1. Serve SVM and Decision tree models using the flask on separate relative URLs. i.e. localhost:8000/svm_predict and localhost:8000/decision_tree_predict. (ip could be different than localhost, in your case)
2. Dockerize the deployment i.e. create dockerfile and build image such that when you do docker run (may be with some more flags), the above two links should be accessible via curl. Write docker_example.sh shell script that includes the full curl commands.
Client:
Server:
3. Deploy on Google Kubernetes Engine
The deployment.yaml file was used as config file for deployment.
Tasks performed during deployment:
- Authenticate Docker on cloud shell.
- Push docker image of the web application to Artifact Registry.
- Create a standard cluster of 3 nodes.
- Authorize cluster to create its entry for in kubeconfig.
- Deploy app image using deployment.yaml and check status.
- Expose deployment on Kubernetes service to public.
- Make requests to running deployment using curl API calls on the EXTERNAL-IP provided by the service.
- Perform manual scaling.
- Create a HorizontalPodAutoScaler for auto-scaling.
- Resize cluster size by changing number of nodes.
- Update app, create image and set the deployment with the new image (version).
- Check background workings like forming of ReplicaSets and termination of Pods.
The results and experiments performed during deployment can be found in kubernetes/deployment_results.txt
Specifically:
Accessible from external-IP:
