Heimdall is an HTTP client that helps your application make a large number of requests, at scale. With Heimdall, you can:
- Use a hystrix-like circuit breaker to control failing requests
- Add synchronous in-memory retries to each request, with the option of setting your own retrier strategy
- Create clients with different timeouts for every request
All HTTP methods are exposed as a fluent interface.
go get -u github.com/gojektech/heimdall
The below example will print the contents of the google home page:
// Create a new HTTP client with a default timeout
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
// Use the clients GET method to create and execute the request
res, err := client.Get("http://google.com", nil)
if err != nil{
panic(err)
}
// Heimdall returns the standard *http.Response object
body, err := ioutil.ReadAll(res.Body)
fmt.Println(string(body))You can also use the *http.Request object with the http.Do interface :
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
// Create an http.Request instance
req, _ := http.NewRequest(http.MethodGet, "http://google.com", nil)
// Call the `Do` method, which has a similar interface to the `http.Do` method
res, err := client.Do(req)
if err != nil {
panic(err)
}
body, err := ioutil.ReadAll(res.Body)
fmt.Println(string(body))You can use the NewHystrixHTTPClient function to create a client wrapped in a hystrix-like circuit breaker:
// Create a new hystrix config, and input the command name, along with other required options
hystrixConfig := heimdall.NewHystrixConfig("google_get_request", heimdall.HystrixCommandConfig{
ErrorPercentThreshold : 20,
MaxConcurrentRequests: 30,
Timeout: 1000,
})
timeout := 10 * time.Millisecond
// Create a new hystrix-wrapped HTTP client
client := heimdall.NewHystrixHTTPClient(timeout, hystrixConfig)
// The rest is the same as the previous exampleIn the above example, there are two timeout values used: one for the hystrix configuration, and one for the HTTP client configuration. The former determines the time at which hystrix should register an error, while the latter determines when the client itself should return a timeout error. Unless you have any special requirements, both of these would have the same values.
You can use the NewHystrixHTTPClient function to create a client wrapped in a hystrix-like circuit breaker by passing in your own custom fallbacks:
The fallback function will trigger when your code returns an error, or whenever it is unable to complete based on a variety of health checks.
How your fallback function should look like you should pass in a function whose signature looks like following
func(err error) error {
// your logic for handling the error/outage condition
return err
}Example
// Create a new fallback function
fallbackFn := func(err error) error {
_, err := http.Post("post_to_channel_two")
return err
}
hystrixConfig := heimdall.NewHystrixConfig("MyCommand", heimdall.HystrixCommandConfig{
Timeout: 1100,
MaxConcurrentRequests: 100,
ErrorPercentThreshold: 25,
SleepWindow: 10,
RequestVolumeThreshold: 10,
FallbackFunc: fallbackFn,
})
timeout := 10 * time.Millisecond
// Create a new hystrix-wrapped HTTP client with the fallbackFunc as fall-back function
client := heimdall.NewHystrixHTTPClient(timeout, hystrixConfig)
// The rest is the same as the previous exampleIn the above example, the fallbackFunc is a function which posts to channel two in case posting to channel one fails.
// First set a backoff mechanism. Constant backoff increases the backoff at a constant rate
backoffInterval := 2 * time.Millisecond
maximumJitterInterval := 5 * time.Millisecond
backoff := heimdall.NewConstantBackoff(backoffInterval, maximumJitterInterval)
// Create a new retry mechanism with the backoff
retrier := heimdall.NewRetrier(backoff)
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
// Set the retry mechanism for the client, and the number of times you would like to retry
client.SetRetrier(retrier)
client.SetRetryCount(4)
// The rest is the same as the first exampleOr create client with exponential backoff
// First set a backoff mechanism. Exponential Backoff increases the backoff at a exponential rate
initalTimeout := 2*time.Millisecond // Inital timeout
maxTimeout := 9*time.Millisecond // Max time out
exponentFactor := 2 // Multiplier
maximumJitterInterval := 2*time.Millisecond // Max jitter interval
backoff := heimdall.NewExponentialBackoff(initalTimeout, maxTimeout, exponentFactor, maximumJitterInterval)
// Create a new retry mechanism with the backoff
retrier := heimdall.NewRetrier(backoff)
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
// Set the retry mechanism for the client, and the number of times you would like to retry
client.SetRetrier(retrier)
client.SetRetryCount(4)
// The rest is the same as the first exampleThis will create an HTTP client which will retry every 500 milliseconds incase the request fails. The library also comes with an Exponential Backoff
Heimdall supports custom retry strategies. To do this, you will have to implement the Backoff interface:
type Backoff interface {
Next(retry int) time.Duration
}Let's see an example of creating a client with a linearly increasing backoff time:
First, create the backoff mechanism:
type linearBackoff struct {
backoffInterval int
}
func (lb *linearBackoff) Next(retry int) time.Duration{
if retry <= 0 {
return 0 * time.Millisecond
}
return time.Duration(retry * lb.backoffInterval) * time.Millisecond
}This will create a backoff mechanism, where the retry time will increase linearly for each retry attempt. We can use this to create the client, just like the last example:
backoff := &linearBackoff{100}
retrier := heimdall.NewRetrier(backoff)
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
client.SetRetrier(retrier)
client.SetRetryCount(4)
// The rest is the same as the first exampleHeimdall also allows you to simply pass a function that returns the retry timeout. This can be used to create the client, like:
linearRetrier := NewRetrierFunc(func(retry int) time.Duration {
if retry <= 0 {
return 0 * time.Millisecond
}
return time.Duration(retry) * time.Millisecond
})
timeout := 1000 * time.Millisecond
client := heimdall.NewHTTPClient(timeout)
client.SetRetrier(linearRetrier)
client.SetRetryCount(4)Heimdall supports custom HTTP clients. This is useful if you are using a client imported from another library and/or wish to implement custom logging, cookies, headers etc for each request that you make with your client.
Under the hood, the httpClient struct now accepts Doer, which is the standard interface implemented by HTTP clients (including the standard library's net/*http.Client)
Let's say we wish to add authorization headers to all our requests.
We can define our client myHTTPClient
type myHTTPClient struct {
client http.Client
}
func (c *myHTTPClient) Do(request *http.Request) (*http.Response, error) {
request.SetBasicAuth("username", "passwd")
return c.client.Do(request)
}And set this with httpClient.SetCustomHTTPClient(&myHTTPClient{client: http.DefaultClient})
Now, each sent request will have the Authorization header to use HTTP basic authentication with the provided username and password.
This can be done for the hystrix client as well
hystrixClient.SetCustomHTTPClient(&myHTTPClient{
client: http.Client{Timeout: 25 * time.Millisecond}})
// The rest is the same as the first example
Further documentation can be found on godoc.org
Can I replace the standard Go HTTP client with Heimdall?
Yes, you can. Heimdall implements the standard HTTP Do method, along with useful wrapper methods that provide all the functionality that a regular Go HTTP client provides.
When should I use Heimdall?
If you are making a large number of HTTP requests, or if you make requests among multiple distributed nodes, and wish to make your systems more fault tolerant, then Heimdall was made for you.
Heimdall makes use of multiple mechanisms to make HTTP requests more fault tolerant:
- Retries - If a request fails, Heimdall retries behind the scenes, and returns the result if one of the retries are successful.
- Circuit breaking - If Heimdall detects that too many of your requests are failing, or that the number of requests sent are above a configured threshold, then it "opens the circuit" for a short period of time, which prevents any more requests from being made. This gives your downstream systems time to recover.
So does this mean that I shouldn't use Heimdall for small scale applications?
Although Heimdall was made keeping large scale systems in mind, it's interface is simple enough to be used for any type of systems. In fact, we use it for our pet projects as well. Even if you don't require retries or circuit breaking features, the simpler HTTP client provides sensible defaults with a simpler interface, and can be upgraded easily should the need arise.
Can I contribute to make Heimdall better?
Please do! We are looking for any kind of contribution to improve Heimdalls core funtionality and documentation. When in doubt, make a PR!
Copyright 2018, GO-JEK Tech (http://gojek.tech)
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.