Cost explorer

Cost by any dimension, drill-down to usage lines, unusual days, forecast, CSV.

What it is

Where the euros come from. Billing → Cost explorer draws your metered usage over time, stacked by any dimension (service, usage type, resource, project, tag, zone, region, the person who created it) and lets you narrow it with filters and follow any bar down to the individual usage lines: which machine, how many hours, at what rate. Next to the chart it lists the days that cost unusually much and why, what changed most against the previous period, a forecast for the month, discounts and credits, the budgets you have set, and every billed resource with a warning when it looks forgotten.

Every view is a link: the dates, grouping and filters live in the page address, so the view you are looking at can be pasted into a ticket.

Amounts are exact. Usage is stored in EUR with eight decimal places and is summed without rounding, so the explorer's total for a calendar month is the invoice's net for that month to the cent. Rounding to cents happens only on the invoice line.

Concepts

Term Meaning
Dimension A way to slice cost. service, usageType (the SKU), resource, resourceType, region, zone, project, tag (with a tag key) and creator.
Group by The dimension the chart is stacked by. The eight largest groups are drawn on their own and the rest are summed into Other.
Filter Keep (is) or drop (is not) the rows whose dimension has one of the given values. Several filters all apply. The empty value selects rows that have none: untagged, no project, unknown creator.
Granularity Hourly (windows up to 14 days), daily, or monthly. Storage is metered once a day; in an hourly view its daily amount is spread evenly over the day's 24 hours so it does not show up as a spike at midnight.
Previous period The window of the same length immediately before the one you chose. Every change figure compares with it.
Unusual day A day whose total is above the median of the 14 days before it by more than 40 % and by more than €1. Days with fewer than 7 days of history are not judged. Each one names the resources that explain the jump. Consecutive unusual days caused by the same resources are shown as one change with its length and total.
Forecast Month-to-date plus the remaining days at the average of the last 7 complete days. A spike inside those 7 days lifts it.
Run rate What a resource cost on the last complete day, times 30: what leaving it running costs per month.
Credit A fixed EUR amount granted to your organization for whole calendar months. Each month's net consumes the credits valid in it, oldest first.
Discount A percentage off usage metered inside its validity window, optionally only for some services. Applied before credits.
Created by The IAM user who created a resource, recorded at creation. Resources created before November 2026 show unknown creator.

Credits and discounts are granted by the platform operator; there is no way to grant them to yourself. The same calculation produces the explorer's net and your invoice's credit lines, so the two agree.

How to

Find out why costs went up

Open Billing → Cost explorer, pick the window, and look at Unusual days: each entry says how far the day was above its usual level and which resources added the difference. Click the day for its usage lines. Top movers lists the resources and groups whose cost changed most against the previous period, in euros.

See what a project or team costs

Set Group by to Project, or to Tag: team, and read the table under the chart. Click a group to filter to it and break it down by resource; click Usage lines for the metered lines behind it.

Find what is still running that nobody uses

The Resources table marks resources that are still being billed and look forgotten:

Warning When
Idle CPU under 2 % on average and under 10 % at peak over the last 7 days.
Detached A volume attached to no machine, still billed for its size.
Untagged A resource that can carry tags but has none, so no project or team owns it.
No owner Untagged and with no recorded creator.
Stopped, still billed Stopped, but billed in the last day.
Deleted, still billed Gone, but billed in the last day.

Tick Only with warnings to see just those, with the monthly cost of leaving them.

Savings: are these servers necessary?

Billing → Savings lists, resource by resource, what costs money without doing much, with the evidence in one sentence, the monthly saving (after your discounts, before VAT) and the action:

Finding When Action
Idle machine Running, CPU averaged under 2 % and never above 10 % over the last 7+ days. Stop
Oversized machine 7+ days of CPU samples, and a smaller size of the same family fits its 95th-percentile CPU with 1.5× headroom — and its 95th-percentile memory with 1.3× headroom when the machine publishes memory (below). Without memory the finding says check first. Resize to the named size
Oversized database Its busiest node's 95th-percentile CPU stayed under 10 % per vCPU for 7+ days and the next smaller class (one step, never further: memory is not measured) fits with 1.5× headroom. Saving per node × nodes. Change the instance class
Stopped server, still billed Stopped for 7+ days while its attached volumes keep billing (a stopped machine is not billed; its disks are). Terminating keeps volumes not marked delete-on-termination: delete them too (after a snapshot, if the data matters). Terminate
Unattached volume Attached to nothing for 7+ days, or never attached and older than a day. Take a snapshot first if it may be needed again. Delete volume
Unused elastic IP Associated with nothing for a day. Elastic IPs are not billed yet, so this saves €0 on today's bill. Release

The action button runs the named API operation with your own permissions; it is recorded in the audit log like any other call. Keep takes a reason and hides the finding for good or for 90 days; everyone sees who kept it and why, and Bring back undoes it. Findings are computed fresh from the last 14 days of monitoring and the current price list each time the page opens.

Memory. The provider does not report memory use. A machine that publishes the custom metric mem_used_percent with the dimension InstanceId=<instance id> (any namespace — the CloudWatch agent's convention) is judged on memory too:

dcloud monitoring custom-metrics put --namespace MyApp/Agent \
  --metric-data '[{"metricName":"mem_used_percent","value":41.5,"dimensions":{"InstanceId":"i-abc123"}}]'

Saved views and the monthly mail

Save this view above the explorer stores the current question under a name, with a range that moves along (Last month, This month, Last 30 days, …) or fixed dates, shared with the organization or private to you. Tick Mail monthly to have it sent on the 2nd for the month before, with its own CSV — e.g. "Shop by service" to the shop team.

The organization's monthly cost report (Billing → Budgets) carries: the bill (gross, discounts, credits, net), what could be saved, why did this change (the five resources whose cost moved most, rising or falling, plus "other" and the discounts/credits line, adding up exactly to the change in the net), unusual days, top movers by service, the costliest services and resources, and budgets. Attached: cost-lines-YYYY-MM.csv (every resource and usage type of the month) and accounting-YYYY-MM.csv. Send again sends a month once more, at most every 10 minutes.

Accounting export (DATEV)

DATEV export (or GET /billing/api/cost-explorer/accounting-export?start=YYYY-MM-DD&end=YYYY-MM-DD, or ?month=YYYY-MM) writes the date range on screen as a DATEV Buchungsstapel that DATEV Rechnungswesen imports as it is: the EXTF header record (format version 700, Buchungsstapel 13, not fixed), the 125 standard columns, Windows-1252, semicolons, decimal comma. Set your Beraternummer and Mandantennummer (from your tax advisor), the first month of your fiscal year and the account length under Billing → Budgets → Monthly cost report; DATEV refuses a file without them. A range across two fiscal years is refused: export the two parts.

One row per month × cost centre × project × service. The columns that carry values:

Column Content
Umsatz (ohne Soll/Haben-Kz) Net amount, before VAT, after discounts, 1234,56
Soll/Haben-Kennzeichen S for charges, H for credits
WKZ Umsatz EUR
Konto, Gegenkonto (…) Your expense and creditor accounts, set under Billing → Budgets → Monthly cost report
Belegdatum The last day of service in that month inside the range, DDMM — never a day still to come
Belegfeld 1 The invoice number, or YYYY-MM before there is one
Buchungstext Service and month, vorläufig while the month is not over; Gutschrift MM/YYYY … for a credit
KOST1 - Kostenstelle The value of the cost-centre tag (default cost-center; configurable)
KOST2 - Kostenstelle The project. A resource in two projects is split evenly between them
Leistungsdatum Same day as Belegdatum, DDMMYYYY

Cents are distributed by largest remainder, so the rows add up exactly to the net — for a whole month, the invoice's subtotal.

Forecast per project, and who runs where

When the window reaches into the current month, the groups table has a forecast column: each group's month so far plus the remaining days at its own last-7-day average (a group younger than 3 days shows "—"). Grouped by Project, a monthly budget on that project shows the forecast as a percentage of the budget.

Organization admins (iam:UpdateOrganization) can also group and filter by Runs on: "Deutsche Cloud (managed)" or the name of the organization's own provider, as in Settings → Infrastructure.

Budget alerts on the event bus

Every budget alert also arrives on the organization's default event bus as source: ews.billing, detail-type: Budget Threshold Crossed, with detail.kind actual or forecast — route it to a queue, a function or an HTTPS endpoint with an event rule. A monthly budget can alert on its forecast too (forecastThresholds, e.g. [100]): when spend so far plus the remaining days at the last 7 days' rate reaches the threshold. Each threshold fires once per period and kind.

Look at one bar in detail

Click a bar. A panel lists every usage line in it, largest first: resource, usage type, quantity and unit, unit price, amount, how many records, and the first and last metered hour.

Export

Two buttons download every (period, resource, usage type) row of the current view:

  • CSV for Excel (German): semicolons, decimal commas, column names in words, dates as 18.11.2026, euros to the cent, UTF-8 with a BOM. A German Excel opens it as columns of numbers you can sum. Rounding each row to the cent can make the sum differ from the explorer's total by a few cents.
  • CSV (machine-readable): commas, decimal points, the column ids below and all eight decimals, so a sum reproduces the explorer's total exactly.

Over the API, ?format=excel-de picks the first (the default is machine):

curl -X POST "$EWS/billing/api/cost-explorer/export?format=excel-de" \
  -H "Authorization: EWS-ACCESS-KEY $KEY" -H 'Content-Type: application/json' \
  -d '{"start":"2026-10-01","end":"2026-10-31","granularity":"day"}' > october.csv
dcloud billing costs query --start 2026-10-01 --end 2026-10-31 --group-by '{"dimension":"tag","tagKey":"team"}'
dcloud billing costs export --start 2026-10-01 --end 2026-10-31 > october.csv

Reference

API

All need billing:GetCostAndUsage. They read and write nothing; the query endpoints are POST because the query is a nested structure.

Method and path What
POST /billing/api/cost-explorer/query Series, groups, previous period, movers, unusual days, forecast, discounts and credits, resources with warnings, filter values, budgets.
POST /billing/api/cost-explorer/lines The usage lines behind a selection; pass bucket (a bar's ISO start) and granularity for one bar.
POST /billing/api/cost-explorer/export CSV.
GET /billing/api/credits Credits and discounts with what each has taken off and what is left.

Query body:

Field Type Default
start YYYY-MM-DD, first day, UTC required
end YYYY-MM-DD, last day (inclusive), UTC required
granularity hour (≤ 14 days), day, month day
groupBy { "dimension": "...", "tagKey": "..." } { "dimension": "service" }
filters [{ "dimension", "values": [...], "tagKey"?, "exclude"? }], at most 20 []
maxGroups 1–50 8

A window is at most 400 days. Amounts in responses are EUR strings with eight decimal places.

CSV columns

period_start, service, service_name, usage_type, usage_type_name, resource_type, resource_id, resource_name, region, zone, projects, created_by, tags, quantity, unit, amount_eur, runs_on. tags is key=value pairs separated by ;, projects likewise. amount_eur is unrounded. The Excel file has the same columns in the same order, named in words ("Amount (EUR, net)"), with the resource type in words too.

Projects and the explorer

A resource that matches several projects is counted under the first by name when grouping by project, so the groups add up to the total. A project filter matches every project the resource is in.

What is not supported

  • Network load balancers (lb11–lb31), elastic IPs, snapshots, NAT gateways and object storage buckets are not metered yet, so they appear nowhere in the explorer or on the invoice. application load balancers are metered (lb.application.hour, service Load balancing, one line per load balancer).
  • Registry, git and queue storage, queue requests and email are metered for the whole organization, not per repository or queue, so they cannot carry tags or a project.
  • The accounting export has no EXTF header (Beraternummer, Mandantennummer): import it as an ASCII file or let your tax advisor add the header.
  • Memory use is judged only for machines that publish mem_used_percent; databases are judged on CPU only.

Troubleshooting

A filtered view shows discounts but no credits. Credits apply to a whole month's bill, not to a slice of it.

The month's total differs from the invoice by credits. The explorer's chart shows usage before discounts and credits; the Discounts and credits card shows the net, which is what the invoice charges.

Did this page help you?