Final Week Updates

Day 1

Collected eviction probabilities from azure portal.

Created API using the data collected. Hosted it. It can be accessed from the following link.

The API can be accessed in the following format:

http://137.135.85.84/price/?operation=probability&size=Standard_DS1_v2

Day 2

Experiment average time to start a virtual machine from spot virtual machine. In this experiment I wrote a python script (code in the Github repository) to start a spot virtual machine. It requires azure cli to be logged in. With this experiment my goal was to check the feasibility of starting another virtual machine when eviction happens and we move into the 30 seconds timeline. With the help of script I conducted 3 experiments, all these experiments suggested that on an average it takes around 2 minutes to start a virtual machine from a virtual machine. Conclusion of the experiment: The 2 minutes required timeline and also the necessity for the cli to be installed inside VM renders the plan to handle eviction by trying to restart another spot virtual machine from the virtual machine useless. However an alternative approach would be to make an API server, that would receive eviction message sent by all spot virtual machines and spot node pools in AKS. This server intern would restart another instance.

Day 3 and Day 4

Document the code, make tutorials and setup guide for the code. Also made a video tutorial for the API server setup, with step by step instructions. Push all the documentation in the repository.

Day 5

Make a presentation. Present before my mentors. Update blog.