All case studies
Cost Optimization
Right-sizing memory across a multi-tenant SaaS cluster
Turning guesswork resource requests into data-backed ones on a multi-tenant Kubernetes platform — reclaiming capacity and stopping restarts.
- 20 → 0
- Auth restarts
- 7-day p95
- Sizing basis
- Reclaimed
- Capacity
// the challenge
A multi-tenant SaaS platform had memory requests set by guesswork across its Kubernetes cluster. Some services were badly over-provisioned and wasting capacity, while the identity service kept restarting under real load.
// our approach
- Analyzed seven days of Prometheus data to find true p95 memory usage per workload.
- Set honest requests and limits based on measured demand instead of round-number guesses.
- Tuned and stabilized the Keycloak identity service, taking it from frequent restarts to none.
- Reclaimed the over-provisioned reservations so the freed capacity could serve real workloads.
// the outcome
The identity service went from around twenty restarts to zero, and right-sizing across the cluster reclaimed significant over-provisioned memory — improving both stability and density on the same hardware.
Have a similar challenge?
Book a 15-minute call and we'll show you where we can help — no pitch, no obligation.