Administering

This section is dedicated to topics in administering the RocketML Enterprise application

RocketML comes with admin functionalities. Access to these functionalities requires separate login credentials. Please check with your IT department for login url and access.

Admins can do the following tasks:

1. Software Updates

2. Add, Manage, Monitor users

   • Note: Adding new users is not part of the current functionality

3. Add, update, monitor Data Sources

4. Add, update, manage, monitor Cloud Compute instances

5. Enable and disable Cloud AI services

6. Update Cloud configuration entries

1. Software Update

There are three types of Software upgrades:

A) RocketML regularly updates its application for enhancements, bug fixes etc. These updates happen automatically, in the background. RocketML may issue release notes for major updates.

B) RocketML periodically updates AMI (Amazon Machine Image for AWS) and VHD (Virtual Hard Disk for Microsoft Azure) that includes new or updated software packages, OS security patches. Software package updates are initiated by RocketML based on changes in the bundled open source softwares within an AMI or VHS. These AMI/VHS updates are specific to each customer instance. RocketML will update AMI/VHS in the background. When an update is made, RocketML will inform Admins about the changes. Users are required to "reset" their projects once to benefit from these updates.

Users are required to "reset" their existing projectVMs. Although not mandatory, it is strongly recommended. When resetting, make sure long run jobs are stopped. Resetting will shut down the compute, apply patches/updates and restart. There will not be any loss of data, code files.

C) Occasionally customer's DevOps/IT departments would like to change security policies and add/delete access of cloud resources by users. These updates are implemented via CFTs (AWS) and ARM templates (Azure). It requires a series of interactions between RocketML developers and Client DevOps including screen share sessions. Specific admin actions/tasks are dependent on each update.

2. Users

Adding New Users

To add new users, admins must make a support request with an email address of the user. The feature for allowing admins to add users without a support request is on RocketML's roadmap.

Start and Stop user created projectVMs

Admins can see all the projectVM's running at any point in time. They can turn off any projectVM by clicking on the toggle switch. This feature is given as a convenience if and when a user forgets to turn off expensive cloud compute instances.

Before turning off any projectVM make sure to check with the appropriate user if it is ok to turn them.

Check on cloud compute costs and usage by user

Admins can click "configure" link on any specific user and check their consumption by clicking on "cost and usage" tab

Configuring resources available for a specific user

Admins can configure user specific resources like budget (free credit), total number of projectVMs allowed, total number of models allowed, total number of cloud instances (like EC2) running simultaneously and storage limit.

We recommend Admins to understand their company's allowed capacity limits for both compute as well as storage when configuring these settings. Typically admins maintain a spreadsheet to map out their capacity and needs in order to allowcate the resources amongsts various needs and users.

3. Data Sources

TL;DR --> Simple Steps to Add a Data Source

• Gather information from CorpIT or respective stakeholders

• Confirm that project VMs can access buckets (check IAM policies) and databases (check VPC/subnet network setup)

• Enter Information into appropriate screens on Admin app, follow on-screen instructions

Add Databases

Admins can add new connection to Databases (structured data) like any Oracle, Mysql, Snowflake or Mariadb etc. These data sources are read-only data for users. In addition, users can always upload their own data using file upload/import functionality provided within Jupyter environment. Typically any database that can be accessed from the VPC and Subnet in which projectVMs are supported. If users and admins need connection to additional data sources, please contact us via a support request.

To add, update, enable databases, admins will need precise information as per the form below

Add, update, enable Databases as Data Source

Add unstructured data source (or files). Eg., AWS S3 or Azure Blob storage

Any source (buckets) that is in the same AWS/Azure account as project resources can be added. Image below shows screen to add an S3 bucket. ProjectVMs will get read-only access to these buckets via an IAM role

Users can also access buckets in other AWS accounts with cross-account permission at the root level.

Optionally, admins can select specific users for access to these buckets.

Nuances 😉

Disabling a Data source will

• Prevents use of Data Source in NEW projectVMs ONLY. Users will stop seeing the data source on their UI

• However, existing projectVMs will continue to have access to the data source, unless the native data source is deleted

After each modification to a data source (add, update, enable, disable), ask your users to test the connections!

4. Compute instances

Overview

• AWS EC2 compute instances are the underlying DSW cloud compute infrastructure. They are charged pay-for-use basis by AWS directly to companies.

• DSW is optimized for certain instance types; Admin app shows the list of instances enabled for access. Each Instance type comes with a limit from AWS. This information is critical for smooth administration of DSW Compute

• Attempts to use instances beyond limits by data scientists will result in failure of workspaces

• To have the instance type limits increased, Corp IT will have to make support request to AWS

Enter & update Instance Limit information

1. AWS EC2 instance limits are set by AWS. Admins need to get instance limit information targeted towards data science workloads from Corp IT and enter this information into Admin App

2. Admins can update limits on Admin App

Monitor Instance type usage against limits

• As data scientists utilize instances, Admin dashboard is updated to reflect the usage

• Admins should request Corp IT to increase the limits if a particular instance type usage reaching limits

Elastic IP Addresses

• Make sure the AWS account in which the project VMs are running have as many Elastic IP addresses as the number of user projects.

Enable/Disable an Instance type

• Enables/Disables compute instance type for entire user base

• Disabling will

  • Allow existing projects to continue to make use of the compute instance type

  • Prevents use of compute instance type in NEW projects ONLY

  • User will stop seeing the compute instance type on their DSW UI

Communication & Post Modification Actions

• If an instance is disable or enabled, Admins need to inform Data Scientists

• If an instance type limits are being reached, data scientists will not see that instance in the master or child node actions related to project VMs

• Contact CorpIT for instance limit increase as needed

5. AI services

Five AWS AI Services are supported based on Data Scientists requests & Corp IT security clearance. Admin responsibilities are straightforward and simple

1. Enable/Disable Service type

   • Enables/Disables AWS Services for entire user base

   • Disabling will

     • Allow existing projects to continue to make use of the AI service

     • Prevents use of AWS Service in NEW projects ONLY

     • User will stop seeing the AI Service type on their DSW UI

2. Post Modification Actions

   • Admins need to inform Data Scientists if a service is enabled/disabled

6. Cloud Configuration Entries