Deployment on Azure AKS¶
You can use the API Deployer Kubernetes integration to deploy your API Services on Azure Kubernetes Service (AKS).
Setup¶
Create your ACR registry¶
Follow the Azure documentation on how to create your ACR registry. We recommend that you pay extra attention to the pricing plan since it is directly related to the registry storage capacity.
Create your AKS cluster¶
Follow Azure’s documentation on how to create your AKS cluster. We recommend that you allocate at least 8GB of memory for each cluster node.
Once the cluster is created, you must modify its IAM credentials to grant it access to ACR (Kubernetes secret mode is not supported). This is required for the worker nodes to pull images from the registry.
Prepare your local az
, docker
and kubectl
commands¶
Follow the Azure documentation to make sure that:
Your local (on the DSS machine)
az
command is properly logged in. As of October 2019, this implies running theaz login --service-principal --username client_d --password client_secret --tenant tenant_id
command. You must use a service principal that has sufficient IAM permissions to write to ACR and full control on AKS.Your local (on the DSS machine)
docker
command can successfully push images to the ACR repository. As of October 2019, this implies running theaz acr login --name your-registry-name
.Your local (on the DSS machine)
kubectl
command can interact with the cluster. As of October 2019, this implies running theaz aks get-credentials --resource-group your-rg --name your-cluster-name
command.
Setup the infrastructure¶
Follow the usual setup steps as indicated in Setting up. In particular, to set up the image registry, in the API Deployer go to Infrastructures > your-infrastructure > Settings, and in the “Kubernetes cluster” section, set the “Registry host” field to your-registry-name.azurecr.io
.
Deploy¶
You’re now ready to deploy your API Services to Azure AKS.
Using GPUs¶
Azure provides GPU-enabled instances with NVidia GPUs. Several steps are required in order to use them for API Deployer deployments.
Building an image with CUDA support¶
The base image that is built by default (see Setting up) does not have CUDA support and cannot use NVidia GPUs.
You need to build a CUDA-enabled base image. To enable CUDA add the --with-cuda
option to the command line:
./bin/dssadmin build-base-image --type apideployer --with-cuda
We recommend that you give this image a specific tag using the --tag
option and keep the default base image “pristine”. We also recommend that you add the DSS version number in the image tag.
./bin/dssadmin build-base-image --type apideployer --with-cuda --tag dataiku-apideployer-base-cuda:X.Y.Z
where X.Y.Z is your DSS version number
Note
This image contains CUDA 10.0 and CuDNN 7.6. You can use
--cuda-version X.Y
to specify another DSS-provided version (9.0, 10.0, 10.1, 10.2, 11.0 and 11.2 are available). If you require other CUDA versions, you would have to create a custom image.Remember that depending on which CUDA version you build the base image (by default 10.0) you will need to use the corresponding tensorflow version.
Warning
After each upgrade of DSS, you must rebuild all base images and update code envs.
If you used a specific tag, go to the infrastructure settings, and in the “Base image tag” field, enter dataiku-apideployer-base-cuda:X.Y.Z
Create a cluster with GPUs¶
Follow Azure documentation for how to create a cluster with GPU.
Add a custom reservation¶
In order for your API Deployer deployments to be located on nodes with GPU devices, and for AKS to configure the CUDA driver on your containers, the corresponding AKS pods must be created with a custom “limit” (in Kubernetes parlance) to indicate that you need a specific type of resource (standard resource types are CPU and memory).
You can configure this limit either at the infrastructure level (all deployments on this infrastructure will use GPUs) or at the deployment level.
At the infrastructure level¶
Go to Infrastructure > Settings
Go to “Sizing and Scaling”
In the “Custom limits” section, add a new entry with key
nvidia.com/gpu
and value:1
(to request 1 GPU)Don’t forget to add the new entry, and save settings
At the deployment level¶
Go to Deployment > Settings
Go to “Sizing and Scaling”
Enable override of infrastructure settings in the “Container limits” section
In the “Custom limits” section, add a new entry with key
nvidia.com/gpu
and value:1
(to request 1 GPU)Don’t forget to add the new entry, and save settings
Deploy¶
You can now deploy your GPU-requiring deployments.
This applies to:
Python functions (your endpoint needs to use a code environment that includes a CUDA-using package like tensorflow-gpu)
Python predictions