Plan recovery

Apply the 3-2-1 pattern

Place multiple copies across distinct storage and one offsite boundary without counting synchronized replicas incorrectly.

The 3-2-1 rule is the classic baseline. Keep three copies of important data, on two kinds of storage, with one copy offsite.

Each number defends against a different disaster. Three copies means one failure never leaves you with a single copy. Two storage types means a bad disk batch or a filesystem bug can’t take everything. One offsite copy means a fire, a flood, or a burglary at one location doesn’t end the story.

Map one dataset

Take a real dataset and write down where its copies live:

copy 1: live laptop data
copy 2: versioned backup on external disk
copy 3: encrypted versioned backup in remote object storage

Notice that the live data counts as copy one. You only need two backups to satisfy the rule. But they must be independent from each other.

Now name the failure each copy survives. Accidental deletion, disk loss, theft, account compromise, site disaster. This is how you verify a backup plan. If two copies fail for the same reason, they count as one:

accidental deletion  -> external disk (has history)
laptop disk death    -> external disk, object storage
house fire           -> object storage only
cloud account hacked -> external disk only

Every threat should have at least one surviving copy. If a row comes up empty, you found the gap before the disaster did.

The counting mistake

The most common error is counting a synchronized mirror as a backup. A mounted mirror shares every deletion, malware infection, and operator mistake with the live data. Delete a file and Dropbox, RAID, or an always-connected rsync target deletes it too. Instantly and faithfully. Ransomware works the same way. It encrypts the mirror right along with the source.

What makes a copy a real backup is version history plus some independence from the live system. An offline disk. Separate credentials. Snapshots the source machine can’t overwrite. Independence matters more than the count.

Plenty of setups have “three copies” that are all one rm -rf away from zero. Try the table on your own data, and count only the copies that would survive each row.

Lesson completed