Keepalived won't forward traffic to BACKUP node after Kubernetes cluster setup

Question

System Structure:

• 10.10.1.86: Kubernetes master node
• 10.10.1.87: Kubernetes worker 1 node; keepalived MASTER node
• 10.10.1.88: Kubernetes worker 2 node; keepalived BACKUP node
• 10.10.1.90: VIP, would load balance to .87 & .88; implemented by keepalived.

This Kubernetes cluster is a dev env, testing collect netflow log.

What I want to achieve is:

1. All router / switch netflow log first output to .90
2. Then use keepalived to load balance (lb_kind: NAT) to .87 & .88, which are two Kubernetes workers.
3. There is NodePort Service to catch these traffic into Kubernetes cluster and do the rest data parsing jobs.

• Something like:

        |                {OS Network}                   |   {Kubernetes Network}

                                K8s Worker -> filebeat -> logstash (deployments)
                              /
<data> -> [VIP] load balance
                              \ 
                                K8s Worker -> filebeat -> logstash (deployments)

• filebeat.yml (have tested that traffics are all fine after filebeat, so I use file output to narrow root cause.)

# cat filebeat.yml
filebeat.inputs:

- type: tcp
  max_message_size: 10MiB
  host: "0.0.0.0:5100"

- type: udp
  max_message_size: 10KiB
  host: "0.0.0.0:5150"




#output.logstash:
#  hosts: ["10.10.1.87:30044", "10.10.1.88:30044"]
output.file:
  path: "/tmp/"
  filename: tmp-filebeat.out

Kubernetes

• Master and Workers are 3 VMs in my private env; not any of GCP or AWS providers.
• Version:

# kubectl version
Client Version: version.Info{Major:"1", Minor:"21", GitVersion:"v1.21.0", GitCommit:"cb303e613a121a29364f75cc67d3d580833a7479", GitTreeState:"clean", BuildDate:"2021-04-08T16:31:21Z", GoVersion:"go1.16.1", Compiler:"gc", Platform:"linux/amd64"}
Server Version: version.Info{Major:"1", Minor:"21", GitVersion:"v1.21.0", GitCommit:"cb303e613a121a29364f75cc67d3d580833a7479", GitTreeState:"clean", BuildDate:"2021-04-08T16:25:06Z", GoVersion:"go1.16.1", Compiler:"gc", Platform:"linux/amd64"}

• Services

# cat logstash.service.yaml
apiVersion: v1
kind: Service
metadata:
  name: logstash-service
spec:
  type: NodePort
  selector:
    app: logstash
  ports:
    - port: 9514
      name: tcp-port
      targetPort: 9514
      nodePort: 30044

• Once data get in Kubernetes, all works fine.
• It was the VIP load balance not forwarding.

Keepalived conf

!Configuration File for keepalived
global_defs {
  router_id proxy1   # `proxy 2` at the other node
}


vrrp_instance VI_1 {
  state MASTER       # `BACKUP` at the other node
  interface ens160
  virtual_router_id 41
  priority 100       # `50` at the other node
  advert_int 1
  virtual_ipaddress {
    10.10.1.90/23
  }
}

virtual_server 10.10.1.90 5100 {
  delay_loop 30
  lb_algo rr
  lb_kind NAT
  protocol TCP
  persistence_timeout 0

  real_server 10.10.1.87 5100 {
    weight 1
  }
  real_server 10.10.1.88 5100 {
    weight 1
  }
}
virtual_server 10.10.1.90 5150 {
  delay_loop 30
  lb_algo rr
  lb_kind NAT
  protocol UDP
  persistence_timeout 0

  real_server 10.10.1.87 5150 {
    weight 1
  }
  real_server 10.10.1.88 5150 {
    weight 1
  }

It works Before Kubernetes cluster setup

• Both .87 & .88 have installed keepalived, and rr (RoundRobin) load balance works fine (TCP and UDP).
• Stop keepalived service (systemctl stop keepalived) when going to setup kubernetes cluster, just in case.

Problem occurred After Kubernetes cluster setup

• Found that only MASTER node .87 can get traffic forwarded, the VIP can't forward to BACKUP node .88.
• The forwarded data from MASTER is successfully catched by kubernetes NodePort and deployments.

Problem test by nc:

• nc: only who holds VIP (MASTER node) can forward traffic, when rr forward to BACKUP, it just shows timeout.
• also tested by nc -l 5100 on both server, only MASTER node got results.

# echo "test" | nc 10.10.1.90 5100
# echo "test" | nc 10.10.1.90 5100
Ncat: Connection timed out.
# echo "test" | nc 10.10.1.90 5100
# echo "test" | nc 10.10.1.90 5100
Ncat: Connection timed out.

Some Info

• Package versions

# rpm -qa |grep keepalived
keepalived-1.3.5-19.el7.x86_64

• Kubernetes CNI: Calico

# kubectl get pod -n kube-system
NAME                                      READY   STATUS    RESTARTS   AGE
calico-kube-controllers-b656ddcfc-wnkcj   1/1     Running   2          78d
calico-node-vnf4d                         1/1     Running   8          78d
calico-node-xgzd5                         1/1     Running   1          78d
calico-node-zt25t                         1/1     Running   8          78d
coredns-558bd4d5db-n6hnn                  1/1     Running   2          78d
coredns-558bd4d5db-zz2rb                  1/1     Running   2          78d
etcd-a86.axv.bz                           1/1     Running   2          78d
kube-apiserver-a86.axv.bz                 1/1     Running   2          78d
kube-controller-manager-a86.axv.bz        1/1     Running   2          78d
kube-proxy-ddwsr                          1/1     Running   2          78d
kube-proxy-hs4dx                          1/1     Running   3          78d
kube-proxy-qg2nq                          1/1     Running   1          78d
kube-scheduler-a86.axv.bz                 1/1     Running   2          78d

• ipvsadm (same result on .87