etcd-io / gofail Goto Github PK

When trying to disable enabled failpoint through http DELETE method, the server hangs up:

curl
-L http://localhost:22381/github.com/coreos/gofail/examples/ExampleLabelsGo
-X DELETE

Encountered when running this command on the examples folder

If failpoint is injected within for-loop, lock may never be released. And delete call would block.

#13 (comment)

OK, defer is broken in general. e.g., this will starve the delete too:

for ctx.Err() == nil {
  // gofail: var abc struct{}
  // fmt.Println("abc")
  whatever()
}

Useful for etcd-io/etcd#17680

If I use a http and setup a failpoint with a long sleep (e.g. 1s), deactivating it runs into a problem.
Execution and failpoint change run under the same lock. Deactivation will be really delayed as it's racing against execution for the same lock. For example in etcd-io/etcd#17680, I want to inject a sleep into watch write loop, results in continuous execution of the endpoint to sleep. As result it can take a long time to deactivate it.

I would expect that failpoint execution only uses lock to copy the failpoint term, and then execution is done outside of critical section, allowing deactivation to be immediate.

Currently there is only unit test. It would be great if we have an e2e test suite, and it's supposed to be triggered on each PR or commits on master.

Now the package path is calculated based on the relative relationship between GOPATH and the directory where the file is located:

With the go mod support after go1.11, source files can be outside of GOPATH. So when I run gofail, I can see the error: "missing package for %q; using %q as failpoint path".

I can think of several ways to get the package import path more accurately:

1. Search parent directories for the go.mod file which contains the import path of the module.
2. Use cmd.Exec to call go list to get the import path.
3. Use reflect in generated foo.fail.go to determine import path.

I personally think that the third method is better as it relies less on the executing environment. It can be done by generating such code:

type __fp_foo struct{}
var __fp_fooBar *runtime.Failpoint = runtime.NewFailpoint(reflect.TypeOf(__fp_foo{}).PkgPath(), "fooBar")

After #69 gets merged, we are going to release v0.2.0.

cc @ArkaSaha30 @henrybear327 @serathius

func (t *terms) eval() (interface{}, error) {
	t.mu.Lock()
	defer t.mu.Unlock()
	for _, term := range t.chain {
		if term.mods.allow() {
			return term.do(), nil
		}
	}
	return nil, nil
}

return term.do(), nil causes the following term to not be executed

I expect gofail version could display the info similar to etcd.

$ etcd --version
etcd Version: 3.5.5
Git SHA: 19002cfc6
Go Version: go1.16.15
Go OS/Arch: darwin/amd64

Gofail currently uses golang 1.20 which is no longer supported. We should update to ensure gofail to golang 1.21.8.

This would involve three updates;

Additionally - We should also keep golangci-lint up to date by bumping the version here

Let's release the first formal version v0.1.0, and update etcd to pin to the specific version.

We start to track changes in https://github.com/etcd-io/gofail/blob/master/CHANGELOG.md

cc @ptabor @serathius @spzala

WDYT?

A go function may return multiple values, such as func a() (string, int).
In this case, how should I define failpoint variables?

Thank you!