// FinOps / Cost Optimization
Cost optimization: from Kubernetes to Cloud Run
Stateless services ran on an always-on GKE cluster sized for peak — idle most of the day, billing all of it. Migrating the right workloads to Cloud Run with scale-to-zero cut GCP compute spend without hurting reliability.
Impact Metrics
55%
GCP compute cost cut
→ 0
Idle cost, scale-to-zero
0
Downtime migrating
Stack Applied
// Context & Background
A web3 product ran everything on a permanently-provisioned GKE cluster. Node pools were sized for peak traffic that arrived a few hours a day, so the cluster sat largely idle the rest of the time — and billed for every idle core.
// The Challenge
Cut the GCP bill meaningfully without re-architecting the whole application or adding cold-start latency to the endpoints that couldn't tolerate it.
// The Roadmap & Approach
Utilization audit of the GKE estate
Profiled every workload's real CPU/memory usage and request patterns to separate the genuinely always-on services from the bursty, stateless ones.
Migrate stateless services to Cloud Run
Moved stateless, request-driven services to Cloud Run with scale-to-zero — so they cost nothing when idle and scale out automatically under load.
Right-size what stays on Kubernetes
Kept stateful and long-running workloads on a much smaller, right-sized GKE node pool — paying only for what genuinely needs to stay warm.
Tune cold starts and keep CI/CD intact
Set minimum-instance counts on latency-sensitive services, fronted everything with Cloudflare, and ported deployments cleanly into the existing CI/CD.
// Validated Results
- ~55% reduction in GCP compute spend
- Non-critical services now scale to zero — idle cost effectively eliminated
- Zero downtime during the migration
- Far less Kubernetes upkeep for the services that moved off it
Facing similar architecture challenges?
Let's talk through your system architecture — a 30-minute review session usually surfaces the quickest reliability and FinOps improvements.