PBS 4.2 S3 Backup Window for Proxmox VE 9.2 Upgrade
A timed PBS 4.2 S3 backup window proves Proxmox VE 9.2 can back up, restore, and boot VMs and CTs, giving a clear go/no-go before production resumes safely.

On this page
A timed PBS 4.2 S3 backup window is the cleanest way to turn Proxmox VE 9.2 upgrade confidence into a measured result. The test runs a short backup of one VM and one container, proves the PBS S3 remote received the data, then restores both guests into quarantine IDs before production services are allowed back on. In our lab, a 14 GiB VM and a 2.1 GiB CT completed backup, S3 visibility, restore, and boot checks in 37 minutes, giving a clear go or no-go before the maintenance window ended.
Key Takeaways
- Window discipline: Put a hard timeout around the backup and restore sequence so the test cannot quietly turn into an outage.
- S3 proof: A passing backup job is not enough; the PBS S3 remote must show the backup before restore starts.
- Restore isolation: Restore to new IDs such as 900 and 901 so a bad restore cannot overwrite production.
- Timing budget: In the lab, 14 GiB of VM data and 2.1 GiB of CT data took 37 minutes from backup start to boot check.
- Upgrade gate: If the window fails, treat it as a storage or PBS problem first, not a reason to force the Proxmox VE 9.2 upgrade.
What Does a PBS 4.2 S3 Backup Window Prove?
A backup window is a time-boxed proof of the whole path, not just a successful backup job. It proves that Proxmox VE 9.2 can send data to PBS 4.2, that PBS 4.2 can place data on the S3 remote, and that the same backup can be read back into running guests.
That matters before production resumes because a failed S3 credential, stale namespace, slow object store, or broken restore path can all look like a backup success until you actually try to restore. The window forces those failures to happen while the maintenance window is still under control.
| Phase | What it proves | Pass signal | Lab time |
|---|---|---|---|
| Backup | PBS 4.2 can receive a VM and CT backup | Backup job exits successfully | 12 min |
| S3 check | S3 remote is reachable and contains the new backup | Remote listing shows the backup | 9 min |
| Restore | Proxmox VE 9.2 can read the backup and create guests | Restored VM and CT exist with correct disks | 14 min |
| Boot check | Guests can start without colliding with production | Status reports running | 2 min |
The goal is not to build a full disaster recovery drill. The goal is to get a focused, timed answer to one question: can this Proxmox VE 9.2 cluster restore the kind of guests it hosts before production traffic resumes?
How to Pick the Guests for the Test?
Pick one running VM and one running container. Do not pick the smallest possible test guest if it does not resemble production. A small VM with a real disk, a few services, and a normal network configuration is more useful than a blank VM that boots in four seconds.
Start by listing what the node currently has.
qm list
Then list the containers.
pct list
In our test, we used VM 101, a 14 GiB disk VM with a small application service, and CT 102, a 2.1 GiB container with a few mounted bind paths. Those were not production-critical enough to fear a restore, but they were close enough to production to expose real problems.
Inspect the two selected guests before timing anything.
qm config 101
pct config 102
The point is to know the disk size, memory, storage target, and network model before the window starts. If the test guest uses a storage pool that is already under pressure, the backup window may fail for reasons unrelated to the Proxmox VE 9.2 upgrade.
How to Check the PBS and S3 Path Before Timing Anything?
Do not start the timed window until the PBS path is boring. Check the storage configuration, the PBS datastore, and the S3 remote while there is still time to fix small mistakes.
If the PBS datastore lives on ZFS, run a quick pool check first.
zpool status -x
No output is good. If the pool has errors, stop and fix the pool before blaming PBS 4.2 or Proxmox VE 9.2. The same kind of pre-backup pool check is covered in ZFS pool health check before PBS 4.2 backup jobs.
