Why is your Databricks bill so high?

Paste one row per cluster — kind, DBUs used, autotermination — and see what each cluster costs, scheduled work paying the all-purpose rate, clusters that never terminate, long autotermination tails and the setting to change.

Nothing is uploaded and Databricks is never contacted. The text box is analysed in this browser and never leaves your device. Savings are modelled from the numbers you paste, which is an estimate; cloud VM cost is not included.

1. Paste your clusters

Rows of cluster, kind (interactive, scheduled or jobs), DBUs this month, autotermination minutes (or never), DBU per hour, starts a month (last two optional)

Prefer a CLI, or want a CI check?

Same logic, zero dependencies, one command — reads a file or stdin:

node cli.mjs analyze clusters.csv
cat clusters.csv | node cli.mjs analyze -
node cli.mjs analyze clusters.csv --terminate 60 --fail-on warn   # exit 1 on any waste
node cli.mjs analyze clusters.csv --json --prices my-prices.json

Get cli.mjs and engine.mjs — Node 18+, nothing to install, no network call. More: Databricks cost: autotermination, Jobs compute and DBU rates.

What's recognised

Input	Handling

Prices are illustrative and dated; edit them with --prices in the CLI. Cloud VM cost, SQL warehouses, serverless, Photon and storage are not modelled.

Common questions

Does this connect to my Databricks workspace?

No. It reads the rows you paste and never contacts Databricks. It cannot see your contract, so every dollar figure is the DBUs you pasted times a per-DBU price you can edit.

How do I get the numbers?

DBUs per cluster: SUM(usage_quantity) grouped by usage_metadata.cluster_id and sku_name over 30 days from system.billing.usage. Autotermination is in each cluster's configuration. DBU per hour is shown in the cluster UI. Starts come from the cluster event log.

Why is scheduled work on all-purpose compute expensive?

All-purpose compute lists at roughly 3.5 times the DBU rate of Jobs compute. A notebook job pointed at a shared all-purpose cluster pays the higher rate for work that a job cluster would do for less. Moving it is a job-definition change; VM cost is the same.

What autotermination should I use?

30 minutes is a common starting point for interactive clusters; shorter saves more but restarts cold. Autotermination off keeps the cluster, and its DBUs and VMs, billing until someone stops it.

Is my data uploaded anywhere?

No. The browser page analyses the text in your own browser and never sends it anywhere. The CLI reads a local file or stdin and makes no network call.