This article addresses the issue with EX switches as the NTP time source for other networking devices on Layer 3 interfaces where VRRP is enabled.
VRRP enables hosts on a LAN to make use of redundant routing platforms on that LAN without requiring more than the static configuration of a single default route on the hosts. The VRRP routing platforms share the IP address corresponding to the default route configured on the hosts. At any time, one of the VRRP routing platforms is the primary (active) and the others are backups. If the primary routing platform fails, one of the backup routing platforms becomes the new primary, providing a virtual default routing platform and enabling traffic on the LAN to be routed without relying on a single routing platform. Using VRRP, a backup EX-series switch can take over a failed default switch within few seconds. This is done with minimum VRRP traffic and without any interaction with the hosts. When NTP is configured on a VRRP enabled interface, the NTP requests from the clients are replied with the real physical ip address of the interface. As the reply contains the source address as the physical IP of the interface, the clients rejects the reply since they were expecting a reply from the VRRP address which is configured as the NTP server. Because of this the clients never gets the NTP offset and is never able to syncronize to the correct time. The packet capture below, taken on the EX switch, shows that the client 10.0.1.2 is sending its request to the VRRP address 10.0.0.1 but the VRRP primary's physical IP, in this case 10.0.0.3 responds to the client's request. The client rejects this response.