咨询 cilium loadbalancer 问题

[sinycn1]

我在本地有 4 台机器 其中 1 master 3 node 利用 kubeadm 搭建 k8s1.29 版本的集群，并按照官方教程安装了 cilium CNI , 官方有提到可以使用 cilium 作为 loadbalancer 我就采用了 BGP 方式设置

apiVersion: "cilium.io/v2alpha1"
kind: CiliumLoadBalancerIPPool
metadata:
  name: "pool"
spec:
  cidrs:
    - cidr: "10.X.X.0/24"
---
	
apiVersion: "cilium.io/v2alpha1"
kind: CiliumBGPPeeringPolicy
metadata:
 name: 01-bgp-peering-policy
spec:
 virtualRouters:
 - localASN: 64512
   exportPodCIDR: false
   neighbors:
    - peerAddress: '10.X.X.1/32'
      peerASN: 64512

然后部署了一个 nginx 的 svc

kind: Pod
metadata:
  name: nginx
  labels:
    app.kubernetes.io/name: proxy
spec:
  containers:
  - name: nginx
    image: nginx:stable
    ports:
      - containerPort: 80
        name: http-web-svc

---
apiVersion: v1
kind: Service
metadata:
  name: nginx-service
spec:
  type: LoadBalancer
  selector:
    app.kubernetes.io/name: proxy
  ports:
  - name: name-of-service-port
    protocol: TCP
    port: 80
    targetPort: http-web-svc

k get svc 的结果 nginx-service LoadBalancer 172.16.7.185 10.X.X.2 80:31206/TCP 2d22h

已知从 10.X.X.30 10.X.X.31 10.X.X.32 10.X.X.33 (1 master 3 node) 执行 curl http://10.X.X.2

但是从其他机器上就不法访问 arp -a 显示 ? (10.X.X.2) at <incomplete> on eno1

各位大神可有解决思路，小弟万分感谢

[DAPTX4869]

其它机器要访问的是 31206 端口?

[pubby]

其他机器是指 k8s 节点以外的机器吗？


对你的网络地址划分不太清楚，尤其这 2 个地方不太理解：

cidr: "10.X.X.0/24" 这里是用于分配 service external-ip 的，不要和你 node 地址段重合
peerAddress: '10.X.X.1/32' 这里应该是一个 k8s 集群外的一个 bgp 路由 ip

没用过 cilium,我们是用 clico 的，集群外面用 bird 部署了 bgp 路由作为 peer ，这样外面就能直连 service external-ip 了

[sinycn1]

@pubby 感谢你的回复，目前问题还没有解决，所以我继续回帖
1.cidr: "10.X.X.0/24" 这里是用于分配 service external-ip 的，不要和你 node 地址段重合
之前确实是使用了同一网段，现在拆分为 10.206.20.0/22 （ node 使用）和 10.206.24.0/22 (external-ip)
peerAddress: '10.206.20.1/32' 这里应该是一个 k8s 集群外的一个 bgp 路由 ip --- 是这样设定的

目前还是没有成功的发布路由 有更新的话我再来回帖
```
cilium bgp routes
(Defaulting to `available ipv4 unicast` routes, please see help for more options)

Node VRouter Prefix NextHop Age Attrs
```

[pubby]

你先看一下 external-ip 分配了没有

然后集群外部需要有个 bgp peer 服务，可以用 bird 部署一个。（举例：192.168.3.100 ）

在外部主路由器上配个路由规则，目标 external-ip 段 走 bgp peer


外部 bgp 服务配置， ：

```code
[root@centos8-bgp-gw ~]# cat /etc/bird.conf
router id 192.168.3.100;

log syslog all;

protocol kernel {
scan time 10;
ipv4 {
#import none;
export all; # insert routes into the kernel routing table
};
merge paths on; # Allow export multipath routes (ECMP)
graceful restart;
}

protocol device {
scan time 60;
}


# Template example. Using templates to define IBGP route reflector clients.
template bgp k8s_nodes {
local 192.168.3.100 as 63400;
neighbor as 63400;

direct;
ipv4 {
import all;
export none;
};
}

# for vm-k8s-master
protocol bgp vm_k8s_master from k8s_nodes {
neighbor 192.168.3.105;
}
# for vm1-node1
protocol bgp vm1_node1 from k8s_nodes {
neighbor 192.168.4.121;
}
# for vm2-node4
protocol bgp vm2_node4 from k8s_nodes {
neighbor 192.168.4.104;
}
# for vm2-node5
protocol bgp vm2_node5 from k8s_nodes {
neighbor 192.168.4.105;
}
```

