README
¶
🚀 goretry
A minimal, composable, production-ready retry library for Go.
goretry provides a clean fluent API for retrying operations in Go. No external dependencies. Race-free. Generics-native.
✨ Features
| Feature | Detail | |
|---|---|---|
| 🎯 | Context first | ctx.Done() is checked between every attempt — no blocking sleeps |
| 🔗 | Fluent API | retry.New().Attempts(5).Backoff(ExponentialJitter(200ms)).Do(ctx, fn) |
| 🧬 | Generics | DoValue[T] returns typed values — no closures outside the retry loop |
| 🛑 | Permanent errors | retry.Permanent(err) stops the loop before wasting more attempts |
| 🔍 | Conditional retry | RetryIf(func(error) bool) — only retry errors you want retried |
| ⏱️ | MaxDelay | Cap any backoff strategy at a maximum wait duration |
| 📈 | 4 backoff strategies | Constant, Linear, Exponential, ExponentialJitter — or bring your own |
| 🪝 | OnRetry hook | Plug in logging or metrics per attempt |
| ⚡ | Zero allocations | ~2.6 ns/op, 0 allocs on the success path |
| 🪶 | Zero dependencies | Pure stdlib. Ships in <200 LOC |
📦 Installation
go get github.com/chmenegatti/go-retry
Requires Go 1.18+
💻 Quick Start
import retry "github.com/chmenegatti/go-retry"
err := retry.New().
Attempts(5).
Backoff(retry.ExponentialJitter(100 * time.Millisecond)).
Do(ctx, func() error {
return callUnstableService()
})
📖 Usage
Returning a value
user, err := retry.DoValue(ctx,
retry.New().Attempts(3),
func() (*User, error) {
return db.FindUser(id)
},
)
Permanent errors — stop immediately from inside fn
err := retry.New().Attempts(5).Do(ctx, func() error {
resp, err := http.Get(url)
if err != nil {
return err // transient — will retry
}
if resp.StatusCode == 401 {
return retry.Permanent(ErrUnauthorized) // fatal — stop now
}
return nil
})
// Original error is always preserved
errors.Is(err, ErrUnauthorized) // true
Conditional retry with RetryIf
err := retry.New().
Attempts(5).
RetryIf(func(err error) bool {
// Only retry network timeouts, not application errors
var netErr net.Error
return errors.As(err, &netErr) && netErr.Timeout()
}).
Do(ctx, fn)
Capped exponential backoff
retry.New().
Attempts(10).
Backoff(retry.Exponential(100 * time.Millisecond)).
MaxDelay(5 * time.Second). // never wait longer than 5s
Do(ctx, fn)
Logging with OnRetry
retry.New().
Attempts(5).
OnRetry(func(attempt int, err error) {
log.Printf("attempt %d failed: %v", attempt, err)
metrics.Inc("retry.attempt")
}).
Do(ctx, fn)
Real-world examples
🌐 HTTP request with server-error retry
body, err := retry.DoValue(ctx,
retry.New().
Attempts(4).
Backoff(retry.ExponentialJitter(500*time.Millisecond)).
MaxDelay(10*time.Second).
RetryIf(func(err error) bool {
var apiErr *APIError
if errors.As(err, &apiErr) {
return apiErr.Code >= 500 // only retry server errors
}
return true
}),
func() ([]byte, error) {
resp, err := http.Get("https://api.example.com/data")
if err != nil {
return nil, err
}
defer resp.Body.Close()
if resp.StatusCode == 401 {
return nil, retry.Permanent(ErrUnauthorized)
}
if resp.StatusCode >= 400 {
return nil, &APIError{Code: resp.StatusCode}
}
return io.ReadAll(resp.Body)
},
)
🗄️ Database operation
err := retry.New().
Attempts(3).
Backoff(retry.Constant(500 * time.Millisecond)).
RetryIf(func(err error) bool {
return isDeadlock(err) // only retry transient DB conflicts
}).
