I'm preparing migration from Xen to Proxmox. Currently I'm testing Proxmox migration features - both live and offline. How does it actually work?
In test environment I prepared 2 clustered Proxmox nodes with separate local storages LVM-thin configured.
root@proxmox-nd1:~# pveversion
pve-manager/5.2-1/0fcd7879 (running kernel: 4.15.17-1-pve)
1. Offline migration
I've created VM with 10 GB (LVM-thin provisioning). When I issue lvs command in proxmox I can see that 20% of virtual disk is being used. This corresponds with disk usage checked with df -h on guest host. I turn off the guest and run migration. There is 2 GB of data, but the migration last about 90 seconds. Between nodes there is 1 GB ethernet link, and when I scp between them I got 112-113 MB/s. ~90 seconds seems like all 10GB of "data" is being transfered. Also during migration when I issue every couple of seconds lvs command I can see how Data% indicator increases. It's not linear. It grows fast and stops for some time, and again grows. Is really "empty" space transfered?
2. Online migration issue.
It seems to be slower than 1GB/s but I found that encryption is causing this. I will test the difference later. Because I use live migration with separate storage i need to use CLI and issue:
qm migrate 100 proxmox-nd2 --online --with-local-disks
Migration works fine but the disk is no longer thin-provisioned. Data% indicator in lvs command output shows 100% and I can confirm this with GUI.
Is there a way to bypass this and schrink guest?
In test environment I prepared 2 clustered Proxmox nodes with separate local storages LVM-thin configured.
root@proxmox-nd1:~# pveversion
pve-manager/5.2-1/0fcd7879 (running kernel: 4.15.17-1-pve)
1. Offline migration
I've created VM with 10 GB (LVM-thin provisioning). When I issue lvs command in proxmox I can see that 20% of virtual disk is being used. This corresponds with disk usage checked with df -h on guest host. I turn off the guest and run migration. There is 2 GB of data, but the migration last about 90 seconds. Between nodes there is 1 GB ethernet link, and when I scp between them I got 112-113 MB/s. ~90 seconds seems like all 10GB of "data" is being transfered. Also during migration when I issue every couple of seconds lvs command I can see how Data% indicator increases. It's not linear. It grows fast and stops for some time, and again grows. Is really "empty" space transfered?
2. Online migration issue.
It seems to be slower than 1GB/s but I found that encryption is causing this. I will test the difference later. Because I use live migration with separate storage i need to use CLI and issue:
qm migrate 100 proxmox-nd2 --online --with-local-disks
Migration works fine but the disk is no longer thin-provisioned. Data% indicator in lvs command output shows 100% and I can confirm this with GUI.
Is there a way to bypass this and schrink guest?
