Skip to content

Latest commit

 

History

History
235 lines (170 loc) · 7.24 KB

README.md

File metadata and controls

235 lines (170 loc) · 7.24 KB
Raw

Delphix CSI driver

Please check the Design Proposal

Kubernetes

Clone this repo into your GOPATH

cd $GOPATH
mkdir -p gitlab.delphix.com/daniel.stolf/
cd gitlab.delphix.com/daniel.stolf/
git clone https://gitlab.delphix.com/daniel.stolf/delphix-csi-driver

Requirements

Build a local multi-node Kubernetes cluster on VMWare

cd hack && vagrant up

Beware that the nodes will have IP addresses 172.168.0.10, 11, 12 etc.

K8S has a pod network with CIDR IP range in 192.168.0.0/24.

If there's a need to change this ranges, please get in contact at the #delphix-csi Slack channel

This will build a cluster running the latest version of Kubernetes, with all the needed feature gates and pre-reqs needed to run a CSI Driver.

Deploy K8S Delphix Pluin

Deploy the K8S Python Plugin to the Engine.

cd plugin
dvp build && dvp upload -e 172.168.26.9 -u admin --password delphix

Add k8s-master Environment to Delphix

Add k8s-master (172.168.0.10) as an environment on Delphix and provision an empty vFile like any other Linux target.

The follwing NFS Addresses are configure to this Environment:

172.168.26.10, 172.16.251.174, 172.168.26.12, 172.17.0.1, 192.168.247.0, 172.16.251.173, 172.168.26.11, 172.17.0.1, 192.168.84.128

This list is proved to work, but it might cover more addresses than necessary.

It is NOT necessary that for environment to be a Kubernetes Cluster Node, we're doing this for the sake of simplicty. It could be a bastion node with access to the cluster, kubectl binary and kubeconfig set.

The plugin will scan for kubectl and configured clusters in $KUBECONFIG environment variable (default to ~/.kube/config)

A Kubernetes Repository and a new source should be visible. Add this source: Kubernetes Repo/Source Discovery

This is an empty source, with nothing inside. We'll clone this to provision new empty volumes to Kubernetes.

Take note of this empty source reference, we're using that later! Kubernetes Empty Source

Deploy

The Manifests in the deploy folder have not been fully tested yet.

Create a Kubernetes Secret with Delphix credential

kubectl create secret generic delphix-creds --from-literal=username=admin --from-literal=password=delphix -n kube-system

Edit deploy/kubernetes/releases/delphix-config.yaml and apply the appropriate configs (Engine URL, group name the volumes should be created in, repository name, source reference we've saved earlier etc).

Create the config map

kubectl apply -f deploy/delphix-config.yaml

Deploy the Driver to Kubernetes. The VolumeSnapshotClass will throw an error at first, but that's because it should be created after the controllerserver is already running

kubectl apply -f deploy/kubernetes/releases/csi-delphix-v0.0.2.yaml

Check that de-controller and de-node pods are running. There should be 1 controller with a stable name (because it's a StatefulSet) and as many node pods as there are Kubernetes worker nodes (because it's a DaemonSet).

$ kubectl get pods -n kube-system
NAME                                      READY   STATUS    RESTARTS   AGE
...
csi-de-controller-0                       4/4     Running   0          91s
csi-de-node-m4l7g                         2/2     Running   0          91s
csi-de-node-vtqpx                         2/2     Running   0          91s
...

Create a Delphix Volume in Kubernetes:

kubectl apply -f examples/kubernetes/pvc.yaml

Create a Snapshot:

kubectl apply -f examples/kubernetes/snapshot.yaml

Restore snapshot (not working properly yet):

kubectl apply -f examples/kubernetes/restore.yaml

Example application - Use Helm to provision a MySQL instance backed by Delphix Volumes:

Install Helm

Deploy MySQL Chart with Helm:

examples/kubernetes/mysql
helm install --name my-release -f values.yaml ./

Next Steps

  • Driver should create a Delphix Environment on its own
  • Driver should get empty volume reference on its own
  • An operator pod should dinamically update NFS Addresses when new nodes are added or removed
  • Sample Delphix SDK plugin showcasing database migrations to Kubernetes

End Goals

  • Feature full driver, with shareable volumes
  • Leverage Kubernetes Operators to expand our supported platforms
  • Migrate standard databases to Kubernetes?
  • What else...?

Build

Just

export GOPRIVATE=gitlab.delphix.com
make

Testing the driver locally

Make sure you've added your Linux environment to Delphix!

Get the CSC tool

Get the Container Storage Client, which will emulate the gRPC calls to the driver

go get github.com/rexray/gocsi/csc

Run the driver in a session

sudo bin/delphixplugin --endpoint tcp://127.0.0.1:10000 --url http://172.168.26.9 --username admin --password delphix --envName k8s-master --sourceRef APPDATA_CONTAINER-224 --groupName K8S-Volumes --repositoryName "Kubernetes kubernetes-admin@kubernetes - v1.16.0" --mountPath "/mnt/provision" --nodeid CSINode -v=5

Get plugin info

On another session

csc identity plugin-info --endpoint tcp://127.0.0.1:10000
"defs.csi.delphix.com"	"596cd4bba146149eb69a65b997806607aea793bc"

Create a Delphix Volume

ret=$(csc controller new --endpoint tcp://127.0.0.1:10000 --cap MULTI_NODE_MULTI_WRITER,mount,nfs --req-bytes 10241024 CSIVolumeName)
ret=$(echo $ret | awk '{print $1 " " $5 " " $8}')
volID=$(echo $ret | awk '{print $1}' | cut -d\" -f2)
sharedPath=$(echo $ret | awk '{print $2}' | cut -d= -f2 | cut -d\" -f2)
volReference=$(echo $ret | awk '{print $3}' | cut -d= -f2 | cut -d\" -f2)

This will save the volID, volReference and sharedPath environment variables, which we will use to provide the context to NodePublish (in a cluster, Kubernetes and the CSI SideCars would do that)

Publish (mount) a volume:

csc node publish --endpoint tcp://127.0.0.1:10000 --cap MULTI_NODE_MULTI_WRITER,mount,nfs \
--target-path /csi-data-dir/$volID \
--vol-context VolDelphixSharedPath=$sharedPath \
--vol-context VolID=$volID \
--vol-context VolDelphixReference=$volReference \
$volID

Check that mount was successful

ls -larthd /csi-data-dir/$volID
df -kh /csi-data-dir/$volID

Take a snapshot

csc controller snap --endpoint tcp://127.0.0.1:10000 --source-volume $volID snap1

Clone from snapshot

kubectl apply -f examples/kubernetes/restore_from_snapshot.yaml

Clone from Volume

Clone from volume isn't working yet. The problem with cloning from snapshot was related to the manifest, not the driver code. This is probably the case again.

Unpublish (umount) Volume

csc node unpublish --endpoint tcp://127.0.0.1:10000 \
--target-path /csi-data-dir/$volID \
$volID

Check that volume was umounted:

ls -larthd /csi-data-dir/$volID
df -kh /csi-data-dir/$volID

Delete Volume

csc controller del --endpoint tcp://127.0.0.1:10000 $volID

Clone from Source or Snapshot

TBD! It's implemented, but csc still doesn't support this.

Rollback

Not in the CSI Spec, so out of scope of this project for now.

Running Kubernetes End To End tests on Delphix Driver

TBD