Path1.Mod1.c - Explore ML Workspace - Environment Setups Flashcards
Single Environment Workspace - Use Case
When you have research-centered scenarios with no need to release artifacts, OR when you’re only deploying inference services across environments ie ONE workspace to accommodate more than one environment
RAF
Single Environment Workspace - Pros
Reduced Azure Footprint; minimal management overhead (unlike multi-environment workspaces)
RBACs Access
Single Environment Workspace - Cons
- Dev vs Prod workflows have different RBACs respectively == environments could be too loose or rigid
- Datastore access considerations. Ex: excessive access to prod data == compromised data quality
“clear…”
Multi-Environment Workspace - Use Case
When you need regulated workspaces with clear separation of duties between environments, and you have users with resource access to those environments. Basically dedicated Workspace Resources per Environment.
Three: Staged Security Training
Multi-Environment Workspace - Pros
- Staged rollouts of workflows and artifacts
- Enhanced security and resource control over downstream environments
- Training scenarios on production data (via user access)
Three: DevOps Risk Data
Multi-Environment Workspace - Cons
- Requires Azure DevOps knowledge and ML engineering expertise
- Increased risk with management and process overhead (ex. giving access to Prod to investigate issues)
- Data mgmt + dev effort to make prod data available for training
One Env w/ Limited Data Access,
One Env w/ Production Data Access setup - Use Case
When you need to segregate dev and prod enviroments (DUR)!
Two: Separate Mgmt
One Env w/ Limited Data Access,
One Env w/ Production Data Access setup - Pros
- Clear separation of both duties AND access between dev and prod
- Lower resource mgmt overhead vs a multi-env deployment
Two: Dep Data
One Env w/ Limited Data Access,
One Env w/ Production Data Access setup - Cons
- Defining a deployment and rollout process for artifacts across workspaces
- Requires data mgmt and dev effort to make prod data available for training in dev