Size your ImageEngine Kube cluster.
Turn a few workload knobs into a starting-point values-override.yaml. These are bench-from numbers — re-measure after deploying. See the sizing guide for the full rationale.
Self-managed clusters have no cloud LB — front the edge with MetalLB or NodePort.
Total edge ingress. Cache layers pare this down: ~90% hit at Varnish, ~75% of the rest at OSC — so 1000/s ≈ 100/s at the backend and ~25/s of image processing.
Fraction of your origin catalog kept in OSC. Processed variants add ~2× more (20 variants/origin at ~10% of origin size each).
Assumes X GiB RAM = X/2 vCPU. Drives node-count estimates, processor pod size, and caps Varnish memory/disk to your nodes.
Per-request structured JSON. High volume — Auto disables it in production or above ~100/s to save I/O, CPU & storage.
One record per origin fetch (cache misses on the cold path) — far lower volume than access logs, so it stays on by default.