Mirror Server

[Lking299]

Hi, i got a question
is possible to have 2 servers with proxmox running in mirror, when 1 of the servers fail the other continue without loss information


Thanks,

Best Regards

 

[Dunuin]

Hi, i got a question
is possible to have 2 servers with proxmox running in mirror, when 1 of the servers fail the other continue without loss information


Thanks,

Best Regards

No.
You always need at least 3 machines for quorum (3 nodes or 2 nodes + 1 qdevice). With ceph as a real shared storage you need 3 ceph hosts. Local storage via ZFS with replication could run on 2 nodes + qdevice buy then you will lose the last minute of data in case of a failover.

 

[UdoB]

without loss information

That is not how it works.

Lets say you have a Cluster with three nodes running different VMs and properly configured networking and shared storage and finally High Availability ("HA"). Now one host dies, perhaps by a power failure for this hardware.

The VMs running on that failed node are not available now. All web server sessions and all other processes are just gone.

A minute (or two) later those VMs will start on another node. This is what HA in PVE context means. They will pick up their tasks depending on the state saved on the shared storage.

If - and only if - you have application layer redundancy this may be completely invisible by a remote user. But that is not a PVE related topic.

No magic involved...

 

[Lking299]

No.
You always need at least 3 machines for quorum (3 nodes or 2 nodes + 1 qdevice). With ceph as a real shared storage you need 3 ceph hosts. Local storage via ZFS with replication could run on 2 nodes + qdevice buy then you will lose the last minute of data in case of a failover.

Ok, thank you very much for your information, i try to get enough quorum

 

[Lking299]

That is not how it works.

Lets say you have a Cluster with three nodes running different VMs and properly configured networking and shared storage and finally High Availability ("HA"). Now one host dies, perhaps by a power failure for this hardware.

The VMs running on that failed node are not available now. All web server sessions and all other processes are just gone.

A minute (or two) later those VMs will start on another node. This is what HA in PVE context means. They will pick up their tasks depending on the state saved on the shared storage.

If - and only if - you have application layer redundancy this may be completely invisible by a remote user. But that is not a PVE related topic.

No magic involved...

Thank you for your answer, i´ll check it

 

[liptech]

Hi, i got a question
is possible to have 2 servers with proxmox running in mirror, when 1 of the servers fail the other continue without loss information


Thanks,

Best Regards

Dear Lking299,

I made a script for this purpose, see if it helps you.
https://github.com/liptech/Proxmox_Scripts/blob/main/quorum2.sh

 

[Dunuin]

If you want to run a cluster you should have at least 2 full PVE nodes + 3rd machine (for example a bare metal PBS, SBC, thin client, VM on some NAS box) as qdevice. With that script you are risking a split brain and potentially data-loss. "pvecm expected 1" should only be used manually after verifying that one of two server is actually dead/powered off and that there isn't a network problem and both running but not able to see each other.

 

[liptech]

If you want to run a cluster you should have at least 2 full PVEnodes + 3rd machine (for example a bare metal PBS, SBC, thin client, VM on some NAS box) as qdevice. With that script you are risking a split brain and potentially data-loss. "pvecm expected 1" should only be used manually after verifying that one of two server is actually dead/powered off and that there isn't a network problem and both running but not able to see each other.

@Dunuin​

You're right, I hadn't thought about this shared storage situation.

​