Breyta

Deila með


Workload profiles in Azure Container Apps

A workload profile determines the amount of compute and memory resources available to the container apps deployed in an environment.

Profiles are configured to fit the different needs of your applications.

Profile type Description Potential use
Consumption Automatically added to any new environment. Apps that don't require specific hardware requirements
Dedicated (General purpose) Balance of memory and compute resources Apps that require larger amounts of CPU and/or memory
Dedicated (Memory optimized) Increased memory resources Apps that need access to large in-memory data, in-memory machine learning models, or other high memory requirements
Dedicated (GPU enabled) (preview) GPU enabled with increased memory and compute resources available in West US 3 and North Europe regions. Apps that require GPU

Note

When using GPU-enabled workload profiles, make sure your application is running the latest version of CUDA.

The Consumption workload profile is the default profile added to every Workload profiles environment type. You can add Dedicated workload profiles to your environment when you create it or after you create it. Workload profiles environments are deployed separately from Consumption only environments.

For each Dedicated workload profile in your environment, you can:

  • Select the type and size
  • Deploy multiple apps into the profile
  • Use autoscaling to add and remove instances based on the needs of the apps
  • Limit scaling of the profile to better control costs

You can configure each of your apps to run on any of the workload profiles defined in your Container Apps environment. This configuration is ideal for deploying microservices where each app can run on the appropriate compute infrastructure.

Note

You can only apply a GPU workload profile to an environment as the environment is created.

Profile types

There are different types and sizes of workload profiles available by region. By default, each Dedicated plan includes a consumption profile, but you can also add any of the following profiles:

Display name Name vCPU Memory (GiB) GPU Category Allocation
Consumption Consumption 4 8 - Consumption per replica
Dedicated-D4 D4 4 16 - General purpose per node
Dedicated-D8 D8 8 32 - General purpose per node
Dedicated-D16 D16 16 64 - General purpose per node
Dedicated-D32 D32 32 128 - General purpose per node
Dedicated-E4 E4 4 32 - Memory optimized per node
Dedicated-E8 E8 8 64 - Memory optimized per node
Dedicated-E16 E16 16 128 - Memory optimized per node
Dedicated-E32 E32 32 256 - Memory optimized per node
Dedicated-NC24-A100 (preview) NC24-A100 24 220 1 GPU enabled per node*
Dedicated-NC48-A100 (preview) NC48-A100 48 440 2 GPU enabled per node*
Dedicated-NC96-A100 (preview) NC96-A100 96 880 4 GPU enabled per node*

* Capacity is allocated on a per-case basis. Submit a support ticket to request the capacity amount required for your application.

Note

The command az containerapp env workload-profile set is no longer available for selecting a workload profile. Instead, use az containerapp env workload-profile add or az containerapp env workload-profile update.

In addition to different core and memory sizes, workload profiles also have varying image size limits available. To learn more about the image size limits for your container apps, see hardware reference.

The availability of different workload profiles varies by region.

Resource consumption

You can constrain the memory and CPU usage of each app inside a workload profile, and you can run multiple apps inside a single instance of a workload profile. However, the total resources available to a container app are less than the resources allocated to a profile. The difference between allocated and available resources is the amount reserved by the Container Apps runtime.

Scaling

When demand for new apps or more replicas of an existing app exceeds the profile's current resources, profile instances might be added.

At the same time, if the number of required replicas goes down, profile instances might be removed. You have control over the constraints on the minimum and maximum number of profile instances.

Azure calculates billing largely based on the number of running profile instances.

Networking

When you use the workload profile environment, extra networking features that fully secure your ingress and egress networking traffic (such as user defined routes) are available. To learn more about what networking features are supported, see Networking in Azure Container Apps environment. For steps on how to secure your network with Container Apps, see the lock down your Container App environment section.

Next steps