Best practices for Deploy
1. Mention clear naming conventions and descriptions for teammates to understand.
Deployments are hot live endpoints that are critical for your business needs. Every endpoint should be clearly described, and its purpose explained. Unlike Jobs, which have IDs, we assign random names to the deployments.
2. Use open-source tools for evaluating and monitoring performance
The performance and metrics of your system can be monitored directly from your code using open-source tools that can be integrated into your forward method on nbox.
3. You can configure rules for auto-scaling.
The autoscaling feature lets you spin up new pods on the fly if your request traffic is high. If you are using a very basic server like FastAPI, you might want to check out Django, which is more effective at utilizing the hardware resources.
4. Analyze the model states to determine what actions should be taken
A deployment could fail for a variety of reasons. Each phase of deployment generates several logs. To determine the exact cause of deployment failures in the NimbleBox workspace, check model logs. Select the ... icon on the model that has failed and click on view model logs to view the logs.