Strange behaviour with HA

[ewuewu]

Hello,

I am running a two node cluster for testing purposes (Vers. 3.0-23 / 957f0862)

Currently I am evaluating the HA feature and there I found a strange behavior:

Whenever I assign HA to stopped VMs those VMs will be started on node A always even they have been located on node B before.

Is this a default behavior or do we have some misconfiguration?

Where can I start searching to avoid this unintentional migration?

I have another test environment (also two node cluster but Version 2.2-32) where a vm starts on that node it was located originally before I assign HA.

Any hints are appreciated.

 

[dietmar]

Whenever I assign HA to stopped VMs those VMs will be started on node A always even they have been located on node B before.

Is this a default behavior or do we have some misconfiguration?

We use the following command to start the HA service:

# clusvcadm -e pvevm:<VMID> -m <node>

Where <node> is the node where the VM is currently located.

So that should really start on the correct node (see 'man clusvcadm').

Are there any hints in /var/log/cluster/rgmanager.log?

 

[ewuewu]

Hello Dietmar,
thank you for your quick reply.
Here are the entries from rgmanager.log
Node A (where the vm was originally located when I activated HA)

Sep 13 11:27:28 rgmanager Reconfiguring
Sep 13 11:27:28 rgmanager Loading Service Data
Sep 13 11:27:30 rgmanager [pvevm] VM 101 is running
Sep 13 11:27:31 rgmanager Stopping changed resources.
Sep 13 11:27:31 rgmanager Restarting changed resources.
Sep 13 11:27:31 rgmanager Starting changed resources.
Sep 13 11:27:31 rgmanager Initializing pvevm:100
Sep 13 11:27:31 rgmanager pvevm:100 was added to the config, but I am not initializing it.
Sep 13 11:27:33 rgmanager Migration: pvevm:100 is running on 2
Sep 13 11:27:39 rgmanager [pvevm] VM 101 is running
Sep 13 11:27:49 rgmanager [pvevm] VM 101 is running

Node B (where the vm has been moved to after clicking HA activate)

Sep 13 11:27:28 rgmanager Reconfiguring
Sep 13 11:27:28 rgmanager Loading Service Data
Sep 13 11:27:29 rgmanager Stopping changed resources.
Sep 13 11:27:29 rgmanager Restarting changed resources.
Sep 13 11:27:29 rgmanager Starting changed resources.
Sep 13 11:27:29 rgmanager Initializing pvevm:100
Sep 13 11:27:29 rgmanager pvevm:100 was added to the config, but I am not initializing it.
Sep 13 11:27:30 rgmanager Starting stopped service pvevm:100
Sep 13 11:27:30 rgmanager [pvevm] Move config for VM 100 to local node
Sep 13 11:27:30 rgmanager Migration: pvevm:101 is running on 1
Sep 13 11:27:31 rgmanager [pvevm] Task still active, waiting
Sep 13 11:27:32 rgmanager Service pvevm:100 started
Sep 13 11:27:33 rgmanager [pvevm] VM 100 is running

 

[ewuewu]

Hello,


I've changed our two node cluster into a licensed 3.1 three node cluster but the described behavior is still the same. Every VM for which we newly enable HA (and Autostart) will be moved to Node A on the first start, no matter on which node it was located before.
If we disable Autostart on the first start, the VM stays on the originally node.
Any ideas?

 

[Raymond Burns]

Hello,


I've changed our two node cluster into a licensed 3.1 three node cluster but the described behavior is still the same. Every VM for which we newly enable HA (and Autostart) will be moved to Node A on the first start, no matter on which node it was located before.
If we disable Autostart on the first start, the VM stays on the originally node.
Any ideas?

Is it the same behavior if the VM is started when activating HA?

 

[ewuewu]

I've learned that activating HA on a running VM is not allowed. I remember it was forbidden under Proxmox 2.x.
Am I wrong?
Is it safe to activate HA on a running VM under Proxmox 3.1?

 

[dietmar]

Is it safe to activate HA on a running VM under Proxmox 3.1?

No, you should stop the VM before activating HA on the VM.