然后 bgp peer 上用 ip r 看一下那些 external-ip 的路由有没有自动生成

[sinycn1]

@pubby 感谢你的 update ，后来问题解决了，我使用 vyos+cilium bgp 实现了 bgp 路由动态发布，主要是需要变更 cilium 的配置，相关信息可以参考这个 issue(自己提交自己解决)
https://github.com/cilium/cilium/issues/32375

[a0xbd4CX0DHC1EuT]

ping 不通是不是同一个问题
```JavaScript
[root@node1 metallb]# tcpdump -i any host 172.27.0.7 -s0 -A
tcpdump: data link type LINUX_SLL2
dropped privs to tcpdump
tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
listening on any, link-type LINUX_SLL2 (Linux cooked v2), snapshot length 262144 bytes
10:26:33.131273 lo In IP node1 > node1: ICMP echo request, id 6, seq 1, length 64
E..Te:@.@.}*..................:f....}....................... !"#$%&'()*+,-./01234567
10:26:33.131305 lo In IP node1 > node1: ICMP node1 protocol 1 port 42460 unreachable, length 92
E..pe;[email protected]:@.@.}*..................:f....}....................... !"#$%&'()*+,-./01234567
10:26:34.165170 lo In IP node1 > node1: ICMP echo request, id 6, seq 2, length 64
E..Th&@[email protected]>...........W......:f............................ !"#$%&'()*+,-./01234567
10:26:34.165194 lo In IP node1 > node1: ICMP node1 protocol 1 port 3927 unreachable, length 92
E..ph'[email protected]&@[email protected]>...........W......:f............................ !"#$%&'()*+,-./01234567
10:26:35.194465 lo In IP node1 > node1: ICMP echo request, id 6, seq 3, length 64
E..Tl&@[email protected]>..................:f....X....................... !"#$%&'()*+,-./01234567
10:26:35.194505 lo In IP node1 > node1: ICMP node1 protocol 1 port 51171 unreachable, length 92
E..pl'[email protected]&@[email protected]>..................:f....X....................... !"#$%&'()*+,-./01234567
10:26:36.218668 lo In IP node1 > node1: ICMP echo request, id 6, seq 4, length 64
E..TnS@[email protected].......:f.....U...................... !"#$%&'()*+,-./01234567
10:26:36.218798 lo In IP node1 > node1: ICMP node1 protocol 1 port 12676 unreachable, length 92
[email protected]@[email protected].......:f.....U...................... !"#$%&'()*+,-./01234567
^C
8 packets captured
21 packets received by filter
0 packets dropped by kernel
[root@node1 metallb]# kubectl get svc -n nginx
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
nginx-service LoadBalancer 10.233.12.235 172.27.0.7 80:32022/TCP 16h
[root@node1 metallb]# curl -I 172.27.0.7
HTTP/1.1 200 OK
Server: nginx
Date: Wed, 08 May 2024 02:32:54 GMT
Content-Type: text/html; charset=UTF-8
Connection: keep-alive
X-Powered-By: PHP/8.2.7

[root@node1 metallb]# ping 172.27.0.7
PING 172.27.0.7 (172.27.0.7) 56(84) bytes of data.
From 172.27.0.7 icmp_seq=1 Destination Port Unreachable
From 172.27.0.7 icmp_seq=2 Destination Port Unreachable
From 172.27.0.7 icmp_seq=3 Destination Port Unreachable
From 172.27.0.7 icmp_seq=4 Destination Port Unreachable
^C
--- 172.27.0.7 ping statistics ---
4 packets transmitted, 0 received, +4 errors, 100% packet loss, time 3100ms

[root@node1 metallb]# kubectl get pod -A | grep cil
kube-system cilium-mbjx2 1/1 Running 2 (57m ago) 24h
kube-system cilium-operator-5547b984f4-5d9c8 1/1 Running 2 (57m ago) 24h
kube-system cilium-operator-5547b984f4-z9kgk 1/1 Running 2 (57m ago) 24h
kube-system cilium-pc8hh 1/1 Running 2 (57m ago) 24h
[root@node1 metallb]# crictl images | grep cilium
quay.io/cilium/cilium v1.15.4 aebfd554d3483 209MB
quay.io/cilium/operator v1.15.4 cf4b9cdd4ba07 36.1MB
```