What you need
Point-in-time recovery (PITR) is part of the dedicated instance, $9 a month per project on any paid plan: the project’s own Postgres server process. The other Postgres features that live on it are described in Postgres databases. Backups on every plan are a separate, simpler feature; PITR is for when the moment you want is between two backups.
Turn it on
Choose PITR when you create the instance (canner db instance create --pg 18 --pitr) or switch it on later from the database’s point-in-time recovery panel. From then on Canner archives every change and takes a base backup on a schedule. The restorable range starts when archiving starts: you cannot recover to a moment before it was on. The panel shows the range you can restore and the health of the archive, and you get an alert if archiving starts failing.
Choose the window
The window is how far back you can go: 1 to 35 days, 7 by default. A longer window keeps more archive, and the archive counts toward your organization’s pooled storage (see plans and limits). Anything older than the window is dropped, and dropped for good. Pick the shortest window that covers how long it takes you to notice a mistake, with room to spare.
Pick a moment and restore
Choose a moment inside the window, or “latest”, and Canner recreates the database as it was then. In the dashboard you enter the time in your browser’s local time zone; the CLI takes an ISO-8601 timestamp (end it with Z for UTC). Then choose where it lands:
- In place (Restore over current data in the Backups panel, or
--in-place): the recovered data replaces the live database in one transaction. Canner snapshots the current contents first (kept 7 days), the connection string does not change, and the app keeps reading the current data until the swap commits. Tables created after the moment you picked are not part of the recovery: they are left as they are, and the result lists them. - Into a new database (the default): nothing existing is touched, and you can inspect the result before you swap it in, as below.
# Put the database back as it was at a moment (UTC), in place canner db pitr-restore orders --to 2026-09-23T14:00:00Z --in-place --yes # ...or recover into a NEW database and look first canner db pitr-restore orders --to 2026-09-23T14:00:00Z canner db pitr-restore orders --to latest
Inspect before you swap
When you recover from the dashboard you can choose Inspect first. Canner restores into a temporary inspect database with its own connection string, which is never injected into your app and deletes itself after 24 hours unless you keep or extend it. Look around with any client, or press Compare with live to see which tables are new, missing or different, with row counts. From the CLI, canner db diff <inspect> compares it and canner db inspect <inspect> --keep keeps it as a database. A wrong guess about the moment costs you nothing: recover again at a different time.
Swap it in
If you want the recovered data in place of the original, copy it over. Canner takes a snapshot of the current contents first, so this step can be undone from the snapshot. The connection string your app uses does not change.
# 2. Once you are happy with the restored copy, put it in place of the original. # --replace snapshots what was there first. canner db copy orders_restored orders --replace
A copy needs the database it writes over to be replaceable, which is why the flag is explicit. If you would rather keep both, do not swap: point a second app or a one-off script at the restored copy, and pull only the rows you need with a query.
Keep a copy off the server
The archive is compressed and encrypted with AES-256 before it is written, with a key unique to your instance. By default it lives on the same server as the instance, so it protects you from mistakes (a bad migration, an accidental delete), not from losing the server. To survive that as well, add an off-site destination, a bucket you own on any S3-compatible service (canner db offsite add …), then send the archive there:
canner db pitr --offsite <destination>
New WAL and base backups are uploaded within minutes, already encrypted, and the copy follows your window: what the window drops locally is deleted from the bucket too. If files are ever missing on the server, a restore fetches them back from the bucket first. Canner keeps the encryption key, so the bucket alone cannot be read. Shipped files count toward your own bucket, not Canner’s storage pool.
Cost
- The dedicated instance: $9 a month per project (any paid plan), including 0.5 GB of guaranteed RAM.
- Storage: the base backups and WAL count toward your organization’s pooled storage limit. Extra storage is $5 a month per 50 GB block.
- The off-site copy: whatever your own bucket provider charges.
Limits and things to know
- An in-place restore leaves tables created after the target time where they are, and names them. Drop them if the database must match that moment exactly.
- An in-place restore rewrites the database in one transaction, so its tables are locked while the swap commits. For a large database, restore into a new one and swap when it suits you.
- The window is 1 to 35 days. You cannot recover to a moment older than the window or from before PITR was turned on.
- Turning PITR off permanently deletes the archive.
- A major Postgres version upgrade of the instance starts a fresh window: after it, you can only recover to moments after the upgrade.
- If the dedicated-instance add-on lapses, the instance is stopped and its data is kept; nothing is deleted automatically.
- Canner runs in a single region (Montreal) with no automatic failover. Off-site copies are how you protect against losing the server.