Kubernetes Cluster Cost Estimator
What a cluster costs a month, at rates you supply. No price table is bundled on purpose: a price list in a static site is stale the week it ships, and a wrong number that looks authoritative is worse than no number.
Take these from your own bill rather than from a list price, which a committed-use or reserved discount makes meaningless.
Or drop a file anywhere on this panel. Nothing is uploaded: the analysis runs in this tab.
The answer appears here
Paste on the left and press Estimate. Nothing leaves this tab.
Formatted
Results
Nothing else to flag.
No formatting problems, and nothing the rules object to. Worth remembering what that covers: this reads the file you pasted, not the account or cluster it will be applied to.
No finding matches that filter.
Common mistakes
These are the ones that fail silently. The config is accepted, nothing raises an error, and the consequence arrives later.
Costing nodes and stopping there
Load balancers, persistent volumes, snapshots, egress and the control plane are frequently a larger share than compute, and none appears on the node bill.
Instead:Include all of them. Egress in particular surprises people.
Sizing from requests rather than usage
Requests are what the scheduler reserves, and over-set requests mean paying for nodes that run nothing. Usage is what the workload needs.
Instead:Compare the two. The gap is the waste.
Using a bundled price table
Rates vary by region, commitment and discount and go stale quickly.
Instead:Enter your own rates, which is why this page asks for them.
Kubernetes schedules on requests, so unrequested capacity is unusable
The gap between what a cluster costs and what its pods request is not headroom. Nothing can be scheduled onto it, because the scheduler places pods by request rather than by usage.
Right-size requests before shrinking the cluster
Low utilisation against requests usually means the requests are wrong rather than that the cluster is too big, and shrinking first turns a cost problem into an availability one. Under-requesting is the other failure and is worse: it causes CPU throttling and OOM kills rather than a larger bill. Measure actual usage first.
Whichever of CPU and memory binds first is the one to size against
A cluster is full when either dimension is exhausted, and the other one is then wasted whatever it costs. This page reports which binds, because optimising the other one changes nothing.
Why no price table is bundled
A list price is wrong for your commitments, your region and your discounts, and a bundled copy is stale the week it ships. This is the same decision the Kafka cost estimator on this site took. Egress in particular is worth entering carefully: cross-zone traffic is the usual hidden cost, and topology-aware routing keeps it in-zone for one annotation.