Databricks Optimization Action Plans & Automation
An action plan is a proposed fix for a finding. It names the change, its safety tier, and the expected savings. LakeSentry is approval-first: every plan requires approval before execution.
Action plan types
Section titled “Action plan types”LakeSentry creates plans for these findings:
| Plan | Source finding | After approval |
|---|---|---|
| Terminate Cluster | Idle Cluster | LakeSentry executes it |
| Cancel Run | Runaway Job | LakeSentry executes it |
| Pause Schedule | Retry Storm | You apply it in Databricks |
| Reduce Auto Termination | Long Auto-Term (disabled or above 120 minutes) | You apply it in Databricks |
| Enable Spot Instances | Spot Candidate | You apply it in Databricks |
| Convert To Single Node | Single-Node | You apply it in Databricks |
| Scale Down Workers | Excess Workers | You apply it in Databricks |
| Upgrade Runtime | Outdated Runtime | You apply it in Databricks |
All other findings give a recommendation without a plan; see the catalog.
Safety tiers
Section titled “Safety tiers”The tier describes the risk of the change. It does not change the approval rule, and it does not show whether LakeSentry can execute the plan.
| Tier | Meaning |
|---|---|
| Auto | Low-risk change. |
| Review | Review the effect on the workload before approval. |
| Manual | LakeSentry provides the evidence and the proposed change. You apply it. |
Lifecycle
Section titled “Lifecycle”- LakeSentry detects a finding.
- It creates a plan when the finding type supports one and the required evidence exists.
- An admin or owner reviews the plan type, safety tier, savings estimate, and the evidence of the finding.
- The admin or owner approves or rejects the plan.
- For Terminate Cluster and Cancel Run, LakeSentry checks that the resource is still in the expected state, then calls the Databricks API. A snoozed or suppressed finding blocks execution.
- Check the plan status on the detail page of the finding. Successful execution resolves the finding.
For every other plan type, apply the change in Databricks, verify the result, then mark the finding resolved. Approval alone changes nothing in Databricks.
Before you approve
Section titled “Before you approve”- Read the recommendation of the finding first. An informational finding can advise to keep the current configuration.
- Assess the effect of the change in Databricks. Follow the change-management process of your team.
- Approvals and rejections are recorded in the Audit Log.