Feature request: Atomic backup mode for single snapshot for volumes + databases

[muebau]

Problem

Currently, volumes and database dumps are backed up as separate restic snapshots. A volume backup produces one snapshot, and each database dump (restic backup --stdin) produces another. This was mentioned also in #106 . This means:

• A restore requires identifying and matching the correct volume snapshot with the correct database snapshot(s)
• There is no guarantee they represent the same point in time
• For applications like Nextcloud — where the data directory and database must be consistent — this creates a real risk of restoring into an inconsistent state

Proposed solution

A new environment variable ATOMIC_BACKUP (true/false, default false).

When enabled, the backup process changes from:

restic backup /volumes              → snapshot 1
restic backup --stdin db1.sql       → snapshot 2
restic backup --stdin db2.sql       → snapshot 3

to:

docker exec db1 dump > /databases/db1/all_databases.sql   (write to disk)
docker exec db2 dump > /databases/db2/mydb.sql            (write to disk)
restic backup /volumes /databases                          → snapshot 1 (everything)
rm -rf /databases                                          (cleanup)

The resulting snapshot contains:

/volumes/{service}/{path}/...            ← volume data
/databases/{service}/all_databases.sql   ← SQL dumps

A single restic restore retrieves everything needed to reconstruct the stack.

Example usage

backup:
  image: ghcr.io/lawndoc/stack-back
  environment:
    ATOMIC_BACKUP: "true"
    # ... other env vars

No changes to container labels needed — the feature is purely opt-in via the backup container's environment.

Behavior details

• Default off: When ATOMIC_BACKUP is unset or false, existing behavior is fully preserved (separate snapshots, --stdin piping). No breaking changes.
• Dump to disk: In atomic mode, each database container's dump command is streamed via docker exec to a file on the backup container's filesystem under /databases/.
• Single restic call: Both /volumes and /databases are passed to one restic backup invocation, producing one snapshot.
• Cleanup: The /databases/ directory is removed after the backup completes (regardless of success/failure).
• Error handling: If a database dump fails, the error is logged and tracked, but remaining dumps and the restic backup still run (consistent with current behavior where one DB failure doesn't block others). The process exits with a non-zero code if any step failed.

Why this matters

For any stack where application data and database state must be consistent (Nextcloud, Gitea, WordPress, etc.), atomic snapshots eliminate the gap between volume and database backups and make restores straightforward — one snapshot, one restore, guaranteed consistency.