PVE 4.3 HA Behavior VM Templates

[ScottR]

Team,

I'm currently working on an application that utilizes a multi-node proxmox cluster to serve template VMs to a variety of users. The VMs that are served, are all linked clones. In previous versions, if a node were to fail, HA would migrate the templates to the remaining nodes. However, with 4.3, it seems that the "start" action on HA resources causes the template to fall into an "error" state. End result, the template is locked on the failed node. The error state existed in the past, but did not interfere with the migration.

Two questions.

1. Has anyone else used a better mechanism for VM templates and HA?
2. Is there any way to mark an HA resource as "do not start"? This would stop the error state, allowing the correct migration.

Thanks in advance.

 

[dietmar]

In previous versions, if a node were to fail, HA would migrate the templates to the remaining nodes.

We never had such functionality. If you want HA, you need to use shared storage for your VMs.

 

[ScottR]

We never had such functionality. If you want HA, you need to use shared storage for your VMs.

I'm not entirely sure what that has to do with what I'm asking. VM Template (KVM) _is_ a VM, it is stored on a central GlusterFS cluster. The only difference is that the template cannot be started. You cannot create linked clones without a template, and you cannot create instantaneous clones or space efficient copies without linked clones. The issue, something I consider to be a bug, is that the HA manager tries to start a template resource (KVM VM), that cannot be started resulting in an error state. The _new_ behavior is that the error state takes it out of an HA action. Given that the VM template is as important a a real VM, since it's a linked clone of a resource, it should be treated correctly.

The main reason this is an issue, is that you _must_ connect over the REST API to the node that contains the template to clone the resource to another machine. This requires that the templates migrate to a live node. The desirable action, is to treat a KVM template _exactly_ the same as an HA VM resource, except that you should not try to start it, as it's a frozen resource.

 

[dietmar]

The main reason this is an issue, is that you _must_ connect over the REST API to the node that contains the template to clone the resource to another machine. This requires that the templates migrate to a live node. The desirable action, is to treat a KVM template _exactly_ the same as an HA VM resource, except that you should not try to start it, as it's a frozen resource.

OK, got it. Please can you open a bug at bugzilla.proxmox.com?

 

[ScottR]

OK, got it. Please can you open a bug at bugzilla.proxmox.com?

Thanks. Here's the reference.

Bug 1160 - 4.3 ha-manager attempts to start kvm templates resulting in ha migration failure
https://bugzilla.proxmox.com/show_bug.cgi?id=1160