What it shows
The cost per hour of each namespace over the last 12 hours, with what its pods requested in CPU and memory and the share of that they used, read as a bar in each cell. Select a namespace and its pods are listed the same way, each with its app label, the node it ran on, that node's instance type and whether the node is on-demand or spot capacity. Filter by pod or by node to see one of them on its own. A row called (idle) holds the part of each node no pod requested or used.
Only the last 12 hours, whatever the range says: each figure is worked out when the page is read, from every pod in the hour, so it is current rather than rolled up. For anything longer, by namespace and workload, use Kubernetes cost.
Questions it answers
- Which pods cost the most in this namespace right now?
- Is this pod using what it requests, or is its CPU request four times its usage?
- What runs on this node, and is it spot or on-demand?
- Which namespaces sit on the expensive instance type?
What it reads
Every hour, for each Kubernetes cluster, one small agent pod installed in the cluster reads what each pod reserved (its CPU and memory requests), what it actually used, and which node it ran on. Names and labels, such as the app and the namespace, come from Kubernetes itself. The agent is open source, in the qloudpulse-public repository, and the only thing installed in a cluster: nothing is installed in your AWS accounts.
Each node's cost for the hour, from the AWS run rate at list price or, on your own servers, from the price set for the cluster, is split between its pods as in the Kubernetes cost report: a pod pays for the larger of what it requested and what it used, CPU and memory apart. Each namespace adds up its pods.
What to do with it
Open it when a cluster's cost moved in the last hours and you want the pod. Sort the pods by cost and read the used-of-requested bars: a wide request with a short used bar is a request to lower, and a pod using more than it requested is one to raise before the node runs short. Pods on a dear instance type that do not need it point at a node selector or a taint to change.
Related reports
- Kubernetes cost: What each namespace and workload costs on your Kubernetes clusters, in AWS and on your own servers, and a month at this rate, by namespace and by cluster.
- AWS waste: Money paid for what is not used, sorted by how easily it goes: unused, overprovisioned or stranded, by category and by service.
- AWS run-rate cost: What AWS costs an hour now, by account, service and environment, priced from what is running rather than read from a bill that arrives a day late.
Every plan has every report, hosted or self-hosted, and the hosted plan is free during early access: see pricing. Like every report, this one splits and filters by account, environment, service and resource, steps back 13 months, downloads as PDF and CSV and keeps the view in its link: what every report does. All reports.