Lack of cgroups v2 Support in Microsoft.Extensions.Diagnostics.ResourceMonitoring

[mobratil]

Description

I am reporting an issue with Microsoft.Extensions.Diagnostics.ResourceMonitoring related to its incompatibility with cgroups v2. This issue is particularly evident when running applications in environments that default to cgroups v2, such as Debian GNU/Linux 12 (Bookworm).

Issue Details

The library seems to be designed to work with cgroups v1, as it attempts to access /sys/fs/cgroup/memory/memory.limit_in_bytes, a path specific to the cgroups v1 hierarchy. In environments using cgroups v2, this path does not exist, leading to System.IO.DirectoryNotFoundException.

Expected Behavior

For environments using cgroups v2, the library should be able to access the appropriate cgroups v2 paths and files to monitor resources.

Actual Behavior

When running in a cgroups v2 environment, the library fails to find the required paths, resulting in an unhandled exception (DirectoryNotFoundException).

Steps to Reproduce

1. Run an application using Microsoft.Extensions.Diagnostics.ResourceMonitoring in a Docker container on a system using cgroups v2 (e.g., Debian GNU/Linux 12).
2. Observe the DirectoryNotFoundException when the library attempts to access cgroups v1 specific paths.

Suggested Solution

It would be beneficial for Microsoft.Extensions.Diagnostics.ResourceMonitoring to include support for cgroups v2, given its increasing adoption in newer Linux distributions.

Additional Context

This issue is crucial for applications that are intended to run in modern Linux environments where cgroups v2 is the default or only available option.

[tarasevichvlad]

Good day. Is there any time frame for solving this problem? In our project this has a very high priority. And it is important for us to understand how quickly it will appear here or whether this package will simply be abandoned.

[madhub]

Most of the Linux distro are moved cgroup v2 & it is also available in the stable version of kubernetes . It would be great if you support new version otherwise library can't be used in these environments

[mobratil]

The issue was resolved by this PR: #5068

[tarasevichvlad]

Hi, @mobratil
I faced error after updating.

"body": "Unable to gather utilization statistics.",
"severity": "Error",
"attributes": {
"exception.message": "Expected proc/stat to start with 'cpu ' but it was 'estore_file 0\nworkingset_nodereclaim 0\npgfault 623476\npgmajfault 73\npgrefill 0\npgscan 0\npgsteal 0\npgactivate 2924\npgdeactivate 0\npglazyfree 0\npglazyfreed 0\nthp_fault_alloc 0\nthp_collapse_alloc 0\ncpu 1843193685 225438 690691436 5724483778 16007065 0 61993087 0 0 0'.",
"exception.stacktrace": "System.InvalidOperationException: Expected proc/stat to start with 'cpu ' but it was 'estore_file 0\nworkingset_nodereclaim 0\npgfault 623476\npgmajfault 73\npgrefill 0\npgscan 0\npgsteal 0\npgactivate 2924\npgdeactivate 0\npglazyfree 0\npglazyfreed 0\nthp_fault_alloc 0\nthp_collapse_alloc 0\ncpu 1843193685 225438 690691436 5724483778 16007065 0 61993087 0 0 0'.\n at Microsoft.Shared.Diagnostics.Throw.InvalidOperationException(String message)\n at Microsoft.Extensions.Diagnostics.ResourceMonitoring.Linux.LinuxUtilizationParserCgroupV2.GetHostCpuUsageInNanoseconds()\n at Microsoft.Extensions.Diagnostics.ResourceMonitoring.Linux.LinuxUtilizationProvider.GetSnapshot()\n at Microsoft.Extensions.Diagnostics.ResourceMonitoring.ResourceMonitorService.ExecuteAsync(CancellationToken cancellationToken)",
"exception.type": "InvalidOperationException",
"logrecord.event.id": 1,
"logrecord.event.name": "HandledGatherStatisticsException",
"{OriginalFormat}": "Unable to gather utilization statistics."
},
"resources": {
"deployment.environment": "Testing",
"telemetry.sdk.language": "dotnet",
"telemetry.sdk.name": "opentelemetry",
"telemetry.sdk.version": "1.8.1"
},
"instrumentation_scope": {
"name": "Microsoft.Extensions.Diagnostics.ResourceMonitoring.ResourceMonitorService"
}

[mobratil]

@tarasevichvlad I'm afraid that this problem is caused by this issue: #5114