Resource ID's in OpenStack are UUIDs which conform to the standard UUID format - [a-zA-Z0-9-]. In some cases, a resource ID is generated which does not fit this format. For example, when DHCP agent dies, while it is being rescheduled, its port is temporarily attached to a dummy device with name 'reserved_dhcp_port' and an update_port() is sent to all the plugins.
On receiving an update_port() with the port's device id as 'reserved_dhcp_port', the Arista ML2 driver sends it to EOS which is rejected as the id does not conform to the standard UUID format. This is generally ok during the normal operation as once the port is attached to the new DHCP agent, another update_port() is called with the actual ID which will be accepted by EOS. However, if the neutron server dies while the dhcp agent is being migrated, upon restart, the Arista ML2 driver performs a sync which fails because of the unexpected UUID format preventing the neutron server from coming up.
Fixed by change - https:/