Read-only VM with stdout

[rafael.silva]

Hi,

I backup some VMs using SSH with vzdump redirecting to stdout, similar to the command below:
ssh root@pve-server vzdump 100 --mode snapshot $VZDUMP_PARAMETERS --stdout > ${SERVER}_${NAME}_$(date +%Y-%m-%d).vma 2> >(tee -a ${SERVER}_${NAME}_$(date +%Y-%m-%d).log >&2)

But I noticed that if the connection goes offline, the machine being dumped starts to show erros and becomes read-only system.

Is it possible to prevent this from happening?

Thanks.

 

[LnxBil]

Is it possible to prevent this from happening?

I asked this myself. I also encountered this with an external NFS server that recently crashed a lot. I often got read only disks that are caused by a blocked/hanging backup.

 

[bbgeek17]

This is because backup is implemented as a shim for write activity. During the backup if client/VM tries to write to disk and that block has not been backed up yet (based on dirty bitmap), PVE will try to backup that block out of order to get a consistent backup view.
In essence, backup connection is in critical path of a write. With IO blocked on write the OS/Kernel will wait for a bit and after a timeout convert the disk to RO as a precaution.
The only short term solution is to have a rock solid backup server and connectivity to it. The ideal solution would be to offload saving the blocks to snapshots on the backend storage.

---

Blockbridge: Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox