reqserving-setup

Request Serving Setup Utility

This utility configures an OpenShift management cluster for HyperShift's dedicated request serving isolation. It is based on the setup logic used in the HyperShift e2e test suite.

Building

To build the utility, run make from this directory:

make

This will produce a reqserving-setup binary.

Running

The utility requires a configured KUBECONFIG with cluster-admin permissions on the management cluster.

Apply Configuration

To apply the configuration directly to the cluster:

./reqserving-setup

Dry Run

To see what resources would be created without applying them, use the --dry-run-dir flag:

./reqserving-setup --dry-run-dir ./manifests

What This Utility Does

The utility performs the following configuration steps:

1. Cluster Autoscaler: Configures the default ClusterAutoscaler resource to enable balanceSimilarNodeGroups and scale-down.
2. Machine Health Check: Creates a MachineHealthCheck named request-serving-mhc in the openshift-machine-api namespace. It monitors nodes labeled with hypershift.openshift.io/control-plane=true.
3. Common MachineSets: Creates "common" MachineSet resources in 3 availability zones for non-request-serving control plane workloads.
   • Taints: hypershift.openshift.io/control-plane=true:NoSchedule
   • Autoscaling: Creates a MachineAutoscaler for each, scaling from 0 to 20.
4. Request Serving MachineSets: Creates a comprehensive set of MachineSet resources across 2 availability zones for 3 different node sizes (small, medium, large).
   • Isolation: 20 pairs of MachineSets are created for each size and zone.
   • Labels: Nodes are labeled with hypershift.openshift.io/request-serving-component=true and hypershift.openshift.io/node-size=<size>.
   • Taints: Nodes are tainted with:
     • hypershift.openshift.io/control-plane=true:NoSchedule
     • hypershift.openshift.io/control-plane-serving-component=true:NoSchedule
   • Autoscaling: Creates a MachineAutoscaler for each, scaling from 0 to 1.

Prerequisites

• The management cluster must be a standalone Highly Available (HA) cluster.
• Currently, this utility is specific to AWS-based management clusters as it relies on AWS MachineSet provider specifications.