Operator Dashboard & Workflow

The Operator is responsible for:

  • Managing the DKube cluster

  • Managing Pools, Groups, & Users

Operator Menu

The Operator screen provides a GUI-based mechanism to navigate through the workflow.

_images/Operator_Menu_R21.png

Menu Item

Function

Dashboard

Overview of the current status of the cluster

OAuth

Activate/update backend authorization

Pools

Create, view, modify, & delete device Pools

Groups

Create, view, modify, & delete User Groups

Users

On-board, view, and delete Users

DVS

Maintain version control system

Configuration

Configure log server (currently not active)

Nodes

View status of cluster nodes

Devices

View status of cluster devices

Storage

View storage statistics for local Kubernetes cluster

Cluster

View services and pods on cluster

Operator Dashboard

_images/Operator_Dashboard_R21.png

The Operator dashboard provides a snapshot of the cluster. This enables the Operator to see the status and health of the cluster, and provides the information needed for proper load balancing.

Operator Workflow

_images/Operator_Block_Diagram.png

Pools, Groups, & Users are managed from their respective menu selections on the left.

There is no requirement for the Operator to allocate Pools, Groups, and Users for single User operation. At installation time, the Operator User is specified, and that User is authenticated and assigned to the Default Group, which is assigned the Default Pool. The Default Pool contains all of the Devices recognized in the cluster. As such, the single Operator/Data Scientist User has access to all of the GPUs while logged in as a Data Scientist.

If this scenario is true in your situation, you can skip ahead to Data Scientist Dashboard & Workflow

If more than one User, Group, or Pool is required, then the following sections apply. It is recommended that the following flow is used to add a Pool, Group, or User.

  • Create a new Pool and assign Devices to the Pool

  • Create a new Group and assign a Pool to the new Group

  • On-board new Users, and assign them to a Group

Pool Management

_images/Operator_Pool_Management.png

The Pool Management screen allows the Operator to view and manage Pools. Initially, all of the GPUs on the system are assigned to the Default Pool. The GPUs in the Pool Management screen are shown for the entire cluster.

Create Pool


_images/Operator_Create_Pool_Popup.png
  • Select the “+ Pool” icon in the top right-hand part of the Pools screen

  • The Create Pool popup will appear

  • Input the name of the Pool

  • Assign the Devices to the Pool

  • Select the Create button

Note

Assigning a Device to a Pool will remove it from the Default Pool

Note

Only a single type of GPU can be assigned to a Pool

Edit Pool

  • Select the Pool to be edited

  • The Edit Pool popup will appear

  • The GPUs assigned to the Pool can be modified

  • When complete, select the Save button

Delete Pool

  • Select the Pool to be deleted from the left-hand checkbox

  • Select the delete icon from the top right-hand side of the section

  • A confirmation box will pop up

  • Accept the confirmation, and the Pool will be deleted

  • GPUs assigned to the deleted Pool will be available for selection by other Pools

Note

If any GPU in the Pool has an active job running, it cannot be deleted

Group Management

_images/Operator_Group_Management.png

The Group management screen shows information about the current Groups, and allows Groups to be created, modified, and deleted.

Create Group

_images/Operator_Group_Create_Popup.png
  • Select “+ Group” icon from the top right-hand side of the Groups screen

  • The Group creation popup will appear

  • Input the name of the Group

  • Select the Users that will be part of the Group, and the Pool that the Group will be associated with

  • Only Users & Pools that are not part of another Group will appear at this time

  • Multiple Users can be assigned, but only a single Pool

  • Select the Create button

Note

Only free Users (Users not already assigned to another Group) can be assigned to a Group. If Users are already assigned to another Group, including the Default Group, they must first be deassigned from that Group, then assigned to the new Group as described above.

Edit Group

_images/Operator_Group_Edit_Popup.png
  • Select the Group to be edited from the left-hand checkbox

  • Only a single Group can be edited at one time

  • Select the Edit icon at the top right

  • The Edit Group popup will appear

  • Users and Pools can be assigned or removed

  • Only available Pools and Users will be shown in the edit box

  • To move a User between Groups, the User must first be removed from its current Group, then assigned to the new Group

  • Select the Save button

Delete Group

  • Remove all Users & Pools from the Group

  • Select the Group to be deleted from the left-hand checkbox

  • Select the delete icon from the top right-hand side of the section

  • A confirmation box will pop up

  • Accept the confirmation, and the Group will be deleted

    • Users assigned to the Group will be available for selection by other Groups

Note

If the Group has Users or Pools assigned, it cannot be deleted

User Management

_images/Operator_User_Management.png

The User Management screen allows Users to be brought on board and assigned to Groups, as well as deleted from DKube.

Add (On-Board) User

_images/Operator_User_Popup.jpg
  • Select the “+ Onboard User” icon in the top right-hand part of the User screen

  • The On-Board User popup will appear

  • The Users that need to be added are selected by the checkboxes, and they can be added to one of the groups by selecting it on the right

    • By default, the Default Group will be selected in the new User popup. If the User is going to be assigned to another Group later (a Group that does not yet exist), it is better to uncheck the Default Group assignment in this step. The User will then show up as available when the new Group is created.

