The target services are Proxmox Backup Server for architecture and operations, and Proxmox on Hetzner for the provider environment. Arrange a meeting.
PBS has a native S3 backend; Hetzner provides an S3-compatible API. This does not establish suitability for every backup workload. As of October 2026.
Contents
- Version and architecture
- Prepare bucket and access
- Create S3 endpoint and datastore
- Check limits and acceptance
- Recovery and alternatives
- Sources
Version and architecture
The S3 backend graduated from technology preview with PBS 4.2 (release notes). PVE still backs up through PBS rather than directly to a bucket. PBS setup remains a prerequisite.
Hetzner Object Storage and Storage Box are different products with different interfaces (Hetzner storage comparison). For file-based Storage Box backups, use the separate gocryptfs guide.
Prepare bucket and access
- Create a separate private bucket in the intended region (create a bucket).
- Generate S3 credentials and check key scope. Hetzner keys apply project-wide by default, rather than to one bucket (S3 keys). Separate projects and credentials accordingly.
- Obtain region and endpoint from the current Hetzner overview, for example
fsn1andfsn1.your-objectstorage.com. Location values must match.
For this procedure, leave Object Lock disabled and do not apply external lifecycle deletion rules to PBS objects. This is a conservative operating rule derived from PBS deletion needs and its GC process, not a claim of a tested Hetzner malfunction. Do not reconfigure an existing locked or production bucket for this purpose.
Create S3 endpoint and datastore
In PBS, open Configuration > Remotes > S3 Endpoints and configure the HTTPS endpoint with region and credentials. For the fsn1 example, enter {{bucket}}.fsn1.your-objectstorage.com in Endpoint and leave path-style disabled (virtual-host addressing). Specify the actual bucket name later in the datastore configuration; PBS substitutes it for {{bucket}} in the hostname. The region endpoint alone without the bucket placeholder is insufficient for this addressing style. PBS needs get, put, list and delete permissions. Then create a new datastore with the S3 backend, bucket and its own persistent local cache. An existing local datastore cannot simply be reused for this (PBS S3 configuration). Manage credentials exclusively in the intended configuration fields.
Then connect the new PBS datastore in PVE. Configure client-side backup encryption separately; secure bucket credentials and encryption keys independently.
Check limits and acceptance
Hetzner describes HDD-based storage and limitations for high request rates and latency-sensitive workloads. Its FAQ also describes possible bottlenecks and timeouts (workload suitability, checked 18 October 2026). PBS vendor support status therefore does not replace acceptance testing of this combination.
| Check | Record |
|---|---|
| Representative backup | Data scope, final job status, errors and retries |
| Verify and restore | Backup snapshot read, key access, isolated application findings |
| Retention and GC | Expected retention, deletion errors, storage usage development |
| Recovery | Access without original PBS, configuration and cache recreated |
| Operating costs | Bucket usage, outgoing traffic and local PBS/cache resources |
The Hetzner product documentation lists storage and additional outgoing traffic among charges; S3 operations themselves are currently free. This does not establish a fixed total cost. Enter results and measured durations only after actually performing the checks.
Recovery and alternatives
Only one PBS instance may operate the same S3 datastore. Taking over requires stopping the original instance and correctly setting the datastore name and reuse options (PBS recovery documentation). Include this in a separate test plan rather than accessing the datastore concurrently during normal operations.
An alternative is a second PBS with its own datastore and offsite sync. Base the decision on documented restore test results and your recovery objectives.
PBS planning and operations and Proxmox on Hetzner establish the agreed scope. This guide promises no provider performance or recovery duration.
