Volume consistency groups
Coordinate recovery points for applications that write across several volumes.
A consistency group defines the set of volumes that must participate in one application recovery point. It prevents a multi-disk restore from silently combining unrelated captures.
Define membership
Create a group under Storage → Consistency groups. Add explicit volume IDs and identify required members. Members must belong to one project and region. A group cannot include a volume whose storage policy lacks group capture support.
The group revision pins membership for each capture. Adding a volume while a capture runs affects the next attempt, not the already resolved member set. Deleting a required member marks future captures blocked until the group is updated; it does not rewrite old completed manifests.
Select the consistency boundary
Crash consistency captures the participating block devices without application coordination. Guest-assisted consistency uses a supported agent to freeze the relevant filesystems. Application consistency additionally runs the configured prepare and resume hooks.
A group spanning multiple guests needs an application-level coordination method. Independently freezing two database servers does not establish a shared transaction boundary. The group records the requested method and the result for each participant so the restore operator can assess it.
Capture execution
Infra resolves the group revision, checks required members, runs preparation, requests capture, and records a group manifest. Each resulting snapshot includes the group ID and capture-attempt ID. A complete manifest is offered for restore only after every required member reaches its usable state.
If a required member fails, the attempt is incomplete. Successful member snapshots can remain for diagnosis and explicit single-volume use, but the console does not label them a completed application recovery point. Resume hooks run for participants that entered preparation, including when another member fails.
Restore a group
Select Restore group to create new volumes from one complete manifest. The restore view maps source IDs to destination IDs and keeps attachment roles separate from device names. It does not attach the restored disks to running production guests automatically.
Validate all restored members together on isolated recovery capacity. Starting an application with one disk from a newer capture can violate its recovery assumptions even when each disk is individually readable.
Membership removal and retention
Removing a volume from the live group leaves completed group manifests intact. Retention evaluates each manifest as a recovery unit; a retained complete group keeps every required member. An administrator cannot delete one required snapshot and continue presenting that manifest as complete.
See Snapshot API for capture requests, backup retention for independent-copy retention, and disaster recovery for application promotion.