Before you start
- Priced clusters. A cluster with no price source reads not priced, never zero; see Get a number for every cluster.
- The
ankraCLI, signed in to your organisation. The steps use it because every organisation has it. billing.manageto add a budget in the last step. Every member can read the figures.- The portal’s Cost area is optional: it is in closed beta and enabled per organisation on request. Where it is on, each step names the tab that shows the same figures.
1. See when the run rate moved
summary shows the current run rate and the projected month end. trend draws the run rate per day. Days when the set of priced clusters changed - a cluster created, deleted, stopped, or losing its price - are listed apart: the line moved there without anything being spent. A rise outside those days is real spend.
In the portal: Cost → Overview.
2. Find the change that moved it
3. Read the cluster
4. Find the workload
namespaces lists the costliest 25 namespaces with their total, latest and peak, so a namespace that grew stands out. object stack prices one Stack on its own; object application and object namespace do the same for an application or a namespace.
In the portal: Cost → Explore spend, or Cost by namespace on the cluster’s Cost page.
You now know which cluster, which change and which workload. If the rise is expected, such as a new service or more traffic, set a budget in step 6 so the new level is watched.
5. Read what you could save
- right-size idle capacity,
- reduce run rate no namespace claims,
- stop a known non-production cluster on weeknights and weekends.
6. Set a budget that warns you
- CLI
- Portal (closed beta)
--amount is per month. --scope can also be organisation, project, environment (a cluster label) or application; a scope holds one budget.Verify
What can go wrong
Next steps
- Cloud Cost - how every figure is priced, rate cards for private capacity, and comparing providers.
- Cost budgets and the autopilot - let Ankra propose cost changes and measure what they saved.
- Power schedules - stop non-production clusters out of hours.