← All break/fix scenarios in Proxmox VE
Backup job runs for 12 hours instead of 30 minutes
Reported symptoms
- ●PBS backup jobs normally take 30 minutes; now they take 12+ hours
- ●PBS task log shows "slow chunk upload" or timeout retries
- ●VM console response is sluggish during the backup
- ●Storage datastore shows free space decreasing rapidly during backup
Evidence
- · pvesh get /cluster/tasks --typefilter vzdump shows the task duration and throughput
- · vmstat 1 shows high iowait (wa column > 30%)
- · iostat -xz 1 shows the backup target disk at 100% utilisation
- · The VM has changed (more RAM, more disks, or a new workload)
Diagnosis and resolutionclick to reveal
Root cause
Several possible causes, in order of likelihood: 1. The VM's working set exceeds the previous backup size; PBS must read every dirty block 2. The PBS datastore is on slow storage (spinning disk, network) 3. The network between PVE and PBS is saturated 4. The backup is using "stop" or "suspend" mode, freezing the VM while waiting on slow storage
Remediation
1. Verify the backup mode: `pvesh get /nodes/<node>/vzdump/<task-id> --output-format json-pretty` Look for "mode: snapshot" (best) vs "stop" or "suspend" 2. Check the VM's dirty block rate: `pvesh get /nodes/<node>/qemu/<vmid>/status/current | jq '.ballooninfo'` Compare with previous backups to see growth 3. Check PBS datastore performance: `iostat -xz 1` on the PBS host (or the underlying disk) 4. Check the network path: `iperf3 -c <pbs-host>` from the PVE host 5. Tune the backup to reduce impact: - Switch to snapshot mode if not already - Set bandwidth limit: `--bwlimit 50000` (KB/s) - Increase PBS chunk size for faster dedup (cautious) 6. If the VM is the bottleneck: - Pause application I/O during backup (if tolerable) - Schedule backup during low-activity hours - Use a separate, faster disk for backup target
Verification
- Backup duration returns to near-baseline (within 2x) - PBS task log shows expected throughput (no timeouts) - VM remains responsive during backup (console ping latency < 100ms) - Datastore utilisation drops back to expected rate
Prevention
- Monitor backup duration trend over time - Alert when backup duration exceeds 2x baseline - Use snapshot mode exclusively for VMs - Size the PBS datastore for the VM growth rate - Periodically test restore (a fast backup is useless if restore is slow)