Kubernetes right-sizing at the container level for fine-tuned application efficiency

Spot by NetApp’s Ocean continuously optimizes Kubernetes clusters with a wide feature-set tackling different aspects of running and managing Kubernetes containers in a cloud environment. 

One such aspect are the container resource requests defined in the cluster (upon which Ocean intelligently bin-packs pods on the underlying cloud VMs). Incorrect assumptions regarding the CPU and Memory required for an application, can incur unnecessary and costly cloud infrastructure waste.

To address this issue, Ocean provides right sizing recommendations based on the actual resource utilization of any running applications, over time. 

Until now recommendations were generated per Kubernetes workload without distinguishing between the different applications defined to run within the workload.

In addition, clusters running a large amount of workloads could get hundreds of recommendations making it difficult to manage and follow.

Today we’re happy to share two improvements added to Ocean’s right-sizing API.

  • Recommendations are now provided by the container. This means, for example, that a pod running 2 containers will get up to two different, independent recommendations, focusing the application owner on the actual application(s) that can benefit from setting different resource requests.
  • A new filtering mechanism has been added so different teams can easily focus only on the recommendations relevant for them. The mechanism is based on Kubernetes native concepts like labels, annotations and namespaces, allowing users to filter the recommendations according to the existing workloads configuration.

For more details about the new API please refer to the relevant Spot documentation for Ocean kubernetes AWS and Ocean GKE.