Checkpoints і rollback (preview)
Checkpoint service defines records, retention behavior, REST/MCP operations, rollback state machine for apps.
No real recovery yet: active snapshot adapter fabricates VM and database snapshot identifiers. Stop/start/health stages simulated, no disk or database state captured or restored. Checkpoint created today is not backup and must not be used as recovery control.
What can be tested
Preview surface lets integration developers exercise:
- manual checkpoint creation and listing;
- pin, unpin, delete, retention metadata;
- ownership checks;
- rollback state transitions; and
- creation of pre-rollback safety record.
Following MCP tools expose that control plane:
| Tool | Preview behavior |
|---|---|
create_checkpoint | Creates record with stub snapshot identifiers |
list_checkpoints | Lists owned records and reported usage |
pin_checkpoint | Exempts record from retention pruning |
delete_checkpoint | Deletes unpinned record |
rollback_app | Exercises simulated rollback sequence |
Do not describe successful response from these tools as recovered app.
Production recovery today
For immutable app deployment, recover from source and separately managed data:
- identify known-good Git revision;
- confirm database or object-store backup exists and can be restored;
- queue new immutable deployment for that revision;
- restore data through service documented recovery procedure; and
- verify application health and user-visible behavior before switching traffic.
Keep backup retention, restore testing, recovery objectives outside checkpoint preview until real adapters shipped and passed restore drill.
What remains
- real VM and database snapshot primitives;
- rollback integration with immutable variant deployment pipeline;
- failure-atomic restore behavior;
- deploy hooks that create verified checkpoints;
- capacity and retention enforcement against real storage; and
- backup/restore drills with captured evidence.