Resilience Policy Reference

ResiliencePolicy centralizes retry/backoff, timeouts, and per-host concurrency caps that the HTTP executor honors. Use built-ins or derive a custom policy with .With(...) to avoid argument-order bugs.

Built-in profiles

• Default — conservative baseline suitable for most control-plane calls.
• Search — lower per-request deadline and retries for fast paging.
• Lookup — slightly longer retries for details endpoints.
• Streaming — long per-request timeout and lower concurrency for ranged downloads.
• Authentication — careful retries; short max backoff.
• Metadata — similar to Lookup but tuned for lightweight metadata calls.

Each profile has:

• Name — profile tag (propagated to gates and telemetry).
• MaxRetries — attempts including the first try.
• RetryBudget — total wall-clock budget across retries.
• PerRequestTimeout — optional hard timeout per HTTP request.
• MaxConcurrencyPerHost — gate for this profile+host pair.
• MaxTotalConcurrencyPerHost — optional aggregate cap for all profiles to the same host.
• InitialBackoff/MaxBackoff/JitterMin/JitterMax — backoff and jitter envelope.

Choosing a profile

• Search/list endpoints → ResiliencePolicy.Search.
• Item details/lookup → ResiliencePolicy.Lookup.
• Auth/token exchange → ResiliencePolicy.Authentication.
• Long downloads/streaming → ResiliencePolicy.Streaming.

Customizing

var tight = ResiliencePolicy.Search.With(
    maxRetries: 3,
    retryBudget: TimeSpan.FromSeconds(12),
    perRequestTimeout: TimeSpan.FromSeconds(8),
    maxConcurrencyPerHost: 4);

using var resp = await http.ExecuteWithResilienceAsync(req, tight, token);

Notes

• Retry-After absolute dates are honored without extra jitter (still clamped by the budget).
• Relative URIs are resolved against HttpClient.BaseAddress inside the executor.
• Profile name is stamped as an option and propagates to host gates and spans.