Changing/Replacing Master node

[SamTzu]

I have been testing Proxmox and I am quite taken with it. Bravo, well done.

There are a few questions I have though, mostly to do with catastrophic hardware/software failure.

How does one replace a failed Cluster Master node?

I tried virtualising the Cluster Master and that works ok, but only on a VZ level. Inside a KVM/Qemu I am unable to create any more KVM based virtual machines. (Qemu container does not seem to support hardware level virtualisation that KVM needs.)

Is there a way to bypass this? I really would like to have virtual Master node for the ease of management.
If not I will just make two Different Clusters. One with Hardware based Cluster Master for KVM's and one with Virtual Cluster Master for VZ's.


Sam

 

[tom]

I have been testing Proxmox and I am quite taken with it. Bravo, well done.

There are a few questions I have though, mostly to do with catastrophic hardware/software failure.

How does one replace a failed Cluster Master node?

I tried virtualising the Cluster Master and that works ok, but only on a VZ level. Inside a KVM/Qemu I am unable to create any more KVM based virtual machines. (Qemu container does not seem to support hardware level virtualisation that KVM needs.)

Is there a way to bypass this? I really would like to have virtual Master node for the ease of management.
If not I will just make two Different Clusters. One with Hardware based Cluster Master for KVM's and one with Virtual Cluster Master for VZ's.


Sam

KVM inside KVM is not possible on Proxmox VE - as far as I remember there is some experimental code in KVM (for AMD cpu´s only).

 

[SamTzu]

Ok, thanx for the fast reply. It was only a way to recover the Master Node easily.

How does one replace a failed Cluster Master node?

Sam

 

[tom]

Ok, thanx for the fast reply. It was only a way to recover the Master Node easily.

How does one replace a failed Cluster Master node?

Sam

Just promote one of the remaining nodes to a new master and tell all other nodes to sync from the new master. (see 'man pveca')

 

[SamTzu]

New Proxmox Cluster Master

Just to clarify the steps...

to remove a master config and rebuild:

1. remove rm /etc/pve/cluster.cfg from all nodes (and master)
2. remove the stored ssh keys: rm /root/.ssh/known_hosts from all nodes (and master)
3. create a new cluster master: pveca -c
4. join nodes to the new master: pveca -a -h IP_MASTER

Is this correct?

What about the running Virtual Machines on host nodes?
Do we have to back them up somewhere for the duration and then restore them?


Sam

 

[tom]

Just to clarify the steps...

to remove a master config and rebuild:

1. remove rm /etc/pve/cluster.cfg from all nodes (and master)
2. remove the stored ssh keys: rm /root/.ssh/known_hosts from all nodes (and master)
3. create a new cluster master: pveca -c
4. join nodes to the new master: pveca -a -h IP_MASTER

Is this correct?

What about the running Virtual Machines on host nodes?
Do we have to back them up somewhere for the duration and then restore them?


Sam

your howto is for a NEW cluster, but as you just need a new master, it is:

1. force a node to a master: pveca -m
2. tell all the other nodes to sync from the new master: pveca -s -h IP_of_new_master

running virtual machines on the nodes do not have to be stopped for this.

 

[SamTzu]

Thank you.
This info is what most admins are looking for.

After doing some disaster recovery in M$ world I just have to love the simplicity of this approach. It's funny how, the older I get, the better I like the simple solutions.

Sam

 

[mangoo]

KVM inside KVM is not possible on Proxmox VE - as far as I remember there is some experimental code in KVM (for AMD cpu´s only).

It is available, as of kvm-85.

 

[SamTzu]

I did the Cluster Master promotion (very easy) and it worked quite well.
I doubt that I will bother to virtualise the Master anymore since disaster recovery is so easy.
Over a decade in the nerd business has thought me the value of first thinking about the migration/migraine before doing anything else.

Now if someone could come up with a way to virtualise the user service accounts (like www&mail).
That would be heaven.


Sam