Unresponsive VM during backups

[JulienMru]

Dear all,

We have a recurring issue where our VM become either very slow or unresponsive during backups. We have a cluster of 6 hypervisors (3 with SAS disks, 2 with SSDs, 1 with SATA disks, all servers have 3 identical disks on a LSI MegaRAID RAID5) backing up to a shared NFS mount (1 Gbps link). All hypervisors have the issue and since we created the cluster (several months ago).

The backup is configured as follow:
- Compression: LZO
- Mode: Snapshot

I agree a small freeze is required to snapshot, but this is not the case. The slowness/unresponsiveness lasts as long as the backup lasts.

We were able to get some logs from the VM during the issue, the following message is displayed (varying CPU ID/time): NMI watchdog: BUG: soft lockup - CPU#14 stuck for 22s

Any idea how to improve it?

Thanks!

 

[wolfgang]

Hi,

I agree a small freeze is required to snapshot,

I assume you are talking about KVM/Qemu VM's. There is no snapshot on file layer the snapshot happens on Qemu level.

Your problem is your NFS is to slow or/and your VM write to much.
To solve this problem limit the backup bandwidth.

See man vzdump

 

[JulienMru]

Dear Wolfgang,

Thanks for your reply. I am indeed talking about KVM/Qemu VM's.

I limited the bandwidth to 10 MBps on the SSD hypervisor and as soon as the backup starts the load peaks on the VM (but not on the hypervisor).

Best regards.

 

[mgiammarco]

Dear Wolfgang,

Thanks for your reply. I am indeed talking about KVM/Qemu VM's.

I limited the bandwidth to 10 MBps on the SSD hypervisor and as soon as the backup starts the load peaks on the VM (but not on the hypervisor).

Best regards.

After reading this thread
https://forum.proxmox.com/threads/vms-freezing-and-unreachable-when-backup-server-is-slow.96521/

I should suggest to do the opposite: NEVER limit the bandwidth and put ssd disks on backup server!