Table of Contents generated with DocToc
The CSI Manila driver is able to create and mount OpenStack Manila shares. Snapshots and recovering shares from snapshots is supported as well (support for CephFS snapshots will be added soon).
| Option | Default value | Description |
|---|---|---|
--endpoint |
unix:///tmp/csi.sock |
CSI Manila's CSI endpoint |
--drivername |
manila.csi.openstack.org |
Name of this driver |
--nodeid |
none | ID of this node |
--nodeaz |
none | Availability zone of this node |
--runtime-config-file |
none | Path to the runtime configuration file |
--with-topology |
none | CSI Manila is topology-aware. See Topology-aware dynamic provisioning for more info |
--share-protocol-selector |
none | Specifies which Manila share protocol to use for this instance of the driver. See supported protocols for valid values. |
--fwdendpoint |
none | CSI Node Plugin endpoint to which all Node Service RPCs are forwarded. Must be able to handle the file-system specified in share-protocol-selector. Check out the Deployment section to see why this is necessary. |
--cluster-id |
none | The identifier of the cluster that the plugin is running in. If set then the plugin will add "manila.csi.openstack.org/cluster: <clusterID>" to metadata of created shares. |
Kubernetes storage class parameters for dynamically provisioned volumes
| Parameter | Required | Description |
|---|---|---|
type |
yes | Manila share type |
shareNetworkID |
no | Manila share network ID |
availability |
no | Manila availability zone of the provisioned share. If none is provided, the default Manila zone will be used. Note that this parameter is opaque to the CO and does not influence placement of workloads that will consume this share, meaning they may be scheduled onto any node of the cluster. If the specified Manila AZ is not equally accessible from all compute nodes of the cluster, use Topology-aware dynamic provisioning. |
appendShareMetadata |
no | Append user-defined metadata to the provisioned share. If not empty, this field must be a string with a valid JSON object. The object must consist of key-value pairs of type string. Example: "{..., \"key\": \"value\"}". |
cephfs-mounter |
no | Relevant for CephFS Manila shares. Specifies which mounting method to use with the CSI CephFS driver. Available options are kernel and fuse, defaults to fuse. See CSI CephFS docs for further information. |
cephfs-clientID |
no | Relevant for CephFS Manila shares. Specifies the cephx client ID when creating an access rule for the provisioned share. The same cephx client ID may be shared with multiple Manila shares. If no value is provided, client ID for the provisioned Manila share will be set to some unique value (PersistentVolume name). |
nfs-shareClient |
no | Relevant for NFS Manila shares. Specifies what address has access to the NFS share. Defaults to 0.0.0.0/0, i.e. anyone. |
Kubernetes PV CSI volume attributes for pre-provisioned volumes
| Parameter | Required | Description |
|---|---|---|
shareID |
if shareName is not given |
The UUID of the share |
shareName |
if shareID is not given |
The name of the share |
shareAccessID |
yes | The UUID of the access rule for the share |
cephfs-mounter |
no | Relevant for CephFS Manila shares. Specifies which mounting method to use with the CSI CephFS driver. Available options are kernel and fuse, defaults to fuse. See CSI CephFS docs for further information. |
Note that the Node Plugin of CSI Manila doesn't care about the origin of a share. As long as the share protocol is supported, CSI Manila is able to consume dynamically provisioned as well as pre-provisioned shares (e.g. shares created manually).
Authentication with OpenStack may be done using either user or trustee credentials.
Mandatory secrets: os-authURL, os-region.
Mandatory secrets for user authentication: os-password, os-userID or os-userName, os-domainID or os-domainName, os-projectID or os-projectName.
Optional secrets for user authentication: os-projectDomainID or os-projectDomainName, os-userDomainID or os-userDomainName
Mandatory secrets for application credential authentication: os-applicationCredentialID or os-applicationCredentialName (when os-userID or os-userName and os-domainName are set), os-applicationCredentialSecret.
Mandatory secrets for trustee authentication: os-trustID, os-trusteeID, os-trusteePassword.
Optionally, a custom certificate may be sourced via os-certAuthorityPath (path to a PEM file inside the plugin container). By default, the usual TLS verification is performed. To override this behavior and accept insecure certificates, set os-TLSInsecure to true (defaults to false).
For a client TLS authentication use both os-clientCertPath and os-clientKeyPath (paths to TLS keypair PEM files inside the plugin container).
Topology-aware dynamic provisioning makes it possible to reliably provision and use shares that are not equally accessible from all compute nodes due to storage topology constraints. With topology awareness enabled, administrators can specify the mapping between compute and Manila availability zones. Doing so will instruct the CO scheduler to place the workloads+shares only on nodes that are able to reach the underlying storage.
CSI Manila uses topology.manila.csi.openstack.org/zone topology key to identify node's affinity to a certain compute availability zone.
Each node of the cluster then gets labeled with a key/value pair of topology.manila.csi.openstack.org/zone / value of --nodeaz cmd arg.
This label may be used as a node selector when defining topology constraints for dynamic provisioning. Administrators are also free to pass arbitrary labels, and as long as they are valid node selectors, they will be honored by the scheduler.
Topology-aware storage class example:
Storage AZ does not influence
the placement of workloads. Compute AZs do.
+-----------+ +---------------+
| Manila AZ | | Compute AZs |
| zone-a | apiVersion: storage.k8s.io/v1 | nova-1 |
+-----------+ kind: StorageClass | nova-2 |
| metadata: +---------------+
| name: cephfs-gold |
| provisioner: cephfs.manila.csi.openstack.org |
| parameters: |
+---------+ availability: zone-a |
... |
allowedTopologies: +------------------------+
- matchLabelExpressions:
- key: topology.manila.csi.openstack.org/zone
values:
- nova-1
- nova-2
Shares in zone-a are accessible only from nodes in nova-1 and nova-2.
