I am deploying my RASA (v2) to Kubernetes. And i am using multiple pods for traffic issues. Now when i want to select specific model to be load it is loaded only to catched POD. Other pods are still using old model. Does RASA has a solution for that? I couldnt see from code.
Yes, there are different approaches you can take to update the model. The most common I’ve seen is to use the rasa run option to pull the model from block storage like S3.
When there’s a new model, you either replace the model in block storage and use k8s rolling updates to restart the pods or you use the Rasa http API to tell each instance to re-load the model.