Databricks Optimization Action Plans & Automation
Action plans are generated from eligible insights. They describe a proposed remediation, expected savings, evidence, and the operational risk before anything is changed.
LakeSentry is approval-first: all generated action plans currently require review and approval before execution.
Current action plan types
Section titled “Current action plan types”LakeSentry can generate plans for these findings:
| Plan type | Typical source insight | Execution today |
|---|---|---|
| Terminate idle cluster | Idle cluster | Executable after approval |
| Cancel runaway run | Runaway job or excessive duration | Executable after approval |
| Pause job schedule | Repeated waste or failures | Review/manual |
| Reduce auto-termination | Long auto-termination | Review/manual |
| Enable spot instances | Spot candidate | Review/manual |
| Convert to single-node | Single-node candidate | Review/manual |
| Scale down workers | Overprovisioned workers | Review/manual |
| Upgrade runtime | Outdated Databricks Runtime | Review/manual |
The action executor currently performs Databricks mutations for terminating clusters and canceling runs. Other plan types are persisted with evidence and proposed changes for operator review until direct execution support is added.
Safety tiers
Section titled “Safety tiers”| Tier | Meaning |
|---|---|
| Tier 0 / Auto | Low-risk action classification. In the current release, Tier 0 plans still require approval before execution. |
| Tier 1 / Review | Requires human review and approval before execution. |
| Tier 2 / Manual | LakeSentry provides instructions/evidence; the operator applies changes manually. |
The UI labels these as Auto, Review, and Manual. Today those labels describe risk and handling, not unattended execution. Most production-impacting infrastructure changes should be treated as review or manual until your organization has built trust in the detector.
Lifecycle
Section titled “Lifecycle”- An insight is detected.
- A background worker creates an action plan when the type is supported and required evidence exists.
- A user reviews the plan, evidence, savings estimate, and blast radius.
- The plan is approved, rejected, or left pending.
- If approved and executable, LakeSentry calls the relevant Databricks API.
- The timeline records success, failure, or manual completion.
Guardrails
Section titled “Guardrails”- Plans are tied to specific insight evidence.
- Missing or contradictory evidence prevents plan generation.
- Execution status is audited.
- Admins can keep risky recommendations manual.
- Treat generated plans as operational changes; follow your team’s change-management process.