Skip to content

Commit

Permalink
ResourceUtilizationHealthCheck - Report both CPU and Memory issues (#…
Browse files Browse the repository at this point in the history 
…5407)

* Report both CPU and Memory issues

* Use dedicated strings for each case and fix capitalization

* Fix test
  • Loading branch information
@willibrandon
willibrandon authored Sep 9, 2024
1 parent 2739017 commit 39f5d27
Show file tree
Hide file tree
Showing 2 changed files with 105 additions and 56 deletions.
56 changes: 43 additions & 13 deletions ...Extensions.Diagnostics.HealthChecks.ResourceUtilization/ResourceUtilizationHealthCheck.cs
View file
Original file line number Diff line number Diff line change
@@ -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.Collections.Generic;
using System.Threading;
using System.Threading.Tasks;
using Microsoft.Extensions.Diagnostics.ResourceMonitoring;
Expand All @@ -14,7 +15,6 @@ namespace Microsoft.Extensions.Diagnostics.HealthChecks;
/// </summary>
internal sealed class ResourceUtilizationHealthCheck : IHealthCheck
{
private static readonly Task<HealthCheckResult> _healthy = Task.FromResult(HealthCheckResult.Healthy());
private readonly ResourceUtilizationHealthCheckOptions _options;
private readonly IResourceMonitor _dataTracker;

Expand All @@ -39,26 +39,56 @@ public ResourceUtilizationHealthCheck(IOptions<ResourceUtilizationHealthCheckOpt
public Task<HealthCheckResult> CheckHealthAsync(HealthCheckContext context, CancellationToken cancellationToken = default)
{
var utilization = _dataTracker.GetUtilization(_options.SamplingWindow);
if (utilization.CpuUsedPercentage > _options.CpuThresholds.UnhealthyUtilizationPercentage)
IReadOnlyDictionary<string, object> data = new Dictionary<string, object>
{
return Task.FromResult(HealthCheckResult.Unhealthy("CPU usage is above the limit"));
}
{ nameof(utilization.CpuUsedPercentage), utilization.CpuUsedPercentage },
{ nameof(utilization.MemoryUsedPercentage), utilization.MemoryUsedPercentage },
};

if (utilization.MemoryUsedPercentage > _options.MemoryThresholds.UnhealthyUtilizationPercentage)
{
return Task.FromResult(HealthCheckResult.Unhealthy("Memory usage is above the limit"));
}
bool cpuUnhealthy = utilization.CpuUsedPercentage > _options.CpuThresholds.UnhealthyUtilizationPercentage;
bool memoryUnhealthy = utilization.MemoryUsedPercentage > _options.MemoryThresholds.UnhealthyUtilizationPercentage;

if (utilization.CpuUsedPercentage > _options.CpuThresholds.DegradedUtilizationPercentage)
if (cpuUnhealthy || memoryUnhealthy)
{
return Task.FromResult(HealthCheckResult.Degraded("CPU usage is close to the limit"));
string message = string.Empty;
if (cpuUnhealthy && memoryUnhealthy)
{
message = "CPU and memory usage is above the limit";
}
else if (cpuUnhealthy)
{
message = "CPU usage is above the limit";
}
else
{
message = "Memory usage is above the limit";
}

return Task.FromResult(HealthCheckResult.Unhealthy(message, default, data));
}

if (utilization.MemoryUsedPercentage > _options.MemoryThresholds.DegradedUtilizationPercentage)
bool cpuDegraded = utilization.CpuUsedPercentage > _options.CpuThresholds.DegradedUtilizationPercentage;
bool memoryDegraded = utilization.MemoryUsedPercentage > _options.MemoryThresholds.DegradedUtilizationPercentage;

if (cpuDegraded || memoryDegraded)
{
return Task.FromResult(HealthCheckResult.Degraded("Memory usage is close to the limit"));
string message = string.Empty;
if (cpuDegraded && memoryDegraded)
{
message = "CPU and memory usage is close to the limit";
}
else if (cpuDegraded)
{
message = "CPU usage is close to the limit";
}
else
{
message = "Memory usage is close to the limit";
}

return Task.FromResult(HealthCheckResult.Degraded(message, default, data));
}

return _healthy;
return Task.FromResult(HealthCheckResult.Healthy(default, data));
}
}
105 changes: 62 additions & 43 deletions ...Extensions.Diagnostics.HealthChecks.ResourceUtilization.Tests/ResourceHealthCheckTests.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@ public class ResourceHealthCheckTests
0UL,
1000UL,
new ResourceUsageThresholds(),
new ResourceUsageThresholds(),
"",
},
new object[]
Expand All @@ -32,6 +33,7 @@ public class ResourceHealthCheckTests
0UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.2 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.2 },
""
},
new object[]
Expand All @@ -41,6 +43,7 @@ public class ResourceHealthCheckTests
2UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.2 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.2 },
""
},
new object[]
Expand All @@ -50,7 +53,8 @@ public class ResourceHealthCheckTests
3UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
" usage is close to the limit"
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
"CPU and memory usage is close to the limit"
},
new object[]
{
Expand All @@ -59,7 +63,8 @@ public class ResourceHealthCheckTests
5UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
" usage is above the limit"
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
"CPU and memory usage is above the limit"
},
new object[]
{
Expand All @@ -68,7 +73,8 @@ public class ResourceHealthCheckTests
5UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.4, UnhealthyUtilizationPercentage = 0.2 },
" usage is above the limit"
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.4, UnhealthyUtilizationPercentage = 0.2 },
"CPU and memory usage is above the limit"
},
new object[]
{
Expand All @@ -77,7 +83,8 @@ public class ResourceHealthCheckTests
3UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2 },
" usage is close to the limit"
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2 },
"CPU and memory usage is close to the limit"
},
new object[]
{
Expand All @@ -86,67 +93,79 @@ public class ResourceHealthCheckTests
5UL,
1000UL,
new ResourceUsageThresholds { UnhealthyUtilizationPercentage = 0.4 },
" usage is above the limit"
new ResourceUsageThresholds { UnhealthyUtilizationPercentage = 0.4 },
"CPU and memory usage is above the limit"
},
new object[]
{
HealthStatus.Degraded,
0.3,
3UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.9, UnhealthyUtilizationPercentage = 0.9 },
"CPU usage is close to the limit"
},
new object[]
{
HealthStatus.Degraded,
0.1,
3UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.9, UnhealthyUtilizationPercentage = 0.9 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
"Memory usage is close to the limit"
},
new object[]
{
HealthStatus.Unhealthy,
0.5,
5UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.9, UnhealthyUtilizationPercentage = 0.9 },
"CPU usage is above the limit"
},
new object[]
{
HealthStatus.Unhealthy,
0.1,
5UL,
1000UL,
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.9, UnhealthyUtilizationPercentage = 0.9 },
new ResourceUsageThresholds { DegradedUtilizationPercentage = 0.2, UnhealthyUtilizationPercentage = 0.4 },
"Memory usage is above the limit"
},
};

[Theory]
[MemberData(nameof(Data))]
#pragma warning disable xUnit1026 // Theory methods should use all of their parameters
public async Task TestCpuChecks(HealthStatus expected, double utilization, ulong _, ulong totalMemory, ResourceUsageThresholds thresholds, string expectedDescription)
#pragma warning restore xUnit1026 // Theory methods should use all of their parameters
{
var systemResources = new SystemResources(1.0, 1.0, totalMemory, totalMemory);
var dataTracker = new Mock<IResourceMonitor>();
var samplingWindow = TimeSpan.FromSeconds(1);
dataTracker
.Setup(tracker => tracker.GetUtilization(samplingWindow))
.Returns(new ResourceUtilization(cpuUsedPercentage: utilization, memoryUsedInBytes: 0, systemResources));

var checkContext = new HealthCheckContext();
var cpuCheckOptions = new ResourceUtilizationHealthCheckOptions
{
CpuThresholds = thresholds,
SamplingWindow = samplingWindow
};

var options = Microsoft.Extensions.Options.Options.Create(cpuCheckOptions);
var healthCheck = new ResourceUtilizationHealthCheck(options, dataTracker.Object);
var healthCheckResult = await healthCheck.CheckHealthAsync(checkContext);
Assert.Equal(expected, healthCheckResult.Status);
if (healthCheckResult.Status != HealthStatus.Healthy)
{
Assert.Equal("CPU" + expectedDescription, healthCheckResult.Description);
}
}

[Theory]
[MemberData(nameof(Data))]
#pragma warning disable xUnit1026 // Theory methods should use all of their parameters
public async Task TestMemoryChecks(HealthStatus expected, double _, ulong memoryUsed, ulong totalMemory, ResourceUsageThresholds thresholds, string expectedDescription)
#pragma warning restore xUnit1026 // Theory methods should use all of their parameters
public async Task TestCpuAndMemoryChecks(HealthStatus expected, double utilization, ulong memoryUsed, ulong totalMemory,
ResourceUsageThresholds cpuThresholds, ResourceUsageThresholds memoryThresholds, string expectedDescription)
{
var systemResources = new SystemResources(1.0, 1.0, totalMemory, totalMemory);
var dataTracker = new Mock<IResourceMonitor>();
var samplingWindow = TimeSpan.FromSeconds(1);
dataTracker
.Setup(tracker => tracker.GetUtilization(samplingWindow))
.Returns(new ResourceUtilization(cpuUsedPercentage: 0, memoryUsedInBytes: memoryUsed, systemResources));
.Returns(new ResourceUtilization(cpuUsedPercentage: utilization, memoryUsedInBytes: memoryUsed, systemResources));

var checkContext = new HealthCheckContext();
var memCheckOptions = new ResourceUtilizationHealthCheckOptions
var checkOptions = new ResourceUtilizationHealthCheckOptions
{
MemoryThresholds = thresholds,
CpuThresholds = cpuThresholds,
MemoryThresholds = memoryThresholds,
SamplingWindow = samplingWindow
};

var options = Microsoft.Extensions.Options.Options.Create(memCheckOptions);
var options = Microsoft.Extensions.Options.Options.Create(checkOptions);
var healthCheck = new ResourceUtilizationHealthCheck(options, dataTracker.Object);
var healthCheckResult = await healthCheck.CheckHealthAsync(checkContext);
Assert.Equal(expected, healthCheckResult.Status);
Assert.NotEmpty(healthCheckResult.Data);
if (healthCheckResult.Status != HealthStatus.Healthy)
{
Assert.Equal("Memory" + expectedDescription, healthCheckResult.Description);
Assert.Equal(expectedDescription, healthCheckResult.Description);
}
}

Expand Down

0 comments on commit 39f5d27

Please sign in to comment.