Do(ctx, func() error {
return db.ExecContext(ctx, "UPDATE orders SET status=? WHERE id=?", "shipped", id)
})
⚡ gRPC transient failure
resp, err := retry.DoValue(ctx,
retry.New().
Attempts(5).
Backoff(retry.ExponentialJitter(100*time.Millisecond)).
RetryIf(func(err error) bool {
code := status.Code(err)
return code == codes.Unavailable || code == codes.DeadlineExceeded
}),
func() (*pb.Response, error) {
return client.Call(ctx, req)
},
)
📐 Backoff Strategies
type Backoff func(attempt int) time.Duration // bring your own!
| Strategy | Formula | Use case |
|---|---|---|
Constant(d) |
d |
Simple polling, queue consumers |
Linear(base) |
attempt × base |
Gradual ramp-up |
Exponential(base) |
base × 2^(attempt-1) |
Standard retry without jitter |
ExponentialJitter(base) |
[d/2, d] random |
Recommended — avoids thundering herd |
⚙️ API Reference
| Method | Default | Description |
|---|---|---|
New() |
— | 3 attempts, 100ms constant backoff |
.Attempts(n) |
3 |
Total call count (including first) |
.Backoff(b) |
Constant(100ms) |
Delay strategy between attempts |
.MaxDelay(d) |
none | Hard cap applied on top of any backoff |
.RetryIf(fn) |
retry all | Predicate — return false to stop retrying |
.OnRetry(fn) |
none | Hook before each sleep (not on final failure) |
.Do(ctx, fn) |
— | Execute and return error |
DoValue[T](ctx, r, fn) |
— | Execute and return (T, error) |
Permanent(err) |
— | Wrap error to abort retry loop immediately |
IsPermanent(err) |
— | Check if any error in the chain is permanent |
🧠 Execution Flow
flowchart TD
Start(["Do / DoValue"]) --> Exec["Execute fn()"]
Exec --> Ok{"err == nil?"}
Ok -->|"Yes"| ReturnNil(["return nil"])
Ok -->|"No"| Perm{"Permanent(err)?"}
Perm -->|"Yes"| ReturnPerm(["return err immediately"])
Perm -->|"No"| Cond{"RetryIf(err)?"}
Cond -->|"false"| ReturnCond(["return err immediately"])
Cond -->|"true"| Last{"Last attempt?"}
Last -->|"Yes"| ReturnErr(["return err"])
Last -->|"No"| Hook["OnRetry hook (if set)"]
Hook --> Sleep["Wait: min(backoff(n), MaxDelay)"]
Sleep -->|"Timer fired"| Exec
Sleep -->|"ctx.Done()"| ReturnCtx(["return ctx.Err()"])
🔬 Comparison
| Library | API Style | Generics | Permanent | RetryIf | MaxDelay | Dependencies |
|---|---|---|---|---|---|---|
| goretry (this) | Fluent/chain | ✅ | ✅ | ✅ | ✅ | 0 |
cenkalti/backoff |
Functional | ❌ | ✅ | ✅ | ✅ | 1 |
avast/retry-go |
Functional opts | ❌ | ✅ | ✅ | ✅ | 0 |
🏗️ Design Philosophy
- Simplicity over features: every method on
Retryis essential and independently useful. - No global state: calling
New()always returns a fresh independent instance. - Context always wins: the library never blocks on a timer longer than necessary.
- Errors are values:
Permanentis just an error wrapper — no special types to import. - Zero allocations on the hot path: successful calls don't allocate.
🧪 Testing & Benchmarks
go test -race -cover ./...
# ok github.com/chmenegatti/go-retry coverage: 98.4% of statements
go test -bench=. -benchmem ./...
# BenchmarkRetrySuccess-12 443903150 2.626 ns/op 0 B/op 0 allocs/op
# BenchmarkRetryFailure-12 5 201000864 ns/op 502 B/op 6 allocs/op
✅ 0 allocations on the success path — the library is invisible when not needed.
🤝 Contributing
- Fork and create a branch:
git checkout -b feat/my-feature - Run tests:
go test -race ./... - Commit:
git commit -am 'feat: ...' - Open a Pull Request
📄 License
MIT — see LICENSE.
Documentation
¶
Overview ¶
Package retry provides a simple and composable retry mechanism for Go.
Package retry provides a minimal, composable retry mechanism for Go. All features are additive and the API is backward compatible.
Index ¶
Examples ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func DoValue ¶
DoValue is a generic helper that executes fn using the provided Retry configuration and returns both the result and any error. This avoids capturing result variables outside the closure when retrying operations that produce a value.
Example:
user, err := retry.DoValue(ctx, retry.New().Attempts(3), func() (*User, error) {
return db.FindUser(id)
})
Example ¶
ExampleDoValue demonstrates returning a typed result from the retry loop.
package main
import (
"context"
"errors"
"fmt"
"log"
"time"
retry "github.com/chmenegatti/go-retry"
)
func main() {
ctx := context.Background()
attempts := 0
result, err := retry.DoValue(ctx,
retry.New().Attempts(3).Backoff(retry.Constant(time.Millisecond)),
func() (string, error) {
attempts++
if attempts < 3 {
return "", errors.New("not ready yet")
}
return "ok", nil
},
)
if err != nil {
log.Printf("failed: %v", err)
return
}
fmt.Println(result)
}
Output: ok
func IsPermanent ¶ added in v1.1.0
IsPermanent reports whether err (or any error in its chain) was wrapped with Permanent.
func Permanent ¶ added in v1.1.0
Permanent wraps err so that the retry loop stops immediately without further attempts. The original error is preserved and can be retrieved with errors.Unwrap or errors.Is/As.
Example:
Do(ctx, func() error {
resp, err := http.Get(url)
if err != nil {
return err // transient — will retry
}
if resp.StatusCode == 401 {
return retry.Permanent(ErrUnauthorized) // fatal — stop immediately
}
return nil
})
Example ¶
ExamplePermanent shows how to instantly stop retrying from inside fn.
package main
import (
"context"
"errors"
"fmt"
retry "github.com/chmenegatti/go-retry"
)
func main() {
ctx := context.Background()
var ErrBadRequest = errors.New("400 bad request")
err := retry.New().Attempts(5).Do(ctx, func() error {
// Wrapping with Permanent signals "don't retry this".
return retry.Permanent(ErrBadRequest)
})
// Unwrap to get the original error.
fmt.Println(errors.Is(err, ErrBadRequest))
}
Output: true
Types ¶
type Backoff ¶ added in v1.0.1
Backoff is a function that, given an attempt number (1-indexed), returns the duration to wait before the next retry.
func Constant ¶ added in v1.0.1
Constant returns a Backoff strategy that always waits the same duration between attempts, regardless of the attempt number.
Example:
r := retry.New().Backoff(retry.Constant(2 * time.Second))
func Exponential ¶ added in v1.0.1
Exponential returns a Backoff strategy where the wait duration doubles with each attempt: delay = base * 2^(attempt-1).
Example:
r := retry.New().Backoff(retry.Exponential(100 * time.Millisecond)) // attempt 1 → 100ms, attempt 2 → 200ms, attempt 3 → 400ms
func ExponentialJitter ¶ added in v1.0.1
ExponentialJitter returns a Backoff strategy similar to Exponential, but adds a random jitter in the range [d/2, d] to avoid the thundering herd problem when many clients retry simultaneously.
Example:
r := retry.New().Backoff(retry.ExponentialJitter(100 * time.Millisecond))
type Retry ¶ added in v1.0.1
type Retry struct {
// contains filtered or unexported fields
}
Retry holds the configuration for a retry operation. Use New() to construct a Retry and chain builder methods before calling Do.
func New ¶
func New() *Retry
New creates a new Retry with sensible defaults: 3 total attempts with a 100ms constant backoff.
Example ¶
ExampleNew demonstrates a basic retry with the default configuration.
package main
import (
"context"
"errors"
"log"
retry "github.com/chmenegatti/go-retry"
)
func main() {
ctx := context.Background()
err := retry.New().
Attempts(3).
Do(ctx, func() error {
// Simulate a transient failure.
return errors.New("service unavailable")
})
if err != nil {
log.Printf("all attempts failed: %v", err)
}
}
Output:
func (*Retry) Attempts ¶ added in v1.0.1
Attempts sets the total number of times fn will be called (including the first attempt). Values less than 1 are treated as 1.
Example:
retry.New().Attempts(5).Do(ctx, fn)
func (*Retry) Backoff ¶ added in v1.0.1
Backoff sets the strategy used to compute the wait duration between attempts. Use the built-in strategies (Constant, Linear, Exponential, ExponentialJitter) or provide a custom func(attempt int) time.Duration.
Example:
retry.New().Backoff(retry.ExponentialJitter(200 * time.Millisecond))
func (*Retry) Do ¶ added in v1.0.1
Do executes fn up to the configured number of Attempts. Between each failed attempt, Do sleeps for the duration returned by the Backoff strategy (capped by MaxDelay if set), or until ctx is cancelled — whichever comes first.
Do returns nil on the first successful execution of fn. Do returns the last error from fn if all attempts fail. Do returns ctx.Err() if the context is cancelled while waiting. Do returns the error immediately (without further retries) if:
- the error was wrapped with Permanent, or
- a RetryIf predicate returns false for that error.
func (*Retry) MaxDelay ¶ added in v1.1.0
MaxDelay caps the delay returned by the backoff strategy. Any computed delay exceeding MaxDelay will be clamped to MaxDelay. This is especially useful with exponential backoffs to prevent excessively long waits.
Example:
retry.New().
Backoff(retry.Exponential(100*time.Millisecond)).
MaxDelay(5 * time.Second)
Example ¶
ExampleRetry_MaxDelay shows capping exponential backoff at a fixed ceiling.
package main
import (
"context"
"errors"
"time"
retry "github.com/chmenegatti/go-retry"
)
func main() {
ctx := context.Background()
retry.New().
Attempts(10).
Backoff(retry.Exponential(100*time.Millisecond)).
MaxDelay(2*time.Second).
Do(ctx, func() error {
return errors.New("fail")
})
}
Output:
func (*Retry) OnRetry ¶ added in v1.0.1
OnRetry registers a callback invoked before each sleep between attempts. The callback receives the current attempt number (1-indexed) and the error from the last execution. It is NOT called after the final failing attempt.
Example:
retry.New().OnRetry(func(attempt int, err error) {
log.Printf("attempt %d failed: %v", attempt, err)
})
func (*Retry) RetryIf ¶ added in v1.1.0
RetryIf sets a predicate that controls whether a given error should trigger a retry. If the predicate returns false, Do returns the error immediately without further attempts. If unset, all errors are retried up to Attempts.
Note: errors wrapped with Permanent always stop the loop, regardless of RetryIf.
Example:
retry.New().RetryIf(func(err error) bool {
return errors.Is(err, io.ErrUnexpectedEOF)
})
Example ¶
ExampleRetry_RetryIf shows how to skip retrying for known fatal errors.
package main
import (
"context"
"errors"
"fmt"
retry "github.com/chmenegatti/go-retry"
)
func main() {
ctx := context.Background()
errFatal := errors.New("fatal: invalid credentials")
err := retry.New().
Attempts(5).
RetryIf(func(err error) bool {
// Only retry transient errors, not authentication failures.
return !errors.Is(err, errFatal)
}).
Do(ctx, func() error {
return errFatal // stopped after 1 attempt
})
fmt.Println(err)
}
Output: fatal: invalid credentials
Source Files