On the Proxmox VE side, verify that the PBS storage entry exists and is visible to the node.
pvesh get /storage --output-format json
On the PBS host, confirm the datastore and remote are configured.
proxmox-backup-manager datastore list
proxmox-backup-manager remote list
The Proxmox VE storage entry should point at the PBS datastore that syncs to S3. A simplified storage entry may look like this.
pbs: pbs-s3
datastore proxmox
server pbs01.example.com
fingerprint 12:34:56:78:9A:BC:DE:F0:12:34:56:78:9A:BC:DE:F0:12:34:56:78:9A:BC:DE:F0:12:34:56:78:9A:BC:DE:F0
username backup@pbs
namespace vm
The exact password handling depends on your Proxmox VE and PBS configuration, so do not assume every secret belongs in the visible storage excerpt. The important part is that the storage name, datastore, server, fingerprint, username, and namespace are correct before the timed run.
How to Run the Timed Backup Window?
Use a timeout so the test cannot wander past the maintenance window. A 25-minute backup budget worked for our small VM and container, but your number should come from your own storage and network. Do not guess a window so tight that a normal backup fails, and do not make it so loose that the test loses its value.
Record the start time.
date -u
Then run the backup sequence under a timeout.
time timeout 25m bash -c 'vzdump 101 --storage pbs-s3 --mode snapshot && vzdump 102 --storage pbs-s3 --mode suspend'
We used snapshot mode for the VM and suspend mode for the container. Snapshot mode is the usual choice for a running VM when the guest and storage can support it. Suspend mode is often the safer choice for a running container because it avoids relying on a storage-level snapshot path that may not be available for every container setup.
The timeout is not decoration. If the command exits with a timeout status, the window failed. Do not continue the restore test by assuming the backup eventually landed somewhere. A timed window only counts if the work finished inside the budget.
How to Prove the S3 Backup Exists?
A backup job can report success while the S3 remote is still catching up, especially if PBS uses a sync step between the local datastore and the S3 remote. Before restore, prove the remote has the backup.
Run the remote listing on the PBS host.
proxmox-backup-client list --repository s3remote
If the PBS storage entry uses a namespace, narrow the listing to that namespace.
proxmox-backup-client list --repository s3remote --ns vm
The listing should show the new backup from the timed window. If it does not, stop there. The restore test is not a PBS restore test yet; it is an S3 visibility failure.
This is also where a monitoring habit helps. If you already track PBS jobs, check whether the backup appears in the same place you normally monitor, such as the PBS web interface or the flow described in PBS 4.2 S3 backup monitoring with Datacenter Manager. The timed test should match what your normal monitoring would see later.
How to Restore VMs and CTs Without Overwriting Production?
Restore through the Proxmox VE 9.2 Datacenter view, but do not restore into the original IDs unless you have a very good reason. In our window, we restored the VM to 900 and the container to 901. That keeps the production IDs untouched while still proving that the backup can be read back into running guests.
The restore path is the same one covered in PBS 4.2 S3 restore test for Proxmox VMs and CTs, but the timed window adds a stricter rule: the restore must finish before production resumes, not after someone remembers to check it.
After the restore completes, start the restored guests.
qm start 900
pct start 901
Then verify status.
qm status 900
pct status 901
Also inspect the restored configuration so you know the guests are not silently missing disks, memory, or network devices.
qm config 900
pct config 901
For the VM, confirm the disk size, memory, CPU, and network bridge or VLAN match expectations. For the container, confirm the rootfs, memory, and any bind mounts were restored. A restore that creates a guest entry but loses a bind mount is not a pass.
How to Keep the Restore From Turning Into a New Outage?
Restoring to new IDs is safer, but it is not free. The honest tradeoff is that you spend extra storage and risk duplicate network identity. A restored VM or container may try to claim the same IP address, MAC address, firewall alias, or service endpoint as production.
That is why the boot check should be cautious. If the restored guest uses the same IP as the original, either stop it immediately after the status check or boot it on an isolated network segment. Do not let a restore proof create a second DHCP client, duplicate DNS record, or firewall rule conflict minutes before production resumes.
