Repository navigation
v0.29.0
Restore on Postgres, and the database choice made final
v0.29.0 brings restore to Postgres installs, makes the database an install runs on a choice made once, and fixes a SQLite bug that could fail batches under steady writes. Postgres support is still in beta.
New
- Restore from the backups page on Postgres, while the server keeps running, from a stored backup or an uploaded
.dump. The backup is loaded into a schema of its own, checked, then swapped in; writes wait for the swap on every server sharing the database. A damaged archive is refused with nothing changed. Sessions revoked since the backup stay revoked. Needspg_restoreandpsqlfrom a PostgreSQL client of August 2025 or later. - The data a restore replaces is kept for 7 days, under Previous data on the backups page, with Put back (an undo) and Delete.
vaultbase backup,doctorandwipework on Postgres.backupwrites the samepg_dumparchive as the backups page.doctorchecks the server version, the schema and search path, the client tools, and hooks that usehelpers.db.wipedrops only Vaultbase's own objects from the schema and changes nothing when anything else depends on them.- The Docker image includes
pg_dump,pg_restoreandpsql(PostgreSQL 18 client), so backups and restore work on Postgres in a container.
Changed
- The database is chosen at install and is final. Once an install has its first admin, the data directory records its database in
database.json, and the server refuses to start on another one: the other kind, an empty one, or another install's. Before, a forgottenVAULTBASE_DATABASE_URLstarted an empty SQLite server with the setup wizard open. - Postgres: code that uses
helpers.dbis refused when saved or enabled, with a422naminghelpers.sql. It could never run on Postgres; before, it saved and then failed every time it ran.
Fixed
- SQLite: batches could fail with
500 database is lockedunder steady writes, and so could field changes, renames, deletes that cascade and the first search index on a collection. They now wait for the lock. - Postgres: long collection and field names lost indexes (Postgres cuts names to 63 bytes), and a schema whose name needs quoting broke search. Existing installs are repaired at their first start.
- Postgres: every start re-ran the schema's CREATE statements, which could stall a server starting next to busy ones.
See the changelog for everything.
Upgrade notes
- An existing install records its database at its first start on this release. Make sure that start goes to the database you mean.
- On Postgres, run
vaultbase doctorafter upgrading: it lists enabled code that useshelpers.db, which now can't be saved or re-enabled until it useshelpers.sql. - Upgrade notes for earlier releases apply if you are coming from before v0.28.0.
Full Changelog: v0.28.1...v0.29.0