使用WireGuard网状网络实现多区域故障转移

# 01 选择三个独立的故障域

Different countries, ideally different transit mixes. Amsterdam, Ashburn and Singapore is the classic triangle. Three is the minimum for quorum — two gives you a split-brain problem rather than redundancy.

# 02 构建网状网络

Each node gets a stable private address. Every node peers with every other node; with three nodes that is three tunnels.

# node A — /etc/wireguard/mesh.conf
[Interface]
Address = 10.10.0.1/24
ListenPort = 51821
PrivateKey = <A private>

[Peer]                       # node B
PublicKey = <B public>
Endpoint = b.example.net:51821
AllowedIPs = 10.10.0.2/32
PersistentKeepalive = 25

[Peer]                       # node C
PublicKey = <C public>
Endpoint = c.example.net:51821
AllowedIPs = 10.10.0.3/32
PersistentKeepalive = 25

# 03 仅将服务绑定到网状网络

The database, the cache and the internal API should listen on 10.10.0.x, never on the public address. This removes an entire class of exposure without a single firewall rule.

# 04 适当复制数据库

Synchronous replication across regions is impractical — a 160 ms round trip becomes the floor for every write. Use asynchronous replication across regions and know your recovery point objective.

# 05 在DNS层进行健康检查

Short TTLs plus health-checked failover records, or anycast if your regions support it. Then run a failure drill: kill a region deliberately, on a weekday, while you are watching.