generate-backup-scripts

Generate Backup Scripts

Translate the strategy document into runnable code. Output lives in backup-plan/scripts/ and, where relevant, in the project's existing ops/, .github/workflows/, or k8s/ directories.

Inputs

• backup-plan/BACKUP-STRATEGY.md — the source of truth. If it doesn't exist, stop and run document-backup-strategy first.

Per-tool skeletons

Match the tool(s) the strategy picked. Always produce: (a) the backup script itself, (b) a matching restore script stub, (c) the scheduling unit, (d) a short README inside backup-plan/scripts/ explaining how to run and verify.

restic → S3-compatible

#!/usr/bin/env bash
set -euo pipefail

: "${RESTIC_REPOSITORY:?}"     # e.g. s3:https://s3.us-east-005.backblazeb2.com/<bucket>/<prefix>
: "${RESTIC_PASSWORD_FILE:?}"
: "${AWS_ACCESS_KEY_ID:?}"
: "${AWS_SECRET_ACCESS_KEY:?}"

TAG="${BACKUP_TAG:-$(hostname -s)}"
SOURCES=( /path/to/data /etc/<app> )
EXCLUDES=( --exclude-caches --exclude '*.tmp' --exclude 'node_modules' )

restic backup "${SOURCES[@]}" "${EXCLUDES[@]}" --tag "$TAG" --one-file-system
restic forget --tag "$TAG" --keep-daily 7 --keep-weekly 4 --keep-monthly 12 --keep-yearly 3 --prune
restic check --read-data-subset=5%

Postgres logical dump + restic

Dump to a FIFO and pipe into restic to avoid landing plaintext on disk:

restic backup --stdin --stdin-filename "${DB}-$(date +%F).sql" \
  --tag pg-${DB} \
  <(pg_dump --format=custom --no-owner --dbname "$DATABASE_URL")

rclone sync (object → object mirror)

rclone sync --fast-list --checksum --transfers 16 \
  primary:bucket/prefix secondary:bucket/prefix \
  --backup-dir secondary:bucket/_versions/$(date +%F)

systemd timer

Produce paired <name>.service (Type=oneshot) and <name>.timer (OnCalendar=...). Always include Persistent=true so missed runs fire on boot.

GitHub Actions / Kubernetes CronJob

Generate idiomatic manifests. Inject secrets via repository secrets or secretRef — never bake credentials into the script.

Hard requirements for every script

• set -euo pipefail (bash) or equivalent strict mode.
• Credentials come from env vars or a keyfile referenced from the environment. Never hard-coded.
• Exits non-zero on any failure so the scheduler records it.
• Logs to stdout/stderr; the scheduler captures. No custom logfiles in /tmp.
• A clear "dry run" path (--dry-run flag or DRY_RUN=1 env).
• A corresponding restore-command comment at the top of the file.

After generation

• Make scripts executable: chmod +x.
• Write a backup-plan/scripts/README.md listing each script, what it does, how to dry-run, and where logs land.
• Cross-link from BACKUP-STRATEGY.md → the scripts.
• Remind the user to test with --dry-run, then do a real run, then verify by listing snapshots / dumped objects.