PVE4.2-2 HA | VM restart when node fails

[amitadm]

Hello,
I configured an HA cluster with three nodes. I created a VM (Windows 7) to test the HA. when the node fails, the VM does not migrate but will restart on another node. I in various production systems with version 3.5, when a node fails the VM's migrate the other nodes.
Can you help me ?

Thanks!!

 

[amitadm]

I need to activate the live migration when node fails. Proxmox 3.5 is able to do this. I would like to know if Proxmox 4 does the same.

 

[tom]

Yes, start reading here:

http://pve.proxmox.com/wiki/High_Availability_Cluster_4.x

 

[amitadm]

I followed to the letter this guide. the system works but restarts VMs when a node fails. I set the following parameters: max_restart = 0, max_relocate = 3. The result does not change.

 

[amitadm]

ha-manager config
vm:100
group TEST
max_relocate 3
max_restart 0
state enabled

ha-manager status --verbose
quorum OK
master mih533 (active, Sat Jul 16 16:17:12 2016)
lrm mih531 (idle, Sat Jul 16 16:17:09 2016)
lrm mih532 (active, Sat Jul 16 16:17:06 2016)
lrm mih533 (idle, Sat Jul 16 16:17:09 2016)
service vm:100 (mih532, started)
full cluster state:
{
"lrm_status" : {
"mih531" : {
"mode" : "active",
"results" : {},
"state" : "wait_for_agent_lock",
"timestamp" : 1468678629
},
"mih532" : {
"mode" : "active",
"results" : {
"W/n+/Tbx83H7JDdLJrLpvw" : {
"exit_code" : 0,
"sid" : "vm:100",
"state" : "started"
}
},
"state" : "active",
"timestamp" : 1468678626
},
"mih533" : {
"mode" : "active",
"results" : {},
"state" : "wait_for_agent_lock",
"timestamp" : 1468678629
}
},
"manager_status" : {
"master_node" : "mih533",
"node_status" : {
"mih531" : "online",
"mih532" : "online",
"mih533" : "online"
},
"relocate_trial" : {
"vm:100" : 0
},
"service_status" : {
"vm:100" : {
"node" : "mih532",
"state" : "started",
"uid" : "jM4C2/4Pa+goUdfUWvpQoQ"
}
},
"timestamp" : 1468678632
},
"quorum" : {
"node" : "mih532",
"quorate" : "1"
}
}

 

[rickygm]

In version 3 does not restart the vm , in the version 4 restart the vm and lost communication while doing

 

[dcsapak]

what was the behaviour on 3.4 (i guess you mean this version, since there is no 3.5 of PVE)?
when a node dies, the only possibility is to restart the vm on another node, since the vm on the now dead node is also not on (how could it be)

so what exactly is the problem?

 

[rickygm]

what was the behaviour on 3.4 (i guess you mean this version, since there is no 3.5 of PVE)?
when a node dies, the only possibility is to restart the vm on another node, since the vm on the now dead node is also not on (how could it be)

so what exactly is the problem?

Hi dcsapak , in version 3.4 when a node die, the vm were located on another node, he did not restart the vm, in this version 4.X I had more 2mn disconnections while the vm is restarted.

 

[dcsapak]

in version 3.4 when a node die, the vm were located on another node, he did not restart the vm

i do not think this is possible. how could a vm still run when the node is gone/offline? the only way is then to start the vm again.
are you sure you dont misinterpret the behaviour? did you change from containers to qemu virtual machines? (containers start much faster)

maybe your vms just take longer to boot ?