vzdump appears to be slowing down VM IO

[SagnikS]

When trying to backup a large VM to a remote server using vzdump (snapshot mode), the writes on the VM appear to be slowed down greatly. Are disk writes affected (maybe writes are slowed down during a backup) when a backup is in progress, to maintain consistency?

 

[leesteken]

When trying to backup a large VM to a remote server using vzdump (snapshot mode), the writes on the VM appear to be slowed down greatly. Are disk writes affected (maybe writes are slowed down during a backup) when a backup is in progress, to maintain consistency?

Yes, it need a consistent filesystem state, which often requires QEMU Guest Agent to be running. Backup to a local Proxmox Backup Server (and sync to a remote one separately) if you want less slow down.

 

[Darkk]

I back up several VMs few times a day to the PBS server without much of a slowdown regardless of VM size. The storage is all ZFS with scheduled replication. If you're using CEPH yes it will slow the VMs down a bit during backups.

 

[SagnikS]

My storage is regular raw files on a EXT4 partition. I'm primarily curious if disk writes are hijacked to write to both the backup and the actual disk, as limited bandwidth seems to aggravate this.

 

[Darkk]

My storage is regular raw files on a EXT4 partition. I'm primarily curious if disk writes are hijacked to write to both the backup and the actual disk, as limited bandwidth seems to aggravate this.

Putting them on ZFS storage will help a great deal as it will make use of ZFS snapshots. And then the backups will save the snapshots without affecting the VMs much. So give that a try. Also, definitely make sure qemu agent is installed as it will coordinate the internal snapshot of the VM during backups.

 

[leesteken]

I'm primarily curious if disk writes are hijacked to write to both the backup and the actual disk

Yes, to do a consistent backup, new writes must not go into the backup and therefore need to be saved somewhere else.

Are disk writes affected (maybe writes are slowed down during a backup) when a backup is in progress, to maintain consistency?

Yes, the new writes needs to be delayed and buffered in order to do a consistent backup of the old data.
In short: Yes, you are right. Writes are diverted to ensure consistency of the backup. And yes, this will probably cause slowdowns. Whether it be because of increased memory usage or because data needs to be stored separately instead of overwriting existing data.

 

[bbgeek17]

https://git.proxmox.com/?p=pve-qemu.git;a=blob_plain;f=backup.txt

 The following steps are always required:

1.) read old data before it gets overwritten
2.) write that data into the backup archive
3.) write new data (VM write)

---

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