When spawning a user's notebook pod, spawner mounts various type of persistent volumes. They includes
- User volume
- Group volume (Project volume)
- Datasets volume
A user volume stores users' own data. When a user spawn jupyter server for the first time, the user's PVC is created.
The capacity is determined in the order
- Use user's specified
- If this setting is not defined, use the max
userVolumeCapacityof all the groups the user belongs to.
- If none of the group set the
userVolumeCapacity, use the system's
The storage class is defined in the helm value.
Group volume (Project volume)
A group volume (project volume) stores the data shared by a group. To enable the group volume, the administrator should enable the
Shared Volume in the group setting. And the volume is created if it doesn't exist at the very firt time of the mount to the group volume.
The storage class is defined in the helm value
jupyterhub: custom: groupVolumeStorageClass: "<rwx-storage-class>"
But in most cases, we use a NFS pvc provisioned by
primehub-groupvolume-nfs. The configuration would look like this
jupyterhub: custom: groupVolumeStorageClass: "" groupVolumeAnnotations: primehub-group-sc: standard
- the group volume is manually provisioned
primehub-groupvolume-nfscontroller would create a NFS server backed by a RWO pvc
- the RWO pvc uses "standard" storage class.
If you didn't specify primehub-group-sc's value ("standard" here) in yaml, it will be set by
Currently, there are these types of dataset would be mounted as persistence volume
- pv (hostpath)
If dataset is type of
pv, the dataset is backed by a pvc. The difference comparing to group volume is
- A dataset can be connected to multiple groups; while group volume only belongs to a group
- The connection from dataset to a group can be read-only or read-write
- Can enable a upload server
Under the hood, the storage class is defined in the helm value
primehub-console: graphql: primehubGroupSC: standard
It will create a NFS server backed by a RWO pvc which is similar to group volume.
If you didn't specify primehubGroupSC's value ("standard" here) in yaml, it will be set by
metadata.namein the CRD and the
spec.volumeNamein the CRD.
A special hack of
pv is set the
volumeName with prefix
hostpath:. In this way, the dataset is not backed by a pvc, instead, the volume is a hostpath volume
|hostpath||path after the |
hostpathis not allowed to edit in the admin dashboard in the future.
If the dataset is type of
git, the dataset is backed by a hostpath and periodically pull the data from a repository. Under the hood, we use gitsync daemonset to sync the data to hostpath
|The mount path of the pv dataset.|
|Whether a symbolic link to the mountPath is required.|
|The host path to put the gitsync result.|
|The mount path of the git dataset|
|Mount only when the launching group connects this dataset|
|Only works in pv type dataset. Create a upload server.|
|Only works when upload server be created. Set http basic auth based on secret.|
Symbolic links in home folder
The home folder is mounted by the user volume. However, in order to locate the project volume and dataset easily, we create symbolic links in the home folder.
- Group Volumes:
- Dataset Volumes:
The symbolic links to a dataset can be removed by setting the annotation
Launch group only
By default, if a user have permissions to access the a group volume or a dataset, the spawner would mount it no matter which group the user select.
But if a group volume or a dataset is configured
Launch Group Only to
no. The volume is only mounted while the user select the group which connects to this volume.