The following sections specify the GARP and RA packet specifications required by
Google Cloud when the haPolicy.fastIPMove backend service parameter is
set to GARP_RA.
GARP packet format for IPv4 traffic
The following fields are required when you use a GARP packet to trigger a new leader.
| Sender MAC address | MAC address of the sender VM instance (the new leader). |
| Sender IPv4 address | The /32 IPv4 address that is moving to the new leader. |
| Target MAC address | Broadcast MAC address: FF-FF-FF-FF-FF-FF |
| Target IPv4 address | Same as the sender IPv4 address. |
| ARP operation | ARPOP_REPLY: 2 |
RA packet format for IPv6 traffic
RA supports advertising multiple IPv6 prefix ranges in a single RA packet by attaching multiple ICMPv6 prefix information options.
When haPolicy.fastIPMove is set to GARP_RA, every IPv6 prefix range in a
single RA packet should be an exact match to a forwarding rule IP address, and
all of those forwarding rules must point to the same regional backend service.
If the set of IPv6 prefix ranges are served by different regional backend services, the RA packet is treated as invalid and doesn't trigger a fast IP move.
The following fields are required when you use an RA packet to trigger a new leader.
| Sender MAC address | MAC address of the sender VM instance (the new leader). |
| Sender IPv6 address | Link-local IPv6 address of the sender VM instance. This is not the IPv6 address that is moving to the new leader. |
| Target MAC address | All node multicast MAC address: 33-33-00-00-00-01 |
| Target IPv6 address | IPv6 all node multicast address: FF02::1 |
| IPv6 hop limit | 255 |
| ICMPv6 type | ND_ROUTER_ADVERT: 134 |
| ICMPv6 options type | ND_OPT_PREFIX_INFORMATION: 3 |
| Prefix information option prefix | The IPv6 prefix that is moving to the new leader. Bits after the prefix must be zero. |
| Prefix information option prefix length | The length of IPv6 prefix that is moving to the new leader. |