Linux Disaster Recovery

Disaster Recovery

Distro support

Two-family skill. Backup/restore strategy and systemd rescue/emergency targets are identical; recovery-time tooling differs at a few critical points — get these wrong and a box won't boot. Body uses Debian/Ubuntu; substitute per this matrix.

RHEL-family note: the two recovery actions that most often differ are GRUB regeneration (grub2-mkconfig, not update-grub) and initramfs rebuild (dracut -f, not update-initramfs). Root is usually XFS — use xfs_repair, and remember XFS cannot be shrunk. See docs/multi-distro/plan.md. In sk-* scripts use the common.sh package primitives.

Use when

• Data has been lost, corrupted, or overwritten and a restore may be required.
• You need to recover databases, application files, or config snapshots from backups.
• You need an emergency recovery checklist during a production incident.

Do not use when

• The problem is only a service outage or bad config that can be fixed in place; use linux-service-management or linux-troubleshooting.
• The task is a routine backup review rather than an actual restore path.
• The task is creating backups — building rsync/tar archives, incremental snapshots, or filesystem (LVM/ZFS/Btrfs) snapshots. That now lives in the 13-backup-and-archiving category: linux-rsync-sync (offsite/incremental rsync), linux-archive-integrity (tar create + verify), and linux-filesystem-snapshots (LVM/ZFS/Btrfs). This skill stays focused on restore and emergency recovery.

Required Inputs

Capability Contract

Read/search access may assess backups and recovery options. Decryption, filesystem repair, database replacement, bootloader/initramfs work, service stops, and destructive restores require explicit authority, protected credentials, and a recovery target. Never overwrite the sole surviving copy.

Degraded Mode

Fallback when restore verification is unavailable: report the narrowest recoverable scope and blocking gap. Missing checksums, keys, isolated space, capacity, or application validation mean the recovery point has not passed.

Decision Rules

Workflow

Before selecting a restore point, use the two-family validation and recovery reference. Qualify the backup and target filesystem, test the smallest isolated restore, preserve the current target, and record application-level validation. If a recovery check is unavailable, report the narrowest recoverable scope and keep the action open; do not treat a successful file extraction as proof that the service or data is recovered.

1. Stabilise the incident, preserve current state, and confirm restore is necessary.
2. Define RPO/RTO, exact scope, target, authority, and rollback; stop if any required decision is absent.
3. Inventory candidate backups and verify timestamp, checksum, decryptability, and pre-incident status.
4. Test the smallest suitable restore in isolation and run application validation.
5. Confirm destructive impact, back up current target, execute the authorised restore, and record commands.
6. Validate data/service/access; on failure recover the pre-restore target and keep the incident open.

Evidence Produced

Quality standards

• Restore the smallest correct scope first when possible.
• Use timestamps and incident facts to choose the backup, not guesswork.
• Verification after restore is mandatory.

Anti-patterns

• Restoring over live data without confirmation. Fix: record authority, target, impact, and before-state backup.
• Choosing the latest backup blindly. Fix: select a verified point before the incident/corruption.
• Ending after the restore command. Fix: run application, integrity, service, and access checks.
• Testing decryption for the first time during production restore. Fix: qualify the backup in isolation first.
• Repairing the wrong filesystem with family assumptions. Fix: detect filesystem and use its supported tooling.
• Overwriting the only surviving copy. Fix: restore into isolated space or clone current state first.

Outputs

Worked Example

A table was corrupted at 10:20 EAT. Verify and decrypt the 10:00 database backup into an isolated database, validate row counts and application queries, then restore only the affected database after approval. Keep a dump of the damaged live database for rollback and investigation.

Book-derived additions

For a mission-aware NGO recovery plan, load ../../07-security-and-hardening/linux-ngo-cyber-resilience/SKILL.md and retain evidence preservation, staged restoration, integrity checks, and credential rotation.

References

• references/backup-strategy.md
• references/restore-procedures.md
• ../../docs/continuous-improvement/two-family-validation-and-recovery.md
• ../../docs/continuous-improvement/incident-learning-standard.md

This skill is self-contained. Every command below works on a stock Debian/Ubuntu or RHEL-family server (substitute per the Distro support matrix above). The sk-* scripts in the Optional fast path section at the bottom are convenience wrappers — never required.

Always confirm before restoring. A restore overwrites existing data. Never start a restore without typing the full word yes at the prompt, even in non-interactive mode.

Step 1: Assess First

# Is this a service crash (restart only) or actual data loss?
sudo systemctl status nginx mysql postgresql php8.3-fpm

# When did it happen?
sudo journalctl --since "2 hours ago" | grep -iE "error|fail|crash" | head -20

Service crash → restart it (linux-service-management), no restore needed. Data loss/corruption → proceed below.

Step 2: Find The Right Backup

# Local backups (7-day retention)
ls -lth ~/backups/mysql/*.gpg 2>/dev/null | head -10

# Google Drive (3-day retention for MySQL)
rclone ls gdrive:<backup-folder> 2>/dev/null | sort | tail -10

# If rclone token expired:
rclone config reconnect gdrive:

Choose the backup closest to before the incident.

Step 3: Restore

Full restore procedure (decrypt → extract → import): See references/restore-procedures.md.

The procedure, condensed:

# Decrypt (enter passphrase when prompted)
gpg --decrypt backup.sql.gz.gpg > backup.sql.gz

# Inspect size and sanity
gunzip -l backup.sql.gz
zcat backup.sql.gz | head -20

# Stop the service that writes to the DB
sudo systemctl stop nginx apache2 php8.3-fpm

# Restore (confirm first!)
zcat backup.sql.gz | mysql -u root -p <database>

# Restart
sudo systemctl start php8.3-fpm apache2 nginx

Emergency Checklist

# 1. Stop affected service to prevent further damage
sudo systemctl stop <service>

# 2. Find best backup (Step 2 above)

# 3. Decrypt → restore → verify (references/restore-procedures.md)

# 4. Restart all services
sudo systemctl start nginx mysql php8.3-fpm apache2

# 5. Re-run security audit
sudo bash ~/.claude/skills/scripts/server-audit.sh

# 6. Clean up
rm -rf ~/restore/

Demo/Dev Reset (Git-Tracked SQL Dump Pattern)

Some apps ship a git-tracked SQL dump as the demo DB source of truth. A reset script drops and recreates from that dump:

ls /usr/local/bin/reset-*           # find available reset scripts
sudo reset-<app>-from-git           # requires typing YES
ls /var/backups/<app>/              # safety backup always created first

Optional fast path (when sk-* scripts are installed)

Running sudo install-skills-bin linux-disaster-recovery installs wrappers for the above:

These are optional wrappers around the commands above.

Demo/Dev Reset (Git-Tracked SQL Dump Pattern)

Some apps ship a git-tracked SQL dump as the demo DB source of truth. A reset script drops and recreates from that dump:

ls /usr/local/bin/reset-*           # find available reset scripts
sudo reset-<app>-from-git           # requires typing YES
ls /var/backups/<app>/              # safety backup always created first

Scripts

This skill installs the following scripts to /usr/local/bin/. To install:

sudo install-skills-bin linux-disaster-recovery