Skip to main content
Each database has a recovery policy for point-in-time, daily, weekly, and monthly retention. Inspect the recovery window before promising an RPO, and test restores instead of treating backup success as proof of recoverability. The primary managed repository requests best-effort placement near Western North America. Add copyRegions to request independently restorable managed snapshots near Eastern North America, Western Europe, Eastern Europe, Asia-Pacific, or Oceania. Area placement is best effort and is not a data-residency guarantee. Korve records each copy as a separate recovery point with its requested area and status, retries it through the durable backup worker, and removes its exact repository objects when the recovery point or database is deleted. Regional copies are periodic snapshots with a target RPO of six hours; continuous WAL-based point-in-time recovery remains in the primary repository. The primary PITR stream and discrete daily, weekly, monthly, or manual recovery points use separate repository prefixes. Bounded PITR housekeeping therefore cannot expire a longer-lived discrete point, whose exact label is removed only when its own retention ends. Regional copy storage is billed after a copy completes. Network egress uses the backup payload bytes reported by the database engine and excludes transport protocol and WAL overhead. Korve provisions a distinct managed repository with the selected best-effort area hint. Production uses one account-scoped, least-privilege object-storage credential for the primary and regional repositories; deployment proves read, write, and delete access to every required repository before the control plane is rolled out.
An on-demand recovery point accepts exactly one placement: omit both placement fields for the primary repository, provide copyRegion for one available managed area, or provide destinationId for one ready customer destination. Managed regional and customer-destination copies can incur transfer usage according to the destination’s billing policy. Korve bills stored backup bytes for managed repositories; storage in a customer destination is billed by that destination. The response records the selected copyRegion or destinationId, so API, CLI, MCP, and dashboard callers can track the exact repository independently. Restores are asynchronous and create an isolated target. Promotion is a separate two-step operation: preview the binding swap to receive a one-use confirmation token, then submit the exact token and binding with explicit confirmation. Do not place the token in shell history; pass --data @file.json. Portable exports can target an organization-owned compatible object destination. Test the destination before starting an export, and account for transfer charges. See Backup destinations, exports, and restores.