Retry is a Go library for facilitating retry logic and backoff. It's highly extensible with full control over how and when retries occur. You can also write your own custom backoff functions by implementing the Backoff interface.

• Extensible - Inspired by Go's built-in HTTP package, this Go backoff and retry library is extensible via middleware. You can write custom backoff functions or use a provided filter.

• Independent - No external dependencies besides the Go standard library, meaning it won't bloat your project.

• Concurrent - Unless otherwise specified, everything is safe for concurrent use.

• Context-aware - Use native Go contexts to control cancellation.

Here is an example use for connecting to a database using Go's database/sql package:

package main

import (
  "context"
  "database/sql"
  "log"
  "time"

  "github.com/sethvargo/go-retry"
)

func main() {
  db, err := sql.Open("mysql", "...")
  if err != nil {
    log.Fatal(err)
  }

  ctx := context.Background()
  if err := retry.Fibonacci(ctx, 1*time.Second, func(ctx context.Context) error {
    if err := db.PingContext(ctx); err != nil {
      // This marks the error as retryable
      return retry.RetryableError(err)
    }
    return nil
  }); err != nil {
    log.Fatal(err)
  }
}

In addition to your own custom algorithms, there are built-in algorithms for backoff in the library.

A very rudimentary backoff, just returns a constant value. Here is an example:

1s -> 1s -> 1s -> 1s -> 1s -> 1s

Usage:

NewConstant(1 * time.Second)

Arguably the most common backoff, the next value is double the previous value. Here is an example:

1s -> 2s -> 4s -> 8s -> 16s -> 32s -> 64s

Usage:

NewExponential(1 * time.Second)

The Fibonacci backoff uses the Fibonacci sequence to calculate the backoff. The next value is the sum of the current value and the previous value. This means retires happen quickly at first, but then gradually take slower, ideal for network-type issues. Here is an example:

1s -> 1s -> 2s -> 3s -> 5s -> 8s -> 13s

Usage:

NewFibonacci(1 * time.Second)

The built-in backoff algorithms never terminate and have no caps or limits - you control their behavior with middleware. There's built-in middleware, but you can also write custom middleware.

To reduce the changes of a thundering herd, add random jitter to the returned value.

b := NewFibonacci(1 * time.Second)

// Return the next value, +/- 500ms
b = WithJitter(500*time.Millisecond, b)

// Return the next value, +/- 5% of the result
b = WithJitterPercent(5, b)

To terminate a retry, specify the maximum number of retries. Note this is retries, not attempts. Attempts is retries + 1.

b := NewFibonacci(1 * time.Second)

// Stop after 4 retries, when the 5th attempt has failed. In this example, the worst case elapsed
// time would be 1s + 1s + 2s + 3s = 7s.
b = WithMaxRetries(4, b)

To ensure an individual calculated duration never exceeds a value, use a cap:

b := NewFibonacci(1 * time.Second)

// Ensure the maximum value is 2s. In this example, the sleep values would be
// 1s, 1s, 2s, 2s, 2s, 2s...
b = WithCappedDuration(2 * time.Second, b)

For a best-effort limit on the total execution time, specify a max duration:

b := NewFibonacci(1 * time.Second)

// Ensure the maximum total retry time is 5s.
b = WithMaxDuration(5 * time.Second, b)

Here are benchmarks against some other popular Go backoff and retry libraries. You can run these benchmarks yourself via the benchmark/ folder. Commas and spacing fixed for clarity.

Benchmark/cenkalti-7      13,052,668     87.3 ns/op
Benchmark/lestrrat-7         902,044    1,355 ns/op
Benchmark/sethvargo-7    203,914,245     5.73 ns/op

• Randomization uses math/rand seeded with the Unix timestamp instead of crypto/rand.
• Ordering of addition of multiple modifiers will make a difference. For example; ensure you add CappedDuration before WithMaxDuration, otherwise it may early out too early. Another example is you could add Jitter before or after capping depending on your desired outcome.