Databricks Cost Optimization
Cut your Databricks bill: OhChimp flags idle clusters, warehouses with no auto-stop, and pricey all-purpose compute, and drafts a fix for each.
What OhChimp optimizes for Databricks
Some of the cost levers OhChimp checks for Databricks. Each one becomes a reviewable plan you approve before anything changes.
Clusters with no auto-termination
An interactive all-purpose cluster set to never auto-terminate keeps billing DBUs while it sits idle. OhChimp lists your clusters and flags the interactive ones with autotermination_minutes at 0. It then drafts a plan to set a 30 to 60 minute window so idle clusters shut themselves down.
Warehouses with auto-stop disabled
A SQL warehouse with auto-stop turned off bills the whole time it waits for the next query. OhChimp reads your warehouse list and flags the ones with auto_stop_mins explicitly at 0. It plans a 10 minute auto-stop so the warehouse parks itself between queries. Serverless warehouses resume in seconds, so the wait costs you little.
All-purpose compute carrying scheduled jobs
All-purpose (interactive) compute is the most expensive DBU tier, and running scheduled jobs on it quietly inflates the bill. OhChimp sums your priced SKU spend from the billing tables and flags when all-purpose compute is more than 40 percent of it. It then recommends moving scheduled work onto cheaper job-compute or serverless-job SKUs.
Spend broken down by SKU
Most cost views stop at the workspace total. OhChimp joins system.billing.usage to your account's own list_prices and ranks spend per SKU over your chosen window, so you can see which compute SKUs actually drive the bill. The dollar figures use your account list prices, so negotiated discounts are not reflected.
Worked example
For example, say OhChimp finds two interactive clusters set to never auto-terminate, a SQL warehouse with auto-stop disabled, and a billing breakdown where all-purpose compute is 55 percent of your priced DBU spend.
It drafts a plan to set 30 minute auto-termination on the clusters, a 10 minute auto-stop on the warehouse, and a recommendation to move scheduled jobs onto job-compute SKUs. You press apply. OhChimp cannot yet measure what a single Databricks plan saves, so it does not mark one VERIFIED. The saving stays an estimate, and when the 30 days of monitoring end the result reads "Inconclusive."
Databricks cost optimization FAQ
How does OhChimp connect to Databricks?
With your workspace host and a personal access token, sent as a read-only Bearer credential. OhChimp checks the token against the SCIM identity endpoint, then lists your clusters and SQL warehouses. When you provide a SQL warehouse id, it runs read-only SELECT queries against the system.billing tables. It never stores application data, secrets, or workload contents.
Does OhChimp need write access or see my data?
No. The token only needs read access to list compute and, for the spend breakdown, SELECT on the system.billing usage and list_prices tables. OhChimp reads compute configuration and billing metadata. It never stores your notebooks, table rows, secrets, or workload contents.
What Databricks costs can OhChimp actually reduce?
The main ones are interactive clusters left running with no auto-termination, SQL warehouses with auto-stop disabled, and scheduled jobs running on expensive all-purpose compute instead of cheaper job-compute SKUs. When you set a SQL warehouse id, OhChimp also breaks spend down by SKU so you can see what drives the bill.
Who applies the changes, and can they be rolled back?
You do. Each fix is a reviewable plan with a confidence score, a risk level, and rollback steps. Nothing changes until you click apply, and OhChimp applies the matching code and infrastructure changes together.
How are the savings verified, and what does it cost?
Today Databricks savings stay estimates. OhChimp cannot yet measure what a single Databricks plan saves, so it does not mark one VERIFIED, and when the 30 days of monitoring end the result reads "Inconclusive." The only savings OhChimp measures today come from lowering the CPU and memory requests of a Kubernetes Deployment, StatefulSet, or DaemonSet. Pricing is a flat monthly fee with no cut of your savings, plus a year-one ROI guarantee that lets you request a refund of your first 12 months of subscription fees if the verified savings on your Costs tab fall short of them. Estimated savings do not count.
What the all-purpose-versus-job-compute lever is worth, priced end to end: Databricks jobs compute lists 73% under all-purpose.
Related integrations
Teams running Databricks usually run these too. OhChimp finds the waste in each and drafts every fix as a plan you approve.
Snowflake
Warehouse sizing, idle credits, query cost
PlanetScale
Plan tier, row reads, and storage
Confluent
Cluster sizing, throughput, and storage