An HA Error

[Terry Xu]

Hi,
I found a bad HA action in PVE version 6.3. When all the corosync links are down in a host, this host's lrm losts agent lock.
After a moment the VM in HA group on this host starts on other host. But this VM is still running on origin host. Conflic happens, two VMs access same disk.
I think those VMs on corosync-link-down host should be stopped. Is it solved in new version? Or other solution?

Thanks!

 

[aaron]

This does sound wrong. If a node with HA guests is losing the connection to the cluster, it should fence itself (hard reset) after about 1 minute. After about another minute, or about 2 minutes after the node was last seen, the HA guests should be started on the remaining cluster.

Proxmox VE 6.3 is quite old by now. Please update it to the latest version. If the problem remains, we can take a closer look.

 

[Terry Xu]

Can I update pve-ha-manager only?

 

[aaron]

Can I update pve-ha-manager only?

Not really.