live migration with discard enabled

[Aleksej]

Hello.
I have troubles with live migration if "discard=on" for VM.
For some time it halts overall system with IO delay 90% and then migration starts normally.
it seems that proxmox at first moving something (the whole disk size), i can see it by network monitor, and than move disk as usual.

I have 1Gbps connection, lve migration takes about 10 minutes of 50G VM disk and about 20 minutes with "discard=on". So via network traffic flows twice.
Maybe i'm doing something wrong?
I tried discard=on on virtio, scsi - the same.
Also i noticed if VM is newly created on host data on onderlying SSD is used as with discard=on, but after live migration or restoring backup it shows 100% usage until i do not poweroff VM and enable discard.

I'm using up-to-date PVE (pve-manager/6.4-13/9f411e79 (running kernel: 5.4.157-1-pve)

 

[spirit]

What is your storage ? Do you have tried with proxmox 7 to compare ?

 

[Aleksej]

I tried with LVM, LVM-thin (the most i use) storage and simple directory (qcow2)

No, with proxmox 7 i didn't try.
I have this issue from old proxmox 5...
I googled about high IO delay but there is a lot of threads but nothing about that...

 

[sg90]

If you have discard enabled it will zero the whole new disk first, hence the delay / high I/O time.

If you can your better to disable discard, reboot and then migrate.

This is a current limitation of QEMU and not PVE