Golang HTTP request retry code

Question

I believe that clean and elegant retry code is notoriously difficult, so I'm wondering how the following code could be improved. The retryGet function should have the same interface and contract as http.Get, except for the number of attempts argument, of course.

func retryGet(url string, maxAttempts int) (*http.Response, error) {
 attempts := 0
 if maxAttempts < 1 {
 return nil, errors.New("maxAttempts must be at least 1")
 }
 for {
 attempts++
 response, err := http.Get(url)
 if err == nil && response.StatusCode == http.StatusOK {
 return response, nil
 }
 delay, retry := shouldRetry(maxAttempts, attempts, response)
 if !retry {
 return response, err
 }
 if err == nil {
 defer response.Body.Close()
 }
 time.Sleep(delay)
 }
}
func shouldRetry(maxAttempts, attempts int, response *http.Response) (time.Duration, bool) {
 if attempts >= maxAttempts {
 return time.Duration(0), false
 }
 delay := time.Duration(attempts) * time.Second
 if response != nil && response.Header.Get("Retry-After") != "" {
 after, err := strconv.ParseInt(response.Header.Get("Retry-After"), 10, 64)
 if err != nil && after > 0 {
 delay = time.Duration(after) * time.Second
 }
 }
 return delay, true
}

Answer 1

I would recommend you to actually strictly match the http.Get signature using a struct:

type Getter func(url string) (*http.Response, error)
type RetryingClient struct {
 Getter Getter
 MaxAttempts int
}
func (rc RetryingClient) Get(url string) (*http.Response, error) {
 ...
}

And instead of directly calling http.Get, the embedded Getter interface gives you much more flexibility:

• for testing
• for using a proxy

---

Why do you defer the response.Body.Close() ? Since it won't be use, you can close it immediately.

---

Using the interface, you could make a function embedded in the struct to compute a default delay instead of hard-coded time.Duration(attempts) * time.Second. For instance: defaultDelay(attempt int) time.Duration