Databricks Waste Detection & Optimization Insights
LakeSentry looks for spend that produced little value, and for configuration changes that can lower cost or operational risk.
Two kinds of findings
Section titled “Two kinds of findings”- Waste findings describe observed behavior: an idle cluster, a warehouse that runs without queries, repeated failed runs, an endpoint without requests.
- Optimization findings evaluate a configuration: a long auto-termination timeout, more workers than the workload needs, an outdated runtime, a table layout that does not match its queries.
Several findings can describe the same resource. Review them together before you estimate the benefit of a change.
Where to look
Section titled “Where to look”| Area | Findings |
|---|---|
| Clusters | Cluster waste and configuration and sizing |
| Jobs and pipelines | Jobs and pipelines |
| SQL warehouses | SQL warehouses |
| Tables and queries | Tables and queries |
| Model serving, Apps, and workspaces | Model serving, Apps, and workspaces |
| Lakebase | Lakebase |
From finding to fix
Section titled “From finding to fix”- Open the finding and review its evidence and savings estimate.
- Apply the recommendation in Databricks. For an idle cluster or a runaway job, you can approve the action plan and LakeSentry applies it.
- Resolve, dismiss, snooze, or suppress the finding; see Choosing a response.