|
| 1 | +// Package retry provides a provider-agnostic retry engine (exponential |
| 2 | +// backoff, context cancellation, Retry-After honoring) plus the two typed |
| 3 | +// errors every connector needs to drive it: RateLimitError and AuthError. |
| 4 | +// |
| 5 | +// Error classification is deliberately NOT centralized here. Each connector |
| 6 | +// knows how its provider signals "rate limited" or "auth rejected" (status |
| 7 | +// codes, headers, GraphQL error types, ...) and maps those onto typed errors |
| 8 | +// or its own RetryOptions.Classify/ExtractRetryAfter functions. This package |
| 9 | +// only understands the typed RateLimitError/AuthError by default; it does not |
| 10 | +// pattern-match provider error strings. |
| 11 | +// |
| 12 | +// This is a leaf package with no dependencies beyond the Go stdlib, so any |
| 13 | +// connector can import it without creating a cycle. |
| 14 | +package retry |
| 15 | + |
| 16 | +import ( |
| 17 | + "context" |
| 18 | + "errors" |
| 19 | + "fmt" |
| 20 | + "time" |
| 21 | +) |
| 22 | + |
| 23 | +// RateLimitError signals that an operation was rejected because of API rate |
| 24 | +// limiting. RetryAfter carries the provider's requested backoff when known; |
| 25 | +// zero means the caller should fall back to exponential backoff. |
| 26 | +type RateLimitError struct { |
| 27 | + RetryAfter time.Duration |
| 28 | + Message string |
| 29 | +} |
| 30 | + |
| 31 | +func (e *RateLimitError) Error() string { |
| 32 | + return fmt.Sprintf("rate limited: %s", e.Message) |
| 33 | +} |
| 34 | + |
| 35 | +// AuthError signals that the credentials themselves were rejected (e.g. HTTP |
| 36 | +// 401). It is never retryable — retrying with the same token cannot help, so |
| 37 | +// callers should park the operation immediately instead of burning attempts. |
| 38 | +type AuthError struct { |
| 39 | + Message string |
| 40 | +} |
| 41 | + |
| 42 | +func (e *AuthError) Error() string { |
| 43 | + return fmt.Sprintf("auth error: %s", e.Message) |
| 44 | +} |
| 45 | + |
| 46 | +// RetryOptions configures WithRetry / WithRetryVoid. |
| 47 | +// |
| 48 | +// Classify and ExtractRetryAfter are per-connector hooks: each integration |
| 49 | +// maps its provider's error responses onto typed errors (or its own richer |
| 50 | +// error types) and supplies matching functions here. Both default to typed- |
| 51 | +// error-only behavior when left nil — no status-code string matching. |
| 52 | +type RetryOptions struct { |
| 53 | + MaxRetries int // Maximum number of retries (default: 3) |
| 54 | + BaseDelay time.Duration // Initial delay between retries (default: 1s) |
| 55 | + MaxDelay time.Duration // Maximum delay between retries (default: 30s) |
| 56 | + |
| 57 | + // Classify reports whether err is transient and worth retrying. Defaults |
| 58 | + // to DefaultClassify when nil. |
| 59 | + Classify func(err error) bool |
| 60 | + |
| 61 | + // ExtractRetryAfter returns the provider-requested backoff for err, or 0 |
| 62 | + // if none applies. Defaults to DefaultExtractRetryAfter when nil. |
| 63 | + ExtractRetryAfter func(err error) time.Duration |
| 64 | +} |
| 65 | + |
| 66 | +// DefaultRetryOptions returns sensible defaults for retry behavior. |
| 67 | +func DefaultRetryOptions() RetryOptions { |
| 68 | + return RetryOptions{ |
| 69 | + MaxRetries: 3, |
| 70 | + BaseDelay: 1 * time.Second, |
| 71 | + MaxDelay: 30 * time.Second, |
| 72 | + } |
