We have a relatively stable 140 node Openstack deployment consisting of hypervisors and bare-metal.

Periodically, we have observed that the baremetals timeout during the PXE.  

We have been able to reproduce this relatively easily by rebooting several baremetals simultaneously.

I believe I have traced the issue to the dnsmasq reloading every time an entry is added / deleted.

Can anyone who has experienced similar issue provide feedback on what you did to work around this ?

We are running a stable Queens release.

thanks,

Fred.