This session introduced an innovative solution for automating the publishing and deployment of SAS models to Azure. The proposed approach streamlines the path-to-production for SAS models by leveraging SAS Container Runtime (SCR) and Azure DevOps release pipelines. The solution ensures a smooth and efficient deployment process by using Azure Service Plans and Web Apps as versatile back-end applications for model scoring. The process begins with creating an Azure App Service Plan and a Web App, establishing the target environment. An Azure publishing destination is set up, followed by configuring a release pipeline in Azure DevOps. The SAS model is published from SAS Model Manager to the Azure publishing destination, packaged as an SCR container image. Subsequently, the release pipeline retrieves the container image and seamlessly deploys it to a designated slot within the Azure Web App. The SCR enables model scoring through a simple REST API call. By employing tags during publishing, the pipeline can deploy the model to the desired slot, allowing individual testing in different slots and enabling seamless production deployment through slot swapping, eliminating downtime. Azure Web Apps support auto-scaling and scaling based on predefined rules, accommodating spikes in scoring requests. Network traffic isolation can also be enforced, ensuring requests are limited to the virtual network's resources.
Presentation slides are attached to this post.
