Case study

The backup loop that proves the dump will load.

A backup that only proves a file exists is not enough. The failure that matters is the one discovered on disaster day: a missing extension, truncated dump, version mismatch or row count that no longer matches production.

Internal infra · backup · continuity
  • 3operational stores covered by the current dumpInternal observation
  • 14 + 8daily and Sunday weekly retentionInternal observation
  • 0production writes during restore testingInternal observation
  • 16.15Postgres version recorded for recoveryInternal observation

How these were measured: Source: operations/systems/K-16-backup-cold-com-kratt-backup-cold.md and K-17-restore-test-com-kratt-restore-test.md record the build dates, green runs, retention, restore behavior and known exception.

Evidence class: Internal system

Reviewed 2026-09-16 · Kratt editorial review

What we shipped

backup-cold dumps the operational stores that would hurt if the box died and verifies row counts against the same run. restore-test then loads those dumps into a throwaway Postgres instance on a separate machine, checks for load errors and matching rows, and destroys the test database after the run.

  1. Daily backup loop with heartbeat and failure warning

  2. Row-count verification inside gzipped dumps

  3. Destination asserted outside any git work tree

  4. 14 daily plus 8 weekly retention pruning

  5. Weekly restore into throwaway Postgres on a non-production port

  6. Restore checks for SQL errors, table presence and row-count parity

  7. Known gaps surfaced on the page rather than hidden

Ownership and handover

Current owner

Kratt owns and operates this internal system.

Kratt's role

Kratt designed, built, and maintains the capability for its own work.

Handover

It is shown as capability evidence, not as a promised client deliverable.

What this page does not claim

  • No database contents, dump paths, host addresses, personal data, Slack webhook or credential material is exposed.
  • The n8n backup gap remains stated instead of claimed solved. The source card says the target exists but is blocked on a human approval and encryption-key handling.
  • No claim that this replaces a formal disaster-recovery program. It proves these dumps exist, verify row counts and load in a separate restore test.

Where is your business
leaking revenue?

A free AI audit finds it, ranked by impact, and it is yours to keep.

Book your free 30-min audit call

Back to all case studies.

kratt

kratt takes AI in companies from an initial idea to a reliable system. Based on the audit, we define where AI pays off and start development where the return is highest.

Book the free audit call
The newsletter

Occasional notes on
what’s actually working.

No spam. Cancel anytime. Occasional notes only.
Book your free 30-min audit call