Serverless Databases
Run whichever database you already use, and start, clone and restore it in milliseconds.
Branch a running database (memory and volume state included) and run the risky migration against the copy.
Checkpoints keep a history of full instance state, so rolling back is a restore rather than a rebuild.
State is snapshotted when the database goes quiet and restored on the next query. Per-tenant databases stop being expensive.
The only thing that can reach the database
Whichever database engine you run: memory and volume state, live
A full clone, memory included; run the risky change here
Watch the recording to see a running Postgres branched in 40 ms, written to without touching the original, and either copy woken in about 16 ms.
Under the hood
The platform features this use case is built on.
Clones a running database instance, including memory and volume state, with at most a few milliseconds of downtime on the original.
Capture the full state of an instance and keep a history, so you can restore from a known good snapshot.
Snapshots memory and volume state when the database is idle and restores it in milliseconds on the next request. Pay only for what you use.
The database can be hidden from the public internet entirely, reachable only by your web server.
Sign up to Unikraft Cloud and follow the guide to run your first serverless database