Backup, restore and consistency checks

A backup is a directory that opens as a store. It holds the last commit before the backup started, and is written while queries and transactions keep running and committing (it reads a snapshot, copying pages a batch at a time without filling the page cache). Persistent databases back up as persistent stores (their pages and metadata); in-memory and hybrid ones as hybrid stores (a snapshot file).

graphbuff backup  --data data --database social --to backups/social-monday
graphbuff restore --from backups/social-monday --data data --database social --overwrite
graphbuff check   --data data --database social          # exit code 1 if inconsistent
graphbuff check   --store some/persistent/store

• Online: the commands above open the data directory, so the server must not be running on it. A running server does the same over HTTP when started with --backup-dir DIR (admins only): POST /db/{name}/backup with {"backup": "monday"} writes DIR/monday, POST /db/{name}/restore with {"backup": "monday", "overwrite": false} creates (or replaces) the database from it, and POST /db/{name}/check checks a persistent database.
• Restore copies the backup (it stays usable) and checks a persistent one before the database is created. An existing database is replaced only if it is empty or with overwrite, and stays as it was if the restore fails.
• Check (graphbuff check, persistent storage only) reads every record and verifies adjacency lists in both directions (links, list ends, endpoints, relationship groups of dense nodes, degrees and self-loop counts), node, relationship, label and type counts, blob chains and free lists, every B+tree (page use, key order, leaf links) against the entries the records call for, index counts, and constraints. Lost space (blocks neither used nor free) is a warning. 200,000 nodes and 400,000 relationships: check 1.0 s, backup 0.4 s (102 MiB).
• Safety: a backup goes into a new or empty directory only, and holds a BACKUP_INCOMPLETE marker until it is complete; stores refuse to open (and restore refuses to read) a directory that still has one.
• From Rust: Database::backup / Database::check, DatabaseManager::backup_database / restore_database / check_database, and GraphStore::backup / GraphStore::check (call them on a snapshot for an online run).

$ docker run --rm -p 127.0.0.1:7475:7475 -p 127.0.0.1:7687:7687 ghcr.io/janvencl/graphbuff --demo movies