Cost Explorer#
Cost Explorer shows what your infrastructure costs, broken down by provider, account, service and tag, with Kubernetes spend split down to namespace and workload.
Prerequisites#
Cost data comes from each provider's billing API, and each needs a permission beyond the metrics ones:
| Provider | Requirement |
|---|---|
| AWS | ce:GetCostAndUsage, and Cost Explorer enabled in the account |
| Azure | Cost Management Reader on the subscription |
| Google Cloud | roles/billing.viewer on the billing account |
| Linode | Included in the account read scope |
| Kubernetes | cost.enabled=true on the agent, see Kubernetes |
Opening it#
Cloud → Cost. Choose a period, then group by provider, account, service, region or tag.
Cloud providers publish cost data daily, not in real time. Today's figures are always incomplete and the last day or two will move. Compare whole days rather than reading the current partial day as a trend.
Views#
By service answers what is expensive. Usually compute and data transfer dominate, and the surprise is normally somewhere else: idle load balancers, orphaned volumes, cross-region transfer.
By account attributes spend across a multi-account estate.
By tag is the one that matters for chargeback, and it only works if tagging is consistent. Use Cloud Explorer to find untagged resources.
Kubernetes cost splits cluster spend across namespaces and workloads by their share of node CPU and memory, including idle capacity as its own line so you can see what you are paying for but not using.
Multi-cloud#
All connected providers appear in one view with a filter per provider, rather than as separate pages. Note that providers categorise services differently, so like-for-like comparison across providers needs care. Comparing your own spend over time is more reliable than comparing one provider's service names against another's.
Alerting on cost#
Create an alert on a cost metric the same way as any other, under Alerts → Create rule. Practical ones:
- Daily spend above an expected threshold, which catches a runaway job the day it starts.
- Month-to-date spend tracking above the same point last month by more than a set percentage.
- Spend appearing in a region you do not normally use, which often indicates a mistake or a compromised credential.
Reducing cost with the data#
- Idle resources: cross-reference Cost Explorer with metrics. An instance costing money with near-zero CPU for a month is a candidate.
- Oversized instances: compare instance size against actual memory and CPU used.
- Lambda memory: ObserveIQ collects maximum memory used per function. A function provisioned at 1024 MB and using 90 MB is paying roughly ten times what it needs.
- Data transfer: usually the least understood line. Group by service and look for transfer charges between regions or availability zones.
- Storage: orphaned volumes and old snapshots accumulate quietly and never appear in a compute review.