  • When the selection has been made, select “On Board”

  • Once the User has been added, the only other supported capability is to delete the user

    • To change which Group a User belongs to, the edit Group function must be used

Note

If the list of Users does not appear, refresh the User list at the top of the On-Board User popup. This can happen when the GitHub authorization has been activated or changed.

Add Users from an LDAP Group

If the authentication mechanism is through LDAP, it is possible to add users through their LDAP Group. If this is enabled, the top right button labeled ” + On Board User By Group” button is active.

_images/Operator_LDAP_Group.png

Selecting the button brings up a popup that shows the LDAP groups and the users within each group. All of the users can be added at once, or specific users from that group can be added manually.

_images/Operator_LDAP_Group_Popup.png

Note

Onboarding users through an LDAP group also creates a DKube group with the same name, and adds the onboarded users to that DKube group

Select User Roles

A User can have any combination of the following roles:

Role

Function

Operator (OP)

Manage the cluster, users, and resources

Data Scientist (DS)

Create, optimize, and deploy models

Different Roles have access to different screens and capabilities, based on the workflow described in more detail at DKube Roles & Workflow

The first user that is on-boarded, either after installation (as a local user) or when a backend authorization is chosen (described in Authentication Credentials ), is an Operator. The Operator has access to all of the screens and capabilties.

User roles can be assigned from the “Users” Menu by selecting the “Role” Button.

_images/Operator_Role_Select.png

A popup provides a set of check boxes for role assignment.

Move User Between Groups

At the initial on-board time, a User can be assigned to a specific Group, or not assigned to any Group (for later assignment). After that, a User can be moved from one Group to another by editing the Group.

  • The User must first be removed from its current Group

  • The target Group can then add the User

Delete User

  • Select the User to be deleted from the left-hand checkbox

  • Multiple Users can be selected

  • Select the delete icon on the top right part of the section

  • A confirmation box will pop up

  • Accept the confirmation, and the User will be deleted from the DKube environment

Note

If the User has an active job running, it cannot be deleted

DVS

_images/DVS_Version_Diagram.png

Versioning within DKube is provided through DVS, as described in the section Versioning . In order to use the version control system, the DVS storage locations must first be configured through the DVS Operator screen. The DVS targets are referenced when the Repo is created, as described in the section Repos

_images/Operator_DVS.png

The DVS screen shows the existing DVS targets. A new target can be created by selecing the “+ DVS” button. There are 2 storage areas to configure:

  • The metadata source

  • The storage source

_images/Operator_DVS_New.png

Metadata Source

The metadata, including the version information, can be stored in several places:

  • Within a DKube folder on the cluster

  • In a Git repository

_images/Operator_DVS_New_Git.png

If a Git repository is the target, the following information is required.

Field

Value

Repo

url for the Git repo

Auth Token

GitHub access token for the repo (see Create GitHub Token)

Storage Source

The data can be stored in the following places:

  • Within a DKube folder on the cluster

  • AWS S3 or a Minio server

  • Google Cloud Storage

_images/Operator_DVS_New_S3.png

GCS

Field

Value

Bucket

Bucket name

Secret Key

JSON file containing the GCP private key

Cluster Management

The cluster management function of the Operator is organized through several screens.

Cluster Credentials

Cluster access is managed from the “OAuth” menu selection on the left hand side of the screen.

_images/Operator_OAuth_Active.png

After installation, the DKube credentials are handled through local authorization. The figure shows a DKube authorized by GitHub. The OAuth menu has 2 functions:

  • Activate the backend authorization for general DKube use

  • Update the backend credentials when required

This is discussed in more detail in section Authentication Credentials

DKube Status Information

The rest of the cluster management functions are accessed from the right hand menu.

_images/Operator_Right_Menu_Open_R21.png

Status information about the DKube installation is available from the “About” menu item.

_images/Operator_Right_Menu_About.png

The “About” popup provides information on the DKube installation version, and on the status of the license. If the license shows that it is running out soon, contact your cluster administrator to update.



_images/Operator_About_Popup.png

DKube License Update

The Dkube license can be updated by selecting “License”.

_images/Operator_Right_Menu_License.png

This will provide a popup that shows the current license information, and allows a new license token to be pasted in.



_images/Operator_Renew_License_Popup.png

DKube OAuth Token

DKube allows access through the use of an OAuth token. This is used when access is required to the DKube cluster. The token is provided through the “Developer Settings” menu item.

_images/Operator_Right_Menu_Developer.png




_images/Operator_Developer_Settings_Popup.png

Node, Device, & Cluster Status

The nodes on the cluster can be viewed from the “Nodes” menu selection on the left hand of the screen. Selecting a node provides a list of the devices on each node, their status & utilization.

_images/Operator_Nodes_R21.png

Details on each Node in the cluster can be viewed through the “Devices” menu selection. This will show the cluster-wide utilization of the device, as well as its health.

_images/Operator_Devices_R21.png