High‑Availability Service Based on Internal VIP in Cloud Computing

In fast‑paced environments, any service interruption is intolerable, so high‑availability (HA) is crucial; UCloud offers an internal VIP‑based HA service that achieves second‑level failover even in complex heterogeneous overlay networks.

The HA principle relies on backup nodes and automatic failover: when a node fails, traffic switches to a healthy node without user perception, emphasizing backup and failover as key points.

Traditional HA in conventional networks often uses Keepalived with a virtual IP (VIP), where a master node provides the VIP and a backup takes over via VRRP negotiation upon failure.

Cloud overlay networks replace traditional architectures, introducing challenges for HA such as ARP, IP addressing, and broadcast handling, requiring new solutions.

UCloud’s internal VIP service integrates with OVS bridges, a Ryu‑based controller, and GRE encapsulation to manage flows, ACLs, routing, and VPC isolation, enabling seamless VIP migration without extra API calls.

Failover involves two steps: (1) election of a new master after the original master fails, and (2) broadcasting the VIP’s new location to the network.

Broadcast mechanisms have evolved: the first generation used simulated broadcast by duplicating packets via multiple OVS outputs; the second generation introduced a broadcast cluster built on DPDK to handle dynamic broadcast domains efficiently; the third generation adds GARP sniffing to inform three‑layer networks of VIP moves, supporting heterogeneous clouds.

Practical examples include an e‑commerce payment system that leverages Keepalived plus internal VIP for rapid failover, and UCloud’s UDB database product that uses a dual‑master semi‑sync architecture with internal VIP to provide seamless HA without user intervention.

The article concludes that HA is complex, requiring not only VIP solutions but also disciplined service segmentation, monitoring, and continuous optimization to ensure reliable operations.