Databricks
One query over system.billing.usage joined to system.billing.list_prices: the day, the SKU, the workspace, the DBUs and the cost. Never a table of yours, never a notebook, never a query anyone ran.
Access
- Kind
- A read-only key you paste
- What to give
- Workspace, token and a SQL warehouse
- Why that one
- A personal access token that can read the system tables and use one SQL warehouse. Valistry runs one query and nothing else.
What it attributes on its own
- From the source
- day · SKU as Databricks bills it · workspace · DBUs consumed · what Databricks charged, at the price that applied while it was used
- Needs a rule or a trace
- the model and the tokens — a serving endpoint's bill is DBUs, not tokens, so neither is on these rows · team, workflow and agent unless a rule maps the SKU or the workspace · outcome
Connecting it
- Check the system tables are on. system.billing has to be enabled on the metastore. An account admin does it once, and it backfills. Open ↗
- Create a personal access token. Settings → Developer → Access tokens. It carries your own read access and nothing more.
- Copy a SQL warehouse id. SQL Warehouses → your warehouse → Connection details. The query is small; the smallest warehouse is enough.
- Paste the three values and click Test and connect. Valistry runs the query over the last seven days, reports which SKUs it found and which it left out, then fills thirty.
A serving endpoint's bill is in DBUs, so this source carries no tokens. Where the same models are also reached through a gateway you connect, that gateway carries the tokens and this carries the bill — but connect only one of the two for the money, or the same work is counted twice.