homelab-frontend/internal/resilience/retry.go

package resilience

import (
	"context"
	"math/rand"
	"net/http"
	"time"
)

// RetryConfig holds retry settings.
type RetryConfig struct {
	// MaxAttempts is the maximum number of attempts (includes initial).
	MaxAttempts int
	// InitialBackoff is the initial backoff duration.
	InitialBackoff time.Duration
	// MaxBackoff is the maximum backoff duration.
	MaxBackoff time.Duration
	// BackoffMultiplier is the exponential backoff multiplier.
	BackoffMultiplier float64
}

// DefaultRetryConfig provides sensible defaults.
func DefaultRetryConfig() *RetryConfig {
	return &RetryConfig{
		MaxAttempts:       3,
		InitialBackoff:    100 * time.Millisecond,
		MaxBackoff:        2 * time.Second,
		BackoffMultiplier: 2.0,
	}
}

// RetryFunc executes a function with blind retry on 5xx.
// Returns the response and any error from the function itself (not retry logic).
type RetryFunc func(ctx context.Context, attempt int) (*http.Response, error)

// DoRetry executes the function with exponential backoff on 5xx responses.
// Returns the final response (could be 5xx if all retries exhausted) and any error.
func DoRetry(ctx context.Context, cfg *RetryConfig, fn RetryFunc) (*http.Response, error) {
	if cfg == nil {
		cfg = DefaultRetryConfig()
	}

	var lastResp *http.Response
	var lastErr error

	for attempt := 0; attempt < cfg.MaxAttempts; attempt++ {
		// Check context before attempting
		select {
		case <-ctx.Done():
			if lastResp != nil {
				lastResp.Body.Close()
			}
			return nil, ctx.Err()
		default:
		}

		resp, err := fn(ctx, attempt)
		if err != nil {
			lastErr = err
			// Don't retry on network errors in the retry loop itself
			// Let caller decide if those should be retried
			return nil, err
		}

		// Success (not 5xx)
		if resp.StatusCode < 500 {
			return resp, nil
		}

		// 5xx — close and retry
		if lastResp != nil {
			lastResp.Body.Close()
		}
		lastResp = resp

		// If this was the last attempt, return the 5xx response
		if attempt == cfg.MaxAttempts-1 {
			return resp, nil
		}

		// Calculate backoff with jitter
		backoff := calculateBackoff(attempt, cfg)
		select {
		case <-ctx.Done():
			resp.Body.Close()
			return nil, ctx.Err()
		case <-time.After(backoff):
			// Continue to next attempt
		}
	}

	return lastResp, lastErr
}

// calculateBackoff computes exponential backoff with jitter.
func calculateBackoff(attempt int, cfg *RetryConfig) time.Duration {
	// Exponential: initial * (multiplier ^ attempt)
	backoff := time.Duration(float64(cfg.InitialBackoff) * (pow(cfg.BackoffMultiplier, float64(attempt))))

	// Cap at max
	if backoff > cfg.MaxBackoff {
		backoff = cfg.MaxBackoff
	}

	// Add jitter: ±20%
	jitterRange := backoff / 5
	if jitterRange <= 0 {
		return backoff
	}

	jitter := time.Duration(rand.Int63n(int64(2 * jitterRange)) - int64(jitterRange))

	return backoff + jitter
}

func pow(base, exp float64) float64 {
	result := 1.0
	for i := 0; i < int(exp); i++ {
		result *= base
	}
	return result
}

// RetryPolicy determines whether to retry based on response and config.
type RetryPolicy struct {
	Retryable bool // Whether this adapter allows retries
}

// ShouldRetry determines if a response should be retried.
func (p *RetryPolicy) ShouldRetry(resp *http.Response) bool {
	if !p.Retryable {
		return false
	}
	return resp != nil && resp.StatusCode >= 500
}