build error due to csi in container-storage-interface API changed

[andyzhangx]

Anyone know how to work around this issue?

root@andy-dev:~/go/src/github.com/kubernetes-csi/drivers# make hostpath
if [ ! -d ./vendor ]; then dep ensure; fi
go build -i -o _output/hostpathplugin ./app/hostpathplugin
# github.com/kubernetes-csi/drivers/pkg/csi-common
pkg/csi-common/nodeserver-default.go:47:66: undefined: csi.NodeGetIdRequest
pkg/csi-common/nodeserver-default.go:47:90: undefined: csi.NodeGetIdResponse
pkg/csi-common/nodeserver-default.go:55:10: undefined: csi.NodeGetIdResponse
pkg/csi-common/utils.go:55:8: cannot use major (type int32) as type uint32 in field value
pkg/csi-common/utils.go:56:8: cannot use minor (type int32) as type uint32 in field value
pkg/csi-common/utils.go:57:8: cannot use patch (type int32) as type uint32 in field value
Makefile:30: recipe for target 'hostpath' failed
make: *** [hostpath] Error 2

Using master branch of github.com/container-storage-interface/spec/lib/go/csi is incorrect now since csi api has changed.
@rootfs

[andyzhangx]

just could not find the correct commit:

git checkout 2571bd3ff69187efbaef51f3eae6235af59b1614
root@andy-dev:~/go/src/github.com/kubernetes-csi/drivers# make hostpath
if [ ! -d ./vendor ]; then dep ensure; fi
go build -i -o _output/hostpathplugin ./app/hostpathplugin
# github.com/kubernetes-csi/drivers/pkg/csi-common
pkg/csi-common/utils.go:97:9: cannot use ids (type *DefaultIdentityServer) as type csi.IdentityServer in argument to s.Start:
        *DefaultIdentityServer does not implement csi.IdentityServer (missing GetPluginCapabilities method)
pkg/csi-common/utils.go:105:9: cannot use ids (type *DefaultIdentityServer) as type csi.IdentityServer in argument to s.Start:
        *DefaultIdentityServer does not implement csi.IdentityServer (missing GetPluginCapabilities method)
pkg/csi-common/utils.go:113:9: cannot use ids (type *DefaultIdentityServer) as type csi.IdentityServer in argument to s.Start:
        *DefaultIdentityServer does not implement csi.IdentityServer (missing GetPluginCapabilities method)
Makefile:30: recipe for target 'hostpath' failed

[andyzhangx]

v0.1.0 works, while there is build error for master branch

[sbezverk]

@andyzhangx Currently CSI master points to CSI Spec 0.2.0 which introduced breaking changes. There is an effort to convert all existing plugins, as well as provisioner/attacher/registrar to support CSI Spec 0.2.0. Here is my PR which addresses hostpath plugin and csi-common 0.2.0 changes:
#56
Please keep in mind that you still need new provisioner and registrar. New external attacher was merged this morning.

[andyzhangx]

@sbezverk thanks, I am going to write a new CSI driver, so should I start with v0.1.0 or just wait for 0.2.0? I still need new provisioner and registrar for 0.2.0, it looks like v0.2.0 is still on-going.

[sbezverk]

@andyzhangx please do not bother doing anything for 0.1.0, registrar and attacher are already merged, provisioner will be merged tomorrow. You can start working for 0.2.0 right away :)

[andyzhangx]

@sbezverk thanks for clarification, I just would like to make it very detailed, so when the following images would be ready in docker hub, do I need to build them myself?

docker.io/k8scsi/csi-provisioner:0.2
docker.io/k8scsi/csi-attacher:0.2
docker.io/k8scsi/driver-registrar:0.2

Is there any example config when 0.2.0 is ready, I got an 0.1.0 example from here:
https://raw.githubusercontent.com/lpabon/csi-workspace/master/demo/hostPath.yaml

[sbezverk]

@andyzhangx the images will be available at quay.io/k8scsi
See this link for my hostpath plugin yaml: https://gist.github.com/sbezverk/2229db8d27968c1a968e3a4ab5dabe14

[andyzhangx]

@sbezverk thanks, you are awesome!

[andyzhangx]

hi @sbezverk you are using image from 192.168.80.240:4000/k8scsi/csi-provisioner:v0.2.0, could you put those images into docker pub? thanks.

[arm64b]

Hello @sbezverk , I've setup an CSI-enabled k8s cluster with kubeadm, but when I create the csi-pod with:

#kubectl create -f https://raw.githubusercontent.com/kubernetes-csi/docs/master/book/src/example/csi-setup.yaml

The command kubectl get pod always shows:

s# kubectl get pod
NAME      READY     STATUS              RESTARTS   AGE
csi-pod   0/4       ContainerCreating   0          3m

Do I need to adapt that yml file to my local environment, but seems there's not env specific settings in that yaml file. Any suggestions?

[andyzhangx]

external-provisioner is crahsing, it says:

#kubectl logs -p csi-pod -c external-provisioner
I0315 08:14:04.529371       1 csi-provisioner.go:68] Building kube configs for running in cluster...
I0315 08:14:04.561105       1 controller.go:84] Connecting to /csi/csi.sock
I0315 08:14:04.561319       1 controller.go:111] Still trying, connection is CONNECTING
I0315 08:14:04.561816       1 controller.go:108] Connected
I0315 08:14:04.561879       1 controller.go:75] GRPC call: /csi.v0.Identity/GetPluginCapabilities
I0315 08:14:04.561961       1 controller.go:76] GRPC request:
I0315 08:14:04.564703       1 controller.go:78] GRPC response: capabilities:<service:<> >
I0315 08:14:04.564749       1 controller.go:79] GRPC error: <nil>
F0315 08:14:04.564792       1 controller.go:198] no plugin controller service support detected

[arm64b]

@andyzhangx are you using the cmd #kubectl create -f https://raw.githubusercontent.com/kubernetes-csi/docs/master/book/src/example/csi-setup.yaml to create the csi-pod? But the output of #kubectl logs -p csi-pod -c external-provisioner in my side is:

# kubectl logs -p csi-pod -c external-provisioner
Error from server (BadRequest): previous terminated container "external-provisioner" in pod "csi-pod" not found

[andyzhangx]

@arm64b yes, i am using the same cmd, you may delete it and recreate.

[arm64b]

Yes, I delete it, the kubectl get pod shows No resources found, then I create it again with kubectl create -f https://raw.githubusercontent.com/kubernetes-csi/docs/master/book/src/example/csi-setup.yaml:

storageclass "csi-hostpath-sc" created
serviceaccount "csi-service-account" created
clusterrole "csi-cluster-role" created
clusterrolebinding "csi-role-binding" created
pod "csi-pod" created

The kubectl get pod output then:

NAME      READY     STATUS              RESTARTS   AGE
csi-pod   0/4       ContainerCreating   0          3s

After that:

# kubectl logs -p csi-pod -c external-provisioner
Error from server (BadRequest): previous terminated container "external-provisioner" in pod "csi-pod" not found

[pohly]

F0315 08:14:04.564792 1 controller.go:198] no plugin controller service support detected

This was found and fixed yesterday, see PR #69. The images need to be updated to reflect that.