Jan Šafránek

/remove-lifecycle rotten

/lifecycle frozen This needs to wait until https://github.com/kubernetes/kubernetes/issues/89173 is fixed

@humblec I remember you did some crazy stuff with your CSI drivers that don't really support ControllerPublish. Do you actually implement a dummy ControllerPublish in your driver and run a...

@humblec would it be possible for you to write a dummy ControllerPublish, so we can remove the dummy attacher mode?

@zhucan, please do not forget to update https://github.com/kubernetes-csi/docs/blob/master/book/src/external-provisioner.md

/lgtm /approve

> It is actually the external-attacher's job to "pick" the correct access mode based on other information (or some heuristic) That **is** the heuristics. See https://github.com/kubernetes-csi/external-attacher/blob/08983ee75bd777ac779ef1e5e7a746ff72be339d/pkg/controller/util.go#L182-L204 For `[ReadOnlyMany, ReadWriteOnce]` PV,...

> According to kubernetes/kubernetes#70505, external-attacher also uses readOnly field in CSIPersistentVolumeSource. > Can we not ignore/override accessmode check in this case? `PV.Spec.CSI.ReadOnly=true` and `AccessMode=[ReadOnlyMany, ReadWriteOnce]` don't go well together -...

Hmm, that's odd. VolumeAttachment is marked for deletion (`finalizers: [external-attacher/com-google-csi-gcepd]`). Question is why the attacher did not remove the finalizer in VolumeAttachment. This could mean that Detach() did not succeed...

@verult, can you please check again if it's still reproducible? I really need the logs.