| 73 | +} |
| 74 | + |
| 75 | +// DefaultClassify retries *RateLimitError and refuses *AuthError; any other |
| 76 | +// error is treated as non-retryable. Connectors with richer transient-error |
| 77 | +// detection (5xx, network failures, provider-specific GraphQL errors, ...) |
| 78 | +// should supply their own Classify via RetryOptions. |
| 79 | +func DefaultClassify(err error) bool { |
| 80 | + if err == nil { |
| 81 | + return false |
| 82 | + } |
| 83 | + var rlErr *RateLimitError |
| 84 | + if errors.As(err, &rlErr) { |
| 85 | + return true |
| 86 | + } |
| 87 | + var authErr *AuthError |
| 88 | + if errors.As(err, &authErr) { |
| 89 | + return false |
| 90 | + } |
| 91 | + return false |
| 92 | +} |
| 93 | + |
| 94 | +// DefaultExtractRetryAfter returns RateLimitError.RetryAfter when err (or |
| 95 | +// something it wraps) is a *RateLimitError, else 0. |
| 96 | +func DefaultExtractRetryAfter(err error) time.Duration { |
| 97 | + var rlErr *RateLimitError |
| 98 | + if errors.As(err, &rlErr) { |
| 99 | + return rlErr.RetryAfter |
| 100 | + } |
| 101 | + return 0 |
| 102 | +} |
| 103 | + |
| 104 | +// WithRetry executes op with exponential backoff, honoring context |
| 105 | +// cancellation and any provider-requested Retry-After delay reported via |
| 106 | +// opts.ExtractRetryAfter. |
| 107 | +func WithRetry[T any](ctx context.Context, op func() (T, error), opts RetryOptions) (T, error) { |
| 108 | + classify := opts.Classify |
| 109 | + if classify == nil { |
| 110 | + classify = DefaultClassify |
| 111 | + } |
| 112 | + extractRetryAfter := opts.ExtractRetryAfter |
| 113 | + if extractRetryAfter == nil { |
| 114 | + extractRetryAfter = DefaultExtractRetryAfter |
| 115 | + } |
| 116 | + |
| 117 | + var result T |
| 118 | + var lastErr error |
| 119 | + |
| 120 | + for attempt := 0; attempt <= opts.MaxRetries; attempt++ { |
| 121 | + result, lastErr = op() |
| 122 | + if lastErr == nil { |
| 123 | + return result, nil |
| 124 | + } |
| 125 | + |
| 126 | + if !classify(lastErr) { |
| 127 | + return result, lastErr |
| 128 | + } |
| 129 | + if attempt >= opts.MaxRetries { |
| 130 | + return result, lastErr |
| 131 | + } |
| 132 | + |
| 133 | + // Exponential backoff: BaseDelay, 2x, 4x, 8x... capped at MaxDelay. |
| 134 | + delay := opts.BaseDelay * time.Duration(1<<uint(attempt)) |
| 135 | + if delay > opts.MaxDelay { |
| 136 | + delay = opts.MaxDelay |
| 137 | + } |
| 138 | + |
| 139 | + // A provider-requested Retry-After overrides backoff, but is still |
| 140 | + // capped at MaxDelay so a runaway "Retry-After: 3600" can't stall |
| 141 | + // the caller for an hour. |
| 142 | + if retryAfter := extractRetryAfter(lastErr); retryAfter > 0 { |
| 143 | + if retryAfter > opts.MaxDelay { |
| 144 | + retryAfter = opts.MaxDelay |
| 145 | + } |
| 146 | + delay = retryAfter |
| 147 | + } |
| 148 | + |
| 149 | + select { |
| 150 | + case <-ctx.Done(): |
| 151 | + return result, ctx.Err() |
| 152 | + case <-time.After(delay): |
| 153 | + } |
| 154 | + } |
| 155 | + |
| 156 | + return result, lastErr |
| 157 | +} |
| 158 | + |
| 159 | +// WithRetryVoid is like WithRetry but for operations that don't return a value. |
| 160 | +func WithRetryVoid(ctx context.Context, op func() error, opts RetryOptions) error { |
| 161 | + _, err := WithRetry(ctx, func() (struct{}, error) { |
| 162 | + return struct{}{}, op() |
| 163 | + }, opts) |
| 164 | + return err |
| 165 | +} |
0 commit comments