Field guideUnclassified · for engineering distribution
- Guide
- BM-RG-02
- Prepared for
- Platform, data and security engineers
- Updated
- Reading time
- 9 min
Modern database hardening.
Abstract
Most estates now span hyperscale clouds, SaaS databases and stores nobody has looked at since they were built. You cannot harden what you have not listed, and you cannot prove hardening you did by hand.
This guide is the order we work in: inventory first, four control pillars, posture enforced as code, a rehearsed response and metrics that show the estate getting safer.
- Each store sits inside its own perimeter
- Classification and owner recorded per store
- A store without a tested restore is flagged
1.Start with a living atlas of data stores
Hardening starts with an accurate list. Inventory cloud-native services, customer-managed instances and legacy appliances, then tag each one with its business criticality, data classification and blast radius, so that engineering, legal and incident response rank the same stores first.
- Pull metadata nightly from the providers’ own inventories (AWS Config, Azure Resource Graph, Google Cloud Asset Inventory) and from your SaaS vendors.
- Give every new data store the baseline policies the day it appears, not after the next review.
- Map each data product to an accountable team and an executive sponsor. A store with no owner is a finding in itself.
2.Four control pillars
Every hardening sprint lands in one of four pillars. Each mixes automation with human review, so controls apply consistently across clouds and an auditor can follow what was decided.
| Pillar | Control | Evidence left behind |
|---|---|---|
| Identity | Just-in-time roles, workload identities and service identities in place of shared credentials. Conditional MFA, key rotation and lifecycle rules for people and machines alike. | Who held which role, for how long, approved by whom. |
| Data guardrails | Encryption, tokenisation and masking set as defaults in Terraform or Pulumi modules, with drift detection against that baseline. | The module version each store was built from, and every drift and its fix. |
| Network perimeters | Private service endpoints and software-defined perimeters instead of a flat network. Over-permissive rules and unexpected ingress paths are found and closed. | The rule set at any date, and each change to it. |
| Resilience and recovery | Immutable backups, restore tests and failure drills, rehearsed with legal and communications in the room. | Last successful restore per store, and the drill record. |
3.Enforce posture as code
Once the pillars are defined, enforce them daily so engineers can spend their time on harder problems. Write the desired state as reusable policy packs that cover identity, network, secrets and backup tooling.
- Run policy-as-code checks in CI/CD and block risky schema or permission changes before they reach production.
- Track remediation across AWS, Azure, Google Cloud and private clouds on one controls dashboard.
- Send database activity streams (CloudTrail, Azure Monitor, Google Cloud audit logs) to detections for privilege escalation, bulk export and ransomware behaviour.
4.Rehearse the breach
A hardened estate still meets determined attackers. Bring incident command, compliance and legal together before it happens, so the response is rehearsed, documented and fast.
- Scenario tests. Simulate a cross-cloud data breach and test the containment steps, regulatory notifications and customer messages, end to end.
- Briefings in plain language. Turn telemetry into an account that a CISO, a data protection officer and a board can all act on.
Run these quarterly, so that you know exactly how you would contain stolen credentials, a rogue administrator or corrupted data, even when the incident crosses providers.
5.Measure, then go again
Tie control health to risk and regulatory outcomes, so leadership can see which teams are improving fastest and where sponsorship is missing.
- Track the time it takes a new service to reach baseline, alongside coverage for encryption, segmentation and verified backups.
- Put penetration-test and red-team findings on the same dashboard to check that controls actually hold.
- Publish a quarterly review: what improved, what is still open and what the next budget buys.