scaling.md

Scale Application Service Adapter

Application Service Adapter gives you options for scaling components for performance or availability. Read this topic for scaling guidance.

Notes on system performance

Application Service Adapter runs on a Kubernetes cluster and uses the Kubernetes API to store and manage resources, so the performance of Application Service Adapter is very dependent on the underlying Kubernetes control plane. If you observe slow performance or timeouts when using Application Service Adapter, a good first step is to ensure that the control plane for your Kubernetes cluster is properly resourced.

For managed clusters that come from a public-cloud IaaS, you generally won't have direct control over how the control plane is provisioned, so you won't be able to scale up control plane resources. If you are managing your own Kubernetes clusters, you must ensure you have provisioned adequate capacity for the apiserver and etcd components. See the Kubernetes documentation as a starting place for understanding production cluster concerns.

Scaling the Application Service Adapter controller managers

Vertical scaling

Application Service Adapter deploys a set of controller managers responsible for reconciling various custom resources into running pods on your Kubernetes cluster. If these controller managers are resource constrained, scaling them up vertically may improve performance. Controller managers maintain an in-memory informer cache of the events that they are responsible for tracking, so it is common for controller managers on busy systems with many objects to require more memory.

Controller manager deployments in the tas-adapter-system namespace are:

• cartographer-builder-runner-controller-manager
• korifi-controllers-controller-manager

To determine whether your controller managers face CPU or memory pressure, you can check the Prometheus metrics for your controllers as described in the Metrics topic.

You can also inspect the running pods for these deployments using the kubectl top command.

If you determine that one or more of your controller managers is resource constrained, you can increase the available resources by setting the optional scaling parameters as described in the Configure installation settings instructions, and then re-running the installation step.

Horizontal scaling

As described in the Failover and redundancy topic, controller managers use leader election to designate a single active instance, so horizontal scaling is not an effective strategy to improve performance of a slow controller. Since the same deployments also serve webhook requests, you may choose to add horizontal scale to improve webhook performance. You may also choose to scale your deployments to improve controller availability. Multi-instance deployments allow rolling deployments so that controllers and webhooks don't become unresponsive during upgrades. Horizontal scaling can be achieved by using the replicas scaling configuration.

Scaling the Application Service Adapter API

If you have confirmed that the Kubernetes control plane and the controller managers both have sufficient resources, but you are still observing slow response times or timeouts from the Cloud Foundry API, scaling up the korifi-api-deployment in the tas-adapter-system namespace may help.

Unlike the controller managers, requests to the pods in this deployment are load balanced, so either horizontal or vertical scaling should improve performance. By default, korifi-api-deployment runs with two replicas, so it should already maintain availability during upgrades unless it is scaled down.

Similar to the controller managers, both vertical (memory and CPU) and horizontal (replicas) scaling can be achieved by changing the installation settings, and then re-running the installation step.