oc create -f test/testdata/app-scenarios fails to POST

[juanvallejo]

Link to last failure: https://ci.openshift.redhat.com/jenkins/job/test_pull_requests_origin_check/4612/consoleFull#-200384109356c25df4e4b05b79524e5d52

Version

oc v1.3.0-alpha.0+ea204df-dirty
kubernetes v1.3.0+57fb9ac
features: Basic-Auth

Steps To Reproduce

1. [test]

Current Result

FAILURE after 2.917s: test/cmd/admin.sh:350: executing 'oc create -f test/testdata/app-scenarios' expecting success: the command returned the wrong error code

Standard error from the command:
The Service "frontend" is invalid.
spec.clusterIP: Invalid value: "172.30.17.154": provided IP is already allocated

Expected Result

Should succeed

EDIT: updated link to last failure

[stevekuznetsov]

@juanvallejo why the assign to me?

[juanvallejo]

@stevekuznetsov Ah was not sure who to assign to this

[stevekuznetsov]

@juanvallejo if unsure, leave assignment and priority label unset, someone like @danmcp will come by to set them soon

[mfojtik]

@sgallagher can you please spend some time figuring out what is the issue here?

[sgallagher]

@juanvallejo Can you point me at a recent example? The original link has expired.

[juanvallejo]

@sgallagher Hm, the build seems to have been deleted, after checking on ci.openshift.redhat.com/jenkins/. Unfortunately this was the only build I can find this error happening

[miminar]

Here's another occurence: https://ci.openshift.redhat.com/jenkins/job/test_pull_requests_origin_check/4618/console#159471615556c25df4e4b05b79524e5d52

[juanvallejo]

Ah here's one that just happened as well: https://ci.openshift.redhat.com/jenkins/job/test_pull_requests_origin_check/4612/consoleFull#-200384109356c25df4e4b05b79524e5d52

[stevekuznetsov]

The failures seem to be this always:

Error from server: The POST operation against Pod could not be completed at this time, please try again.

[liggitt]

Looks like that message is coming from https://github.com/openshift/origin/blob/master/vendor/k8s.io/kubernetes/pkg/api/rest/create.go#L106

// CheckGeneratedNameError checks whether an error that occurred creating a resource is due
// to generation being unable to pick a valid name.
func CheckGeneratedNameError(strategy RESTCreateStrategy, err error, obj runtime.Object) error {
    if !errors.IsAlreadyExists(err) {
        return err
    }

    objectMeta, kind, kerr := objectMetaAndKind(strategy, obj)
    if kerr != nil {
        return kerr
    }

    if len(objectMeta.GenerateName) == 0 {
        return err
    }

    return errors.NewServerTimeoutForKind(kind.GroupKind(), "POST", 0)
}

that's a terrible error... it looks like it's converting an AlreadyExists error into a server timeout error if it has a generate name

[liggitt]

@smarterclayton looks like the sample data for petset is the culprit... it's creating a petset and all the pods that go with it, so the pod creates will race with the petset controller, and are likely to fail with an AlreadyExists error

[liggitt]

bumping priority, I think we'll hit this flake a lot

[liggitt]

reopening until #10240 goes in as well