).ConfigureAwait(false);
-
             }
             catch (Exception ex) when (!cancellationToken.IsCancellationRequested)
             {
diff --git a/tests/Aspire.Hosting.Tests/Health/ResourceHealthCheckServiceTests.cs b/tests/Aspire.Hosting.Tests/Health/ResourceHealthCheckServiceTests.cs
index 53a18c44c6..3c87f14f68 100644
--- a/tests/Aspire.Hosting.Tests/Health/ResourceHealthCheckServiceTests.cs
+++ b/tests/Aspire.Hosting.Tests/Health/ResourceHealthCheckServiceTests.cs
@@ -1,6 +1,7 @@
 // Licensed to the .NET Foundation under one or more agreements.
 // The .NET Foundation licenses this file to you under the MIT license.
 
+using System.Diagnostics;
 using Aspire.Hosting.Utils;
 using Microsoft.Extensions.DependencyInjection;
 using Microsoft.Extensions.Diagnostics.HealthChecks;
@@ -11,6 +12,98 @@ namespace Aspire.Hosting.Tests.Health;
 
 public class ResourceHealthCheckServiceTests(ITestOutputHelper testOutputHelper)
 {
+    [Fact]
+    public async Task HealthCheckIntervalSlowsAfterSteadyHealthyState()
+    {
+        using var builder = TestDistributedApplicationBuilder.Create(testOutputHelper);
+
+        AutoResetEvent? are = null;
+
+        builder.Services.AddHealthChecks().AddCheck("resource_check", () =>
+        {
+            are?.Set();
+
+            return HealthCheckResult.Healthy();
+        });
+
+        var resource = builder.AddResource(new ParentResource("resource"))
+                              .WithHealthCheck("resource_check");
+
+        using var app = builder.Build();
+        var rns = app.Services.GetRequiredService<ResourceNotificationService>();
+
+        var abortTokenSource = new CancellationTokenSource(TimeSpan.FromSeconds(120));
+
+        await app.StartAsync(abortTokenSource.Token);
+
+        await rns.PublishUpdateAsync(resource.Resource, s => s with
+        {
+            State = KnownResourceStates.Running
+        });
+        await rns.WaitForResourceHealthyAsync(resource.Resource.Name, abortTokenSource.Token);
+
+        are = new AutoResetEvent(false);
+
+        // Allow one event to through since it could be half way through.
+        are.WaitOne();
+
+        var stopwatch = Stopwatch.StartNew();
+        are.WaitOne();
+        stopwatch.Stop();
+
+        // Delay is 30 seconds but we allow for a (ridiculous) 10 second margin of error.
+        Assert.True(stopwatch.ElapsedMilliseconds > 20000);
+
+        await app.StopAsync(abortTokenSource.Token);
+    }
+
+    [Fact]
+    public async Task HealthCheckIntervalDoesNotSlowBeforeSteadyHealthyState()
+    {
+        using var builder = TestDistributedApplicationBuilder.Create(testOutputHelper);
+
+        AutoResetEvent? are = null;
+
+        builder.Services.AddHealthChecks().AddCheck("resource_check", () =>
+        {
+            are?.Set();
+
+            return HealthCheckResult.Unhealthy();
+        });
+
+        var resource = builder.AddResource(new ParentResource("resource"))
+                              .WithHealthCheck("resource_check");
+
+        using var app = builder.Build();
+        var rns = app.Services.GetRequiredService<ResourceNotificationService>();
+
+        var abortTokenSource = new CancellationTokenSource(TimeSpan.FromSeconds(120));
+
+        await app.StartAsync(abortTokenSource.Token);
+
+        await rns.PublishUpdateAsync(resource.Resource, s => s with
+        {
+            State = KnownResourceStates.Running
+        });
+        await rns.WaitForResourceAsync(resource.Resource.Name, KnownResourceStates.Running, abortTokenSource.Token);
+
+        are = new AutoResetEvent(false);
+
+        // Allow one event to through since it could be half way through.
+        are.WaitOne();
+
+        var stopwatch = Stopwatch.StartNew();
+        are.WaitOne();
+        stopwatch.Stop();
+
+        // When not in a healthy state the delay should be ~3 seconds but
+        // we'll check for 10 seconds to make sure we haven't got down
+        // the 30 second slow path.
+        Assert.True(stopwatch.ElapsedMilliseconds < 10000);
+
+        await app.StopAsync(abortTokenSource.Token);
+    }
+
     [Fact]
     public async Task ResourcesWithoutHealthCheckAnnotationsGetReadyEventFired()
     {