Files & maintenance

Preserve receipts and storage ownership across checkpoints, rewrites and verified recovery.

On this page

Storage protection and lifecycle

The write-ahead log uses authenticated data, commits and a durable head. Segmented history rotates without changing global revisions or receipts. Checkpoints preserve a consistent state plus its tail before covered history is reclaimed.

Physical generations register record, index, TTL, receipt and catalog stores in the selected protection mode. Page roots identify committed trees; private unreachable frames can be reclaimed by a verified generation rewrite. Public memory/disk/hybrid policy selection remains unfinished.

Use the common rewrite path

Embedded Database::rewrite and ScheduledDatabase::rewrite share the generation lifecycle. Standalone qiravad rewrite --data <stopped-instance> uses configured custody and the existing instance lock. Do not run the server and a stopped-instance command against the same live directory.

A concurrent change can reject a prepared rewrite candidate. Online catch-up, resumable upgrades, rolling cluster upgrade and shard rebalance still require their full protocols and failure acceptance.

Verified backup and restore

Offline full backup/restore validates authenticated WAL, checkpoint, page and object inventories. Restore creates a fresh owner credential while preserving database revisions and receipts. Use the version-matched backup/restore runbook shipped with the binary; retain and test the independent recovery authority.

A stopped-instance backup is not a live migration or replication protocol. Broader physical power-loss, exhaustion, cold data, retention and recovery drills remain production gates.