The storage VM was started last, but PVE's guest autostart
(pve-guests.service) is a one-shot unit with no retry: if it fires before
pvestatd manages to log into the iSCSI portal (node.startup=manual, so
open-iscsi won't auto-relogin on its own), any onboot VM on lab-lvm just
stays stopped. Start pvestore first and confirm NFS/iSCSI are actually
serving before booting the PVE nodes, then force iSCSI login + LVM
activation and retry any still-stopped onboot VMs as a safety net.
Modular libvirt daemon sockets and the manual iptables NAT/forward rules
don't survive a reboot; this re-arms them (safe to do before any VM is
started) and starts the three existing VMs.
