5 backup mistakes

Quick reference

1. ❌ Never restore-test — A backup that never restored is a hope, not a copy. ✅ Drill one restore a month — One path, off the backup host, timed and logged.

2. ❌ Keep one copy, one region — The region that holds the bucket can be the region that fails. ✅ Copy to a second region — Cross-region replication with its own credentials.

3. ❌ No retention policy — Every backup is kept, so the bill grows and the restore slows. ✅ Retention: 7d, 4w, 12m — Seven daily, four weekly, twelve monthly, then delete.

4. ❌ Unencrypted backups — A stolen archive is the production database in a bag. ✅ Encrypt at write, key separate — The key lives in KMS, not on the backup host.

5. ❌ No owner on the backup — Nobody knows what is inside, so nobody deletes it. ✅ Owner and purpose tag — A name and a restore purpose, checked each quarter.

Explained

The five

1. Never restore-test. A backup that never restored is a hope, not a copy. Drill one restore a month: one path, off the backup host, timed and logged. \2. Keep one copy, one region. The region that holds the bucket can be the region that fails. Copy to a second region with its own credentials. \3. No retention policy. Every backup is kept, so the bill grows and the restore slows. Seven daily, four weekly and twelve monthly is a start; delete the rest on a schedule. \4. Unencrypted backups. A stolen archive is the production database in a bag. Encrypt at write and keep the key in KMS, not on the backup host. \5. No owner on the backup. Nobody knows what is inside, so nobody deletes it. Put an owner and a restore purpose in the tag, and check it each quarter.

What they share

Each one is invisible while the job is green. The first restore is where they all arrive at once.

In one line

Restore one file

$tar -xzf backup.tgz -C /tmp/r