swiss-ai-hub-backup
The centralized backup, restore, and PostgreSQL-maintenance service for Swiss AI Hub — a self-contained Dagster instance that snapshots every stateful service to S3.
What is Swiss AI Hub?
Swiss AI Hub is an open-source, self-hosted AI platform for enterprises. One
docker compose up starts ~30 integrated containers across several stateful stores — PostgreSQL, FerretDB, Milvus,
Neo4j, ClickHouse, Valkey, and NATS JetStream. This package keeps that data safe.
What is this package?
swiss-ai-hub-backup is the platform's backup/restore and database-maintenance plane. It runs as its own
independent Dagster instance (separate from the data pipelines, with its own SQLite storage) and:
- Backs up PostgreSQL (×2), Milvus, Neo4j, ClickHouse, Valkey, and NATS JetStream to S3 (SeaweedFS) on a schedule — gracefully stopping and restarting the managed containers around each run for consistent snapshots.
- Restores any service from a chosen backup timestamp.
- Maintains the platform PostgreSQL online: prunes verbose Dagster
event_logs, tunes autovacuum, and runspg_repack— so deployments stay bounded over time without downtime.
Each stateful service has a BackupHandler (postgres, milvus, neo4j, clickhouse, valkey, nats); the whole
thing is wired into a Dagster asset graph by backup_definitions(). Because it operates on the storage layer and needs
to stop containers, it requires read access to the Docker socket (/var/run/docker.sock), which it uses to discover
platform containers via their com.docker.compose.project label.
Unlike the other Swiss AI Hub packages, this is an operational service, not a library you build agents/APIs on. It is licensed AGPL-3.0-or-later (the rest of the SDK is Apache-2.0).
Should you use this package?
Most operators don't install it directly — it ships with the platform as the backup-* containers (a gRPC code
server, a daemon, and a webserver UI on :3004). You'd reach for this PyPI package to run the backup plane
standalone, embed its logic, or extend it — for example, adding a BackupHandler for a stateful service of your own.
What it does
| Job | Schedule | Stops containers? |
|---|---|---|
| Full backup (all services → S3) | daily | Yes (consistent snapshots) |
| Restore (service ← chosen timestamp) | on demand | Yes |
event_logs cleanup + autovacuum tuning |
weekly | No (online-safe) |
pg_repack (reclaim disk) |
monthly | No (online-safe) |
Installation
pip install swiss-ai-hub-backup
# or
uv add swiss-ai-hub-backup
Requires Python 3.13.
Quick start
The backup plane is a Dagster code location built by backup_definitions():
# my_backup/__init__.py
from swiss_ai_hub.backup.dagster.definitions import backup_definitions
defs = backup_definitions() # 26 assets, 4 jobs: backup, restore, cleanup, repack
Inspect and run it with the Dagster UI (it keeps its own state in DAGSTER_HOME):
export DAGSTER_HOME=/tmp/backup-dagster && mkdir -p "$DAGSTER_HOME"
set -a && source .env && set +a # S3 + DB credentials, BACKUP_* settings
dagster dev -m my_backup # http://localhost:3000
From the UI you can materialize the online-safe maintenance jobs (cleanup, pg_repack) against a running stack
without disruption. The full backup/restore jobs stop and restart containers, so run those deliberately — and note
they need access to the Docker socket and to all the stateful services. dagster definitions validate -m my_backup
loads the whole code location without running anything (a fast CI/sanity check).
Settings are not auto-loaded from the environment. Connection and
BACKUP_*settings are read only when constructed, so export them in the process that runs Dagster (set -a && source .env && set +a).
How it's deployed
In production the backup plane runs as three containers from one image, forming a self-contained Dagster instance:
| Container | Role | Notes |
|---|---|---|
backup-code |
Dagster gRPC code server (dagster api grpc … :4266) |
mounts /var/run/docker.sock:ro to stop/start containers; on data + storage |
backup-daemon |
Dagster daemon | runs the schedules and sensors; on data |
backup-webserver |
Dagster UI (:3004) |
inspect runs, trigger restores; on proxy + data |
Because it needs the Docker socket and the platform's stateful services, the canonical deployment is the platform's own
backup compose. See infra/deployment and the
documentation for the full container setup, retention config, and the
BACKUP_* environment variables. If you run your own variant, mirror that three-container shape and grant the code
server read access to the Docker socket.
Extending — add a service to back up
Implement the BackupHandler ABC for your service in services/, then register it in HANDLER_FACTORIES — the Dagster
asset wiring picks it up automatically (handlers are synchronous by design):
from swiss_ai_hub.backup.services.base import BackupHandler
class MyServiceHandler(BackupHandler):
def backup(self, context) -> ...: ... # dump your service's state to S3
def restore(self, context) -> ...: ... # restore it from a backup
If the handler needs Docker access, type-hint a DockerManager parameter in __init__ and the factory injects it. The
maintenance subsystem follows the same pattern (MaintenanceHandler + CLEANUP_HANDLER_NAMES). See the
documentation for the full handler contract.
Links
- Source & issues: https://github.com/bbvch-ai/aihub-core
- Documentation: https://bbvch-ai.github.io/aihub-core/
License
AGPL-3.0-or-later — see packages/backup/LICENSE. Note this differs from the Apache-2.0 SDK packages; for the full per-package license matrix, see LICENSES.md.
Part of Swiss AI Hub. Built in Switzerland by bbv Software Services.
2 maintainers
Metadata
2 maintainers
Release files for swiss-ai-hub-backup 0.323.4
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| swiss_ai_hub_backup-0.323.4.tar.gz | 55.4 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| swiss_ai_hub_backup-0.323.4-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 130.5 kB
Release files / swiss_ai_hub_backup-0.323.4.tar.gz
| Download URL | swiss_ai_hub_backup-0.323.4.tar.gz |
|---|---|
| Size | 55.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
92db7466df3e81e6a62843500ff5f572f928aee1bb7be664d105391545d21e12
|
|
BLAKE2b-256 checksum How to use checksums |
9503e9db60afae758609ec61f7d0c6528dc7a3c202e6f9e23af2b80473fc4335
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.13.0 {"installer":{"name":"uv","version":"0.13.0","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / swiss_ai_hub_backup-0.323.4-py3-none-any.whl
| Download URL | swiss_ai_hub_backup-0.323.4-py3-none-any.whl |
|---|---|
| Size | 75.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
8f3405e869fe998aee15b5ef8e025ebfedf0bceb56605f8ac8806f43dfd233d2
|
|
BLAKE2b-256 checksum How to use checksums |
50abd9416d05327438c116b1f5c80c08ada88dee28bac773e0903b3415233035
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.13.0 {"installer":{"name":"uv","version":"0.13.0","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
