Add network usage Health Checks

[evgenyfedorov2]

In addition to the existing health checks on CPU and Memory usage provided by the Microsoft.Extensions.Diagnostics.HealthChecks.ResourceUtilization package, please add ones for Network usage.

[amadeuszl]

I'm trying to assess the measurements we could use to make it work. Did you had any counts/measurements in mind when creating the ticket?

1. Usage of bandwidth could be used, but we have to add that, because it's not measured now. So service owner can verify what would be the appropriate thresholds for the service, unless there's OS info on percentage of the bandwidth usage.
2. Tracking the number of open connections is possible, but setting precise thresholds is challenging. Maybe if it's 0?
3. Latency measurements could be nice, but we don't measure it now as well, and I think this should be handled more on application layer, rather than resources.

[evgenyfedorov2]

Short-term

1,2. Measurements used here and here can be used to begin with. Similar to CPU and Memory usage health checks where we ask what CPU and memory usage value is worrying (and the answer is "80% and greater") and create Health Checks based on that, we should ask the same question to each network usage counter (or even create new metrics of the ObservableGauge instrument and ask the same question).

There are network metrics in .NET which we can take advantage of too.

3. Latency is available here and here. It is a shame there is no public documentation (or I just can't find it?). I am not sure if it can be leveraged for the purposes of Health Checks, though. Feel free to contact @ananthsr internally as the original author of the library.

Long-term

It needs more research. For instance, what are typical alert thresholds in network usage area for (micro)services at Microsoft? We could just use that data and signal it via ASP.NET Core HealthChecks too.

[amadeuszl]

Below is proposed summary for that feature that can be discussed further by the community.

Key Features to Implement

• ResourceUtilizationHealthCheckExtensions:

  • AddNetworkingHealthCheck() and AddProbingHealthChecks() methods, with overloads similar to existing resource utilization health checks.
  • No default configuration—service owners must explicitly configure all options due to varying service/hardware needs.

• Probing for Outgoing Requests:

  • Each endpoint can have its own probe configuration (URL, window, min time between calls).
  • Health is determined by the percentage of successful calls over a configurable time window (e.g., last 5 minutes).
  • Configurable thresholds for "Healthy", "Degraded", and "Unhealthy" states (e.g., Healthy >99.99%, Degraded <99.99%, Unhealthy <99.9%).

• TCP Connections Health for Incoming Requests:

  • Monitors the number of established TCP connections to detect issues (too many = resource exhaustion, too few = possible network failure).
  • Alerts based on SYN_RECEIVED state to catch issues like SYN flooding or slow handshakes.
  • Configurable min/max thresholds for connection states.

• Platform Awareness:

  • Differentiates between Linux/Windows and host/container environments to ensure accurate metric collection.

• Performance Considerations:

  • Uses a circular buffer to store recent metric snapshots, minimizing CPU/memory overhead.
  • Ensures probes are not run too frequently (configurable minimum interval).

Design Tradeoffs

• No default values—everything must be configured by the service owner.