SREday 2023

14 - 15 Sep London, UK
Site Reliability Engineering Conference

The efficient way of Autoscaling Your Workload in Kubernetes

Henrik Rexed

Struggling with limited observability data for HPA? Tired of provisioning resources that lead to unnecessary costs? The Keptn's Metric Server solves this by exposing new metrics in K8s. Join me to see how to scale workloads with efficient HPA rules.

Have you ever faced the challenge of limited accessibility to observability data for Horizontal Pod Autoscaler (HPA) or are you tired of provisioning resources that lead to unnecessary costs? As you likely know, HPA leverages metrics exposed by the Kubernetes Metrics API for scaling decisions. Consequently, many organizations typically construct their autoscaling rules based on fundamental metrics like CPU and memory usage. In Kubernetes, creating an autoscaling rule that guarantees both the reliability of applications and the cost efficiency only based on CPU usage can be challenging. As organizations expand their Kubernetes footprint, they often utilize more than one observability tool or switch tools altogether. How can we effectively utilize our observability data in autoscaling rules? The Keptn project offers the Metric Server, which addresses this challenge by seamlessly consolidating all observability data within Kubernetes. Join us for an informative presentation where we will demonstrate how the Keptn Metrics Server effortlessly scrapes metrics, facilitating the autoscaling of Kubernetes workloads in conjunction with the HorizontalPodAutoscaler. Through a real example showcasing inefficient HPA rules, we will explain how the Keptn Metric Server can enhance and optimize the autoscaling process utilizing metrics from Istio/Envoy, but also the cost of your workload. Discover the value of the Keptn Metrics Operator, which serves as a unified entry point for metrics data, regardless of its source. This operator streamlines the autoscaling process for your workloads. Come and learn the best practices to define the right Autoscaling policies

Henrik is a Cloud Native Advocate at Dynatrace. Prior to Dynatrace, Henrik has worked more than 15 years as Performance engineer .. Henrik Rexed Is Also one of the Organizer of the KCD Austria, the Conference named WOPR and the owner of the Youtube Channel IsitObservable.

Sponsors & Partners

Want to become a sponsor? Get in touch!