The fault package provides go http middleware that makes it easy to inject faults into your service. Use the fault package to reject incoming requests, respond with an HTTP error, inject latency into a percentage of your requests, or inject any of your own custom faults.
The fault package works through standard go http middleware. You first create an Injector
, which is a middleware with the code to be run on injection. Then you wrap that Injector
in a Fault
which handles logic about when to run your Injector
.
There are currently three kinds of injectors: SlowInjector
, ErrorInjector
, and RejectInjector
. Each of these injectors can be configured through a Fault
to run on a small percent of your requests. You can also configure the Fault
to blocklist/allowlist certain paths.
See the usage section below for an example of how to get started and the godoc for further documentation.
This package is useful for safely testing failure scenarios in go services that can make use of net/http
handlers/middleware.
One common failure scenario that we cannot perfectly simulate is dropped requests. The RejectInjector
will always return immediately to the user, but in many cases requests can be dropped without ever sending a response. The best way to simulate this scenario using the fault package is to chain a SlowInjector
with a very long wait time in front of an eventual RejectInjector
.
This project is in a stable and supported state. There are no plans to introduce significant new features however we welcome and encourage any ideas and contributions from the community. Contributions should follow the guidelines in our CONTRIBUTING.md.
// main.go
package main
import (
"net/http"
"time"
"github.com/github/go-fault"
)
var mainHandler = http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
http.Error(w, http.StatusText(http.StatusOK), http.StatusOK)
})
func main() {
slowInjector, _ := fault.NewSlowInjector(time.Second * 2)
slowFault, _ := fault.NewFault(slowInjector,
fault.WithEnabled(true),
fault.WithParticipation(0.25),
fault.WithPathBlocklist([]string{"/ping", "/health"}),
)
// Add 2 seconds of latency to 25% of our requests
handlerChain := slowFault.Handler(mainHandler)
http.ListenAndServe("127.0.0.1:3000", handlerChain)
}
This package uses standard go tooling for testing and development. The go language is all you need to contribute. Tests use the popular testify/assert which will be downloaded automatically the first time you run tests. GitHub Actions will also run a linter using golangci-lint after you push. You can also download the linter and use golangci-lint run
to run locally.
The fault package has extensive tests that are run in GitHub Actions on every push. Code coverage is 100% and is published as an artifact on every Actions run.
You can also run tests locally:
$ go test -v -cover -race ./...
[...]
PASS
coverage: 100.0% of statements
ok github.com/github/go-fault 0.575s
The fault package is safe to leave implemented even when you are not running a fault injection. While the fault is disabled there is negligible performance degradation compared to removing the package from the request path. While enabled there may be minor performance differences, but this will only be the case while you are already injecting faults.
Benchmarks are provided to compare without faults, with faults disabled, and with faults enabled. Benchmarks are uploaded as artifacts in GitHub Actions and you can download them from any Validate Workflow.
You can also run benchmarks locally (example output):
$ go test -run=XXX -bench=.
goos: darwin
goarch: amd64
pkg: github.com/github/go-fault
BenchmarkNoFault-8 684826 1734 ns/op
BenchmarkFaultDisabled-8 675291 1771 ns/op
BenchmarkFaultErrorZeroPercent-8 667903 1823 ns/op
BenchmarkFaultError100Percent-8 663661 1833 ns/op
PASS
ok github.com/github/go-fault 8.814s
This project is licensed under the MIT License.