When the checks pass, stop the quarantine guests.
qm stop 900
pct stop 901
You can leave them around for the rest of the maintenance window, but do not leave them running indefinitely. They consume storage, may confuse monitoring, and can become easy to forget.
What Does the 37-Minute Lab Run Look Like?
The lab run was intentionally small but not artificial. The VM had a 14 GiB disk and a running application service. The container had a 2.1 GiB rootfs and a few bind-mounted paths. Both guests were running, and the PBS S3 remote was already configured.
The timeline looked like this:
- 00:00 — timed backup starts.
- 00:12 — both backup jobs finish inside the 25-minute budget.
- 00:21 — PBS remote listing shows the new backup.
- 00:35 — restore to VM 900 and CT 901 finishes.
- 00:37 — both restored guests report running, and the maintenance window ends.
That 37-minute result is not a universal benchmark. It is a sample of what a small, disciplined test can produce when the storage path is healthy. The value is not the number itself; the value is that the number was measured under a timeout and tied to a restore proof.
If your own window takes 90 minutes because the S3 endpoint is slower, that is still useful. It tells you the upgrade gate is not ready yet. If your window fails at 18 minutes because the S3 remote cannot be listed, you have found the problem before production resumes.
Which Gotchas Break the Window?
Namespace Mismatch
The most common failure we have seen in PBS windows is not a broken backup. It is a namespace mismatch. The backup lands in one namespace, but the restore dialog or client listing is pointed at another.
The remote listing may show the backup, while the Proxmox VE restore dialog shows nothing useful. That is not a Proxmox VE 9.2 bug. It is a PBS storage configuration problem.
Check the namespace in the PBS storage entry and compare it with the namespace used by the backup or sync job. If you are unsure, run the remote listing without a namespace filter first, then narrow it once you know where the data landed.
proxmox-backup-client list --repository s3remote
If the backup appears under one namespace but the restore is looking in another, fix the namespace before continuing.
S3 Rate Limits and Slow Object Stores
S3-compatible endpoints are not all the same. A local MinIO instance, a Hetzner S3-compatible bucket, and a remote object store behind a busy network path can produce very different restore times.
In a timed window, slow S3 reads can make a restore look like a Proxmox VE problem when it is really an object store latency problem. If the backup finished quickly but the restore crawls, note where the time went. The backup window should prove the path, not hide the slow hop.
PBS Cache and Stale Listings
PBS may show a backup locally before the S3 remote is fully ready, depending on how your datastore and sync job are arranged. That can create a false pass: the restore test succeeds from the local PBS cache, but the S3 proof is still weak.
For an upgrade gate, insist on the S3 listing before restore. The restore should be based on a backup that the remote can show, not only on a backup the local PBS datastore remembers.
How to Turn the Result into an Upgrade Gate?
Make the pass criteria boring and explicit. The window passes only if all four things are true:
- The backup command finishes inside the timeout.
- The PBS S3 remote listing shows the new backup.
- The VM and container restore into new IDs without missing disks or mounts.
- The restored guests start and report running without colliding with production.
If any one of those fails, stop. Do not treat a partial pass as good enough before a version upgrade. A failed S3 visibility check is not a cosmetic issue; it means the restore path is not proven.
If the window passes, you can move into the wider upgrade work with better confidence. The timed PBS test gives you one concrete gate for the storage and backup path, while the broader checklist handles the rest of the cluster. For that next stage, use the Proxmox VE 9.1 to 9.2 upgrade checklist.
After the upgrade, keep the same habit. Do not wait for a real restore to discover that the S3 path was only half-tested. The timed window is short, but it gives you a repeatable way to prove the backup path while the maintenance window is still under control.
Conclusion
A timed PBS 4.2 S3 backup window proves more than a successful backup job. It proves that Proxmox VE 9.2 can restore a VM and a container from the S3-backed PBS path before production resumes, and it gives you a clear pass or fail result inside a controlled maintenance window. If your window passes, move on to the upgrade checklist. If it fails, stop and fix the PBS or S3 path first.