Enabling topology awareness in Kubernetes
CSI Manila's runtime configuration file is a JSON document for modifying behavior of the driver at runtime.
Schema:
- Root object:
Attribute Type Description nfsNfsConfigConfiguration for NFS shares. Optional. NfsConfig:Attribute Type Description matchExportLocationAddressstringWhen mounting an NFS share, select an export location with matching IP address. No match between this address and at least a single export location for this share will result in an error. Expects a CIDR-formatted address. If prefix is not provided, /32 or /128 prefix is assumed for IPv4 and IPv6 respectively. Optional.
In Kubernetes, you may store this configuration in a ConfigMap and expose it to CSI Manila pods as a volume. Then enter the path to the file populated by the ConfigMap into --runtime-config-file. Demo ConfigMap is located in examples/manila-csi-plugin/runtimeconfig-cm.yaml. If you're deploying CSI Manila with Helm, setting csimanila.runtimeConfig.enabled to true will take care of the setup.
The CSI Manila driver deals with the Manila service only. All node-related operations (attachments, mounts) are performed by a dedicated CSI Node Plugin, to which all Node Service RPCs are forwarded. This means that the operator is expected to already have a working deployment of that dedicated CSI Node Plugin.
A single instance of the driver may serve only a single Manila share protocol. To serve multiple share protocols, multiple deployments of the driver need to be made. In order to avoid deployment collisions, each instance of the driver should be named differently, e.g. csi-manila-cephfs, csi-manila-nfs.
The deployment consists of two main components: Controller and Node plugins along with their respective RBACs. Controller plugin is deployed as a StatefulSet and runs CSI Manila, external-provisioner and external-snapshotter. Node plugin is deployed as a DaemonSet and runs CSI Manila and csi-node-driver-registrar.
Note: Snapshotting feature now requires a separate snapshot controller and CRDs to be deployed in the cluster. Please follow the related official installation instructions for external-snapshotter before continuing.
Deploying with Helm
This is the preferred way of deployment because it greatly simplifies the difficulties with managing multiple share protocols.
CSI Manila Helm chart is located in charts/manila-csi-plugin.
First, modify values.yaml to suite your environment, and then simply install the Helm chart with $ helm install ./charts/manila-csi-plugin.
Note that the release name generated by helm install may not be suitable due to their length. The chart generates object names with the release name included in them, which may cause the names to exceed 63 characters and result in chart installation failure. You may use --name flag to set the release name manually. See helm installation docs for more info. Alternatively, you may also use nameOverride or fullnameOverride variables in values.yaml to override the respective names.
Manual deployment
All Kubernetes YAML manifests are located in manifests/manila-csi-plugin.
First, deploy the RBACs:
kubectl create -f csi-controllerplugin-rbac.yaml
kubectl create -f csi-nodeplugin-rbac.yaml
Deploy the CSIDriver CRD:
kubectl create -f csidriver.yaml
Before continuing, csi-controllerplugin.yaml and csi-nodeplugin.yaml manifests need to be modified: fwd-plugin-dir needs to point to the correct path containing the .sock file of the other CSI driver.
Next, MANILA_SHARE_PROTO and FWD_CSI_ENDPOINT environment variables must be set. Consult --share-protocol-selector and --fwdendpoint command line flags for valid values.
Finally, deploy Controller and Node plugins:
kubectl create -f csi-controllerplugin.yaml
kubectl create -f csi-nodeplugin.yaml
Successful deployment should look similar to this:
$ kubectl get all
NAME READY STATUS RESTARTS AGE
pod/openstack-manila-csi-controllerplugin-0 3/3 Running 0 2m8s
pod/openstack-manila-csi-nodeplugin-ln2xk 2/2 Running 0 2m2s
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
service/openstack-manila-csi-controllerplugin ClusterIP 10.102.188.171 <none> 12345/TCP 2m8s
NAME DESIRED CURRENT READY UP-TO-DATE AVAILABLE NODE SELECTOR AGE
daemonset.apps/openstack-manila-csi-nodeplugin 1 1 1 1 1 <none> 2m2s
NAME READY AGE
statefulset.apps/openstack-manila-csi-controllerplugin 1/1 2m8s
...
To test the deployment further, see examples/csi-manila-plugin.
If you're deploying CSI Manila with Helm:
- Set
csimanila.topologyAwarenessEnabledtotrue - Set
csimanila.nodeAZ. This value will be sourced into the--nodeazcmd flag. Bash expressions are also allowed.
If you're deploying CSI Manila manually:
- Run the external-provisioner with
--feature-gates=Topology=truecmd flag. - Run CSI Manila with
--with-topologyand set--nodeazto node's availability zone. For Nova, the zone may be retrieved via the Metadata service like so:--nodeaz=$(curl http://169.254.169.254/openstack/latest/meta_data.json | jq -r .availability_zone)
See examples/csi-manila-plugin/nfs/topology-aware for examples on defining topology constraints.
The table below shows Manila share protocols currently supported by CSI Manila and their corresponding CSI Node Plugins which must be deployed alongside CSI Manila.
| Manila share protocol | CSI Node Plugin |
|---|---|
CEPHFS |
CSI CephFS : v1.0.0 |
NFS |
CSI NFS : v1.0.0 |
If you'd like to contribute to CSI Manila, check out docs/manila-csi-plugin/developers-csi-manila.md to get you started.