Comment 11 for bug 1779304

> Means that shaker_frktis_agent_2,3,4 was not lost,

That's good, it means that networking is configured properly, image is correct and has shaker-agent.

My suggestion is to check whether all VMs get proper metadata via cloud-init. It may be a case that metadata server is overloaded and does not reply within default cloud-init timeout (which is, as I recall correctly, just 10 seconds). If it is the case then the issue should not appear with lower number of VMs.