klod68/gen-resilience-policy

name: gen-resilience-policy description: "Use when adding retry or resilience handling to an HTTP client or external dependency call. Generates a Polly v8 policy with exponential backoff, transient error detection, jitter, and an observability callback. Also invoke when the user mentions: add retry, resilience policy, Polly policy, transient fault handling. Domain: Code Generation, Resilience. Level: Intermediate."

Generate a resilience policy for the following operation.

Context

• Operation type: ${1|Database CRUD,Database batch,HTTP API call,Message queue,File I/O,Background job,Health check|}
• Provider: ${2:e.g. SQL Server, PostgreSQL, HttpClient, RabbitMQ}
• Idempotent (safe to retry without side effects): ${3|Yes,No|}
• Caller timeout: ${4:e.g. 30 seconds}
• Environment: ${5|User-facing API,Background job,Batch process,Health check|}

Required output — generate ALL of the following:

1. Policy definition

Place in a ResiliencePolicies static class (create if not present):

/// <summary>
/// [Description of what operation this policy protects.]
/// Total worst-case retry window: [calculated — sum of all max delays].
/// </summary>
public static readonly IResiliencePolicy [PolicyName] = ResiliencePolicyFactory.Create(
    maxRetryCount:    [N],
    initialDelay:     TimeSpan.FromMilliseconds([N]),
    maxDelay:         TimeSpan.FromSeconds([N]),
    backoffMultiplier: 2.0,
    useJitter:        true,
    onRetry: (ex, attempt, delay) =>
        Log.Warning(
            "Retry {Attempt} for {Operation} after {Delay}ms: {Error}",
            attempt, "[OperationName]", delay.TotalMilliseconds, ex.Message));

2. Retry timeline table

| Attempt | Base delay | With jitter (±25%) | Cumulative max | |---|---|---|---| | 1 | | | | | 2 | | | | | N | | | |

3. Sizing rationale

Document why these values were chosen given the operation type, idempotency, caller timeout, and environment. Include the worst-case total window calculation.

4. Validation checklist

• [ ] Total retry window ≤ caller timeout (${4})
• [ ] Jitter is enabled (thundering herd prevention)
• [ ] Non-idempotent operation has ≤ 1 retry
• [ ] OperationCanceledException is never retried
• [ ] Permanent errors (auth, constraint, syntax) are excluded from transient detection
• [ ] onRetry callback logs attempt number, delay, and exception message

Sizing reference

| Environment | Max retries | Initial delay | Total window | |---|---|---|---| | User-facing API | 2–3 | 50–100 ms | < 2 s | | Background job | 3–5 | 500 ms–1 s | < 1 min | | Batch process | 5–10 | 1–5 s | < 5 min | | Non-idempotent | 0–1 | any | minimal | | Health check | 0 | — | fail fast |

Constraints

• Follow all rules in the workspace root instruction file
• Policy instances are public static readonly — stateless, created once, thread-safe
• Never retry OperationCanceledException — always propagate cancellation
• onRetry uses structured logging templates — no string interpolation
• Never log sensitive data in the retry callback