VM not able to migrate in realtime

[Ayush]

Hello Team,

We have a simple setup of 3 nodes running proxmox 8.1.4 with under ceph quency 17.2.7 as shared storage. We try to create a HA for our vm . So in case of any node goes down our VM can be live migrated to another node .
We try to test this scenario and we intentionally reboot one node. But our vm get in to freeze state until our box comeback online again. When we check the uptime of vm it shows that it is rebooted.
What is the way to achieve live migration ? We follow proper documentation as per documents but it is not working for us.

 

[Azunai333]

If you change
Datacenter > Options > HA Settings
to shutdown_policy=migrate VMs will be migrated automatically migrated to another node on shutdown/reboot.

 

[bbgeek17]

to shutdown_policy=migrate VMs will be migrated automatically migrated to another node on shutdown/reboot.

The "shutdown" is not exactly what OP is doing to test, they are doing "reboot". That said, if the node comes up fast enough to fit into "timeout", then the VM will still stay on the original node:
https://forum.proxmox.com/threads/shutdown-policy-failover.61987/#post-284169

---

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

 

[LnxBil]

So in case of any node goes down our VM can be live migrated to another node .

That is not possible on a crash, maybe you mean just migrated? If there is nothing alive on the failed node, there cannot be a live migration.

The "shutdown" is not exactly what OP is doing to test, they are doing "reboot". That said, if the node comes up fast enough to fit into "timeout", then the VM will still stay on the original node:

Yes, exactly. If you test, please press reset, send an NMI or crash your linux intentionally in order to simulate a real hardware-induced reboot.

 

[Ayush]

But if Ceph is the under line shared storage and it is working replicated mode. Then in case of node down why it is not possible to do live migration. As ceph still have related data that is required to run vm. I know this could be a beginner query but I still want to know the reason behind this.

 

[Ayush]

If you change
Datacenter > Options > HA Settings
to shutdown_policy=migrate VMs will be migrated automatically migrated to another node on shutdown/reboot.

But is it a live migration ?

 

[bbgeek17]

But if Ceph is the under line shared storage and it is working replicated mode. Then in case of node down why it is not possible to do live migration. As ceph still have related data that is required to run vm. I know this could be a beginner query but I still want to know the reason behind this.

because besides data on disk you also need memory state of the VM. Otherwise how can VM stay "up"?
When both nodes are up - PVE transfers live memory state, pauses VM for a moment, finishes differential memory state transfer and starts the VM on target side with that state (all done within milliseconds or few seconds at most).

But is it a live migration ?

yes it will be live, but _only_ if you actually execute a shutdown operation on the hypervisor, ie type "poweroff". So that the hypervisor will go through _orderly_ shutdown. Same for "reboot=" policy.
If your hypervisor dies unexpectedly - there is no live migration with PVE. There is some work going on for that, but its not here yet.

---

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

 

[LnxBil]

But if Ceph is the under line shared storage and it is working replicated mode. Then in case of node down why it is not possible to do live migration. As ceph still have related data that is required to run vm.

A live migration needs the source to be (a)live, not dead ... If a node crashes, the VM is started on another node. This is called a failover.