Distinguish Snapshot from Backup
A snapshot is a fast pointer to blocks on the same device; a backup is an independent copy you can restore from after that device is gone. Conflating them means a single disk or zone failure destroys "both copies".
Procedure
- Audit every copy you call a backup and label it: same volume? same array? same account? same region? Any "yes" is a snapshot, not a backup.
- State the failure each copy is meant to survive: block corruption, disk death, array loss, ransomware (needs immutability), accidental delete (needs retention), region loss (needs offsite).
- For every failure with no independent copy, add one. Minimum viable backup is: offbox, off-account, offline or immutable.
- Keep snapshots for their real job — fast rollback of hours-old mistakes and clone-and-test — with a short retention (typically 24-72h).
- Keep backups for disaster recovery with a retention that matches the business requirement, verified by restore.
- Make backups immutable or air-gapped: S3 Object Lock in compliance mode, an offline tape, or a pull-only repo the source cannot delete. A backup the source can
rm is erased by the same attacker.
- Encrypt backups with a key held separately from the source, so a compromised source account cannot read them.
- Document, per dataset, the RPO (snapshot age) and RTO (restore time), and confirm they match the business requirement.
Pitfalls
- Snapshots on the same physical array: a controller or power failure takes the live data and every "backup" at once.
- Ransomware that encrypts the live volume and then the mounted snapshot target, because both were writable from the host.
- Using a filesystem snapshot of a running database as a backup; the on-disk state is crash-consistent, not app-consistent, and may not replay.
- Assuming a managed provider's automatic snapshot is offsite; often it lives in the same region and account.
- Never testing the restore, so the "backup" is a hypothesis, not a proven fallback.
- Retaining snapshots indefinitely until they silently fill the array and block writes.
- Encrypting backups with the same key in the same secret store as production, so one leak compromises both.
Verification
aws s3api get-object-lock-configuration --bucket acme-backups
aws s3api list-objects-v2 --bucket acme-backups --query 'Contents[].Key' --max-items 5
Every dataset has at least one copy in a different account/region, locked or air-gapped, and the drill in rehearse-a-restore-on-a-timer restores from it.
Report: "Audited N copies: X are snapshots on the source array, Y are independent backups; Y copies are immutable and offsite, RPO h / RTO h."
1---2name: distinguish-snapshot-from-backup3description: Use when a storage plan treats snapshots as a backup. Separates point-in-time copies from restorable, off-array backups and closes the gap before an incident.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/distinguish-snapshot-from-backup6---78# Distinguish Snapshot from Backup910A snapshot is a fast pointer to blocks on the same device; a backup is an independent copy you can restore from after that device is gone. Conflating them means a single disk or zone failure destroys "both copies".1112## Procedure13141. Audit every copy you call a backup and label it: same volume? same array? same account? same region? Any "yes" is a snapshot, not a backup.152. State the failure each copy is meant to survive: block corruption, disk death, array loss, ransomware (needs immutability), accidental delete (needs retention), region loss (needs offsite).163. For every failure with no independent copy, add one. Minimum viable backup is: offbox, off-account, offline or immutable.174. Keep snapshots for their real job — fast rollback of hours-old mistakes and clone-and-test — with a short retention (typically 24-72h).185. Keep backups for disaster recovery with a retention that matches the business requirement, verified by restore.196. Make backups immutable or air-gapped: S3 Object Lock in compliance mode, an offline tape, or a pull-only repo the source cannot delete. A backup the source can `rm` is erased by the same attacker.207. Encrypt backups with a key held separately from the source, so a compromised source account cannot read them.218. Document, per dataset, the RPO (snapshot age) and RTO (restore time), and confirm they match the business requirement.2223## Pitfalls2425- Snapshots on the same physical array: a controller or power failure takes the live data and every "backup" at once.26- Ransomware that encrypts the live volume and then the mounted snapshot target, because both were writable from the host.27- Using a filesystem snapshot of a running database as a backup; the on-disk state is crash-consistent, not app-consistent, and may not replay.28- Assuming a managed provider's automatic snapshot is offsite; often it lives in the same region and account.29- Never testing the restore, so the "backup" is a hypothesis, not a proven fallback.30- Retaining snapshots indefinitely until they silently fill the array and block writes.31- Encrypting backups with the same key in the same secret store as production, so one leak compromises both.3233## Verification3435 aws s3api get-object-lock-configuration --bucket acme-backups36 aws s3api list-objects-v2 --bucket acme-backups --query 'Contents[].Key' --max-items 53738Every dataset has at least one copy in a different account/region, locked or air-gapped, and the drill in `rehearse-a-restore-on-a-timer` restores from it.3940Report: "Audited N copies: X are snapshots on the source array, Y are independent backups; Y copies are immutable and offsite, RPO <n>h / RTO <n>h."