Plan compute & storage

Before you start

Collect measurements from an existing environment or a representative load test. Where no measurement exists, label your estimate and schedule a test before a production decision.

List what needs resources

Record every application component separately: web/API, background workers, scheduled jobs and databases. Include replicas, per-container requests and limits, persistent data, image architecture, expected traffic and external dependencies. Separate production and test environments in the inventory.

Calculate requests and rollout headroom

Illustrative workload calculation, not a compute specification or price quote:

Scheduling uses resource requests. Limits constrain runtime use; they are not extra reserved capacity. Node allocatable capacity is lower than nominal machine capacity after system reservations. Spread, volume placement and the largest Pod also matter: a total that fits across nodes may still leave a Pod unschedulable.

Confirm capacity and quotas

The plan's maximum node count is a management limit, not an included allowance of free machines. Ask for the characteristics and recurring cost of Compute S/M/L/XL in your proposal. Confirm the namespace quota, supported volume options and whether the chosen topology can meet your workload requirements.

After handover, read the same limits in the Clusters and Usage views. The browser should show the selected plan limit, current capacity, storage and any proposal-specific restriction. An empty table does not confirm unlimited capacity.

Set a change budget

Agree who may request additional workers, larger compute, storage growth or external endpoints. Include recurring cost and any one-off work in the approval. Re-measure after the first workload and before expanding. Keep a resource inventory even when the infrastructure is managed for you.