Logging in DI

A policy registered in a DI container writes log records through ILogger with no calls or lambdas from you: the library understands each event and emits the matching record.

services.AddLogging();
services.AddResilience(name: "payments", policy: Resilience.Http);

// Nothing else to call. The policy logs under "NResilience.payments", which is the category
// an appsettings.json filter matches.

A policy you build yourself is not logged until you say so. See Logging.

Filter per policy

Each policy logs under its own category: NResilience for a policy with no name, and NResilience.<name> otherwise. For a registered policy the name is the registration name; for an HttpClient it is the client name.

{
  "Logging": {
    "LogLevel": {
      "Default": "Information",
      "NResilience": "Warning",
      "NResilience.payments": "Debug"
    }
  }
}

The category is fixed when the listener attaches. Because an HTTP client derives a policy per host, the category does not include the host - otherwise a client talking to fifty hosts would create fifty categories. The host-scoped name still appears in the Policy field of every record for structured output queries.

Use ResilienceLogging.CategoryFor(name) to get the category in code.

Set the profile per policy

Set the profile per policy from a section, or for the whole process with AddResilienceLogging.

{
  "Resilience": {
    "payments": {
      "Preset": "Http",
      "Logging": "Verbose"
    },
    "reports": {
      "Preset": "Http",
      "Logging": "Off"
    }
  }
}

An explicit WithLogging call in a configure callback overrides the automatic listener added by a container.

AddResilienceLogging also sets the process-wide knobs a section does not carry, including sampling: services.AddResilienceLogging(o => o.Sampling = LogSampling.OneIn(20)) keeps one twentieth of the steady state and all of an incident. A section that overrides Logging for one policy keeps those knobs. See Sample the steady state.

Provenance

Binding a configuration section is silently partial. Event 1020 reports the effective policy once per resolution at Debug; a reload produces a new entry, showing exactly what changed.

Event 1021 is a Trace companion that dumps the classifier's state, so you can see what the policy will retry without reading the source. It is guarded to cost nothing when Trace is disabled.

Next steps

• Logging: The profiles, the levels, and how to instrument a policy you built yourself.
• Event IDs: The full table, which is the contract an alert is built on.
• Options reference: Logging and the rest of the bindable shape.