Duplicati can now restore specific PostgreSQL tables
Duplicati 2.4.1.103 adds selective PostgreSQL table backup and restore, which is useful for self-hosted Laravel applications.
The real issue is recovering only what is needed
When a Laravel application runs on PostgreSQL, an incident does not always justify restoring the whole database. The problem may be isolated to one table, a bad import, or data that needs checking before it goes back into production. Restoring everything in that situation adds risk, time, and unnecessary work.
Duplicati 2.4.1.103 Canary, released on November 20, 2026, adds PostgreSQL as a backup source. The useful part is that it can store each table as a separate entry in the backup. That makes it possible to restore selected tables instead of treating the database as one indivisible unit.
A more practical restore path for self-hosted Laravel
For a Laravel project I run in a homelab, this fits well with backups that already separate application code, persistent files, and the database. I can continue backing up the application directory and its volumes while giving PostgreSQL a level of granularity that previously required custom pg_dump workflows. It does not replace a full recovery strategy, but it makes smaller and more isolated incidents easier to handle.
Selective restoration is enabled with --postgresql-restore-mode=tables. According to the release documentation, this mode restores the rows from selected tables while they are being recovered, without writing temporary data the size of the entire database to disk. That matters on smaller machines, where free local storage is usually more limited than remote backup capacity.
This release supports PostgreSQL versions 14 through 18. It also requires the client programs matching the server version to be installed, unless their location is supplied through --postgresql-client-path. Before relying on automation, I would check that detail inside the container or host running Duplicati, not only on the machine hosting PostgreSQL.
A small configuration worth documenting
These are examples of options available in this release. They are not a complete backup job by themselves: the connection, source, and destination still need to be configured in Duplicati. The point is to keep the backup limits and dependencies documented alongside the actual job.
--postgresql-client-path=/usr/lib/postgresql/16/bin
--postgresql-max-database-size=20GB
--postgresql-restore-mode=tables--postgresql-max-database-size can stop a backup when the database has grown beyond the expected limit. It is not an integrity check and it does not replace storage monitoring, but it can prevent a silently growing database from turning the backup window into another problem. If PostgreSQL uses verified TLS, Duplicati can also receive the root certificate through --postgresql-sslrootcert-content.
Ordering backup jobs without another orchestrator
The same release can run one backup after another, either only when the previous job succeeds or regardless of the outcome. In a homelab, ordering can be useful when jobs share network bandwidth, local disks, or the same remote destination. For example, I can run the PostgreSQL backup first and then collect files uploaded by the application.
Duplicati can also start several jobs from one backup and refuses loops in those chains. That does not make it a general-purpose workflow engine, but it removes the need for scripts whose only purpose was to enforce a simple order. I would keep these chains short and clearly named, because hidden dependencies make restores harder than independent schedules.
Once the backup queue is empty, the server can shut down or restart the computer, or run the script configured in post-queue-processing-script. That is useful for an external verification step, a custom notification, or powering down a backup node that only runs during that window. The script should be idempotent and should understand that an empty queue does not necessarily mean every job produced the desired result.
The fine print before adopting it
This is a Canary release intended for testing. I would not install it directly on the node protecting the only useful copy of a production database. I would test it with a real restore into an isolated PostgreSQL instance, using representative data and checking that the application works with the restored result.
Restoring selected tables does not automatically solve table dependencies, foreign keys, sequences, or the assumptions the application makes about the complete data set. If one table depends on another that was not restored, the result may be incomplete or inconsistent for Laravel. Before using this during an incident, I would define which tables must be restored together and validate that set in a test environment.
A granular restore should not be confused with a complete continuity plan either. I still need verified full backups of the database, the .env file, secrets, user uploads, and deployment configuration. Recovering a table is useful; recovering the application is what matters.
My take
Adding PostgreSQL as a backup source and restore destination makes Duplicati more compelling for self-hosted Laravel applications. Selecting tables and avoiding temporary storage equal to the full database size addresses a concrete operational limitation, not a feature-list novelty. Combined with backup chains and actions after the queue finishes, it can simplify a homelab without adding another automation service for basic workflows.
The responsible part remains the same: test restores, retain a full backup, and do not confuse Canary with stable. If it works with my tables, PostgreSQL version, and remote destination, this is worth adding to the backup plan. If a recovery has not been tested, it is not a plan yet; it is only hope.
Source: https://forum.duplicati.com/t/release-2-4-1-103-canary-2026-11-20/23000 Official documentation: https://github.com/duplicati/duplicati/releases/tag/v2.4.1.103_canary_2026-11-20