Skip to content

Latest commit

 

History

History
250 lines (189 loc) · 7.9 KB

File metadata and controls

250 lines (189 loc) · 7.9 KB

Kubernetes Development

Manual deployment

To build and install a development version of the driver:

GCE_PD_CSI_STAGING_IMAGE=gcr.io/path/to/driver/image:dev   # Location to push dev image to
make push-container

# Modify controller.yaml and node.yaml in ./deploy/kubernetes/dev to use dev image
GCE_PD_DRIVER_VERSION=dev
./deploy/kubernetes/deploy-driver.sh

To bring down driver:

./deploy/kubernetes/delete-driver.sh

Testing

E2E tests:

The following environment variables will be required for setup and all subsequent test runs.

export PROJECT=$USER-e2e                                             # GCP project used for testing
export GCE_PD_SA_NAME=$PROJECT-pd-sa                                 # Service account name
export IAM_NAME=$GCE_PD_SA_NAME@$PROJECT.iam.gserviceaccount.com     # Service account IAM name
export GCE_PD_SA_DIR=~/credentials                                   # Service account directory
export GOOGLE_APPLICATION_CREDENTIALS=$GCE_PD_SA_DIR/cloud-sa.json   # Service account key

One time setup

The following steps will need to be completed once per project.

Create service account

To create a service account with correct permissions, your test project must:

  1. Be assigned Project IAM Admin and Role Administrator IAM roles.
  2. Be part of an organization that allows creation of service account keys.
    • For example, a Google internal developer project will not work for E2E testing because the org restricts SA key creation.
export ENABLE_KMS=true
export ENABLE_KMS_ADMIN=true
./deploy/setup-project.sh
Create SSH key
  1. Follow instructions to generate an SSH key with the following parameters:
    • KEY_FILENAME: "google_compute_engine"
    • USERNAME: "$USER"
  2. Add your public key (cat ~/.ssh/google_compute_engine.pub) to your GCP project metatdata.

Running tests

./test/run-e2e-local.sh
Debugging

Local stderr will show you all GCE calls and SSH commands that were performed by the test.

GCE failures

You may see errors during instance creation. It is likely your service account is missing permissions.

  1. Make sure the resources are showing up in pantheon under [Compute Engine > VM Instances].
  2. Look for errors in pantheon under [Logging > Logs Explorer].
  3. Double check and re-run the service account creation step.
SSH failures

You may see errors during driver deployment.

  1. Double check that you have an SSH key file in the correct directory.
  2. gcloud compute ssh $VM_NAME --zone $ZONE into the VM instance and attempt to execute the last few commands and look for errors.

Sanity tests

./test/run-sanity.sh

Unit tests:

./test/run-unit.sh

Performance tests

Performance tests are run in automated testing frameworks, but you may wish to run them locally to benchmark changes. For a given test configuration, you need to modify the test-config.yaml file automatically generated from running one of the run-k8s-integration... or run-windows-k8s-integration.sh scripts to enable performance tests.

DriverInfo:
  ...
  PerformanceTestOptions:
    ProvisioningOptions: # deploy 50 1Gi PVCs and measure latency/throughput
      VolumeSize: 1Gi
      Count: 50
      ExpectedMetrics:
        AvgLatency: 10000000000 # 10 seconds
        Throughput: 5

You may modify the parameter values to customize the test.

You also need to modify the StorageClass file pointed to in test-config.yaml to set volumeBindingMode: Immediate, as the performance tests only support this mode.

We will be running the Kubernetes integration tests directly from its repository. Install kubetest and set up a test cluster with the driver installed. Now, cd into $GOPATH/src/k8s.io/kubernetes and run

# pwd=k/k
kubetest <custom flags based on your provider> \
  --check-version-skew=false \
  --test \
  --test_args="--ginkgo.focus=External.Storage.*volume-lifecycle-performance --allowed-not-ready-nodes=10 --node-os-distro=<linux or windows> --storage.testdriver=<path-to-test-config>"

Dependency management

Use dep

dep ensure

To modify dependencies or versions change ./Gopkg.toml

Debugging

We use https://github.com/go-delve/delve and its feature for remote debugging. This feature is only available in the PD CSI Controller (which runs in a linux node).

Requirements:

Steps:

  • Build the PD CSI driver with additional compiler flags.
export GCE_PD_CSI_STAGING_VERSION=latest
export GCE_PD_CSI_STAGING_IMAGE=image/repo/gcp-compute-persistent-disk-csi-driver
make build-and-push-multi-arch-debug
  • Update deploy/kubernetes/overlays/noauth-debug/kustomization.yaml to match the repo you wrote above e.g.
images:
- name: gke.gcr.io/gcp-compute-persistent-disk-csi-driver
  newName: image/repo/gcp-compute-persistent-disk-csi-driver
  newTag: latest
  • Delete and deploy the driver with this overlay
./deploy/kubernetes/delete-driver.sh && \
  GCE_PD_DRIVER_VERSION=noauth-debug ./deploy/kubernetes/deploy-driver.sh

At this point you could verify that delve is running in the controller logs:

API server listening at: [::]:2345
 2021-04-15T18:28:51Z info layer=debugger launching process with args: [/go/src/sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/bin/gce-pd-csi-driver --v=5 --endpoint=unix:/csi/csi.sock]
 2021-04-15T18:28:53Z debug layer=debugger continuing
  • Enable port forwading of the PD CSI controller of port 2345
kubectl -n gce-pd-csi-driver get pods | grep controller | awk '{print $1}' | xargs -I % kubectl -n gce-pd-csi-driver port-forward % 2345:2345
  • Connect to the headless server and issue commands
dlv connect localhost:2345
Type 'help' for list of commands.
(dlv) clearall
(dlv) break pkg/gce-pd-csi-driver/controller.go:509
Breakpoint 1 set at 0x159ba32 for sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/gce-pd-csi-driver.(*GCEControllerServer).ListVolumes() ./pkg/gce-pd-csi-driver/controller.go:509
(dlv) c
> sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/gce-pd-csi-driver.(*GCEControllerServer).ListVolumes() ./pkg/gce-pd-csi-driver/controller.go:509 (hits goroutine(69):1 total:1) (PC: 0x159ba32)
Warning: debugging optimized function
   504:         }
   505: }
   506:
   507: func (gceCS *GCEControllerServer) ListVolumes(ctx context.Context, req *csi.ListVolumesRequest) (*csi.ListVolumesResponse, error) {
   508:         // https//cloud.google.com/compute/docs/reference/beta/disks/list
=> 509:         if req.MaxEntries < 0 {
   510:                 return nil, status.Error(codes.InvalidArgument, fmt.Sprintf(
   511:                         "ListVolumes got max entries request %v. GCE only supports values between 0-500", req.MaxEntries))
   512:         }
   513:         var maxEntries int64 = int64(req.MaxEntries)
   514:         if maxEntries > 500 {
(dlv) req
Command failed: command not available
(dlv) p req
*github.com/container-storage-interface/spec/lib/go/csi.ListVolumesRequest {
        MaxEntries: 0,
        StartingToken: "",
        XXX_NoUnkeyedLiteral: struct {} {},
        XXX_unrecognized: []uint8 len: 0, cap: 0, nil,
        XXX_sizecache: 0,}
(dlv)

See kubernetes-sigs#742 for the implementation details