Beyond Cluster Autoscaler with Spotinst

18:00 - 18:30: Gathering, food and drinks

18:30 - 19:10: Autoscaling Kubernetes Clusters
Our managed Kubernetes service can reduce costs and increase resource utilization thanks to our homegrown custom cluster autoscaler. In this session, we will discuss the design considerations for scaling Kubernetes clusters, such as pod resource requirements (CPU, MEM, GPU, Custom resources) and pod dependencies (Volume Claims, Affinity, Node-selectors, Tolerations). We’ll also present our custom cluster autoscaler that implements these principles.

19:20 - 19:45: More than Autoscaling
In this session we will introduce additional considerations and complementary concepts to cluster autoscaling:

• Pod Sizing: How to use and identify the correct pod resource requests and limits for your application, how to identify misconfigurations, and how to utilize the Virtual Pod Autoscaler. We will also share our learnings based on real-world data that we collected and analyzed.
• Running immediate workloads: scaling the cluster is not always the best solution to accommodate a pending pod. Pod priority, and managed headroom are viable alternatives to consider.
  (In this part we would also like to open the discussion to hear from the audience)

19:45 - 20:00: Ocean - Spotinst Managed Kubernetes Service
Wrapping up the meetup with a demonstration of Ocean, our managed Kubernetes service that implements some of the concepts discussed.

About the speakers:
Ori Yemini, Principal Engineer, Spotinst
Aviv Shukron, Product Team Lead, Spotinst
Aharon Twizer, CTO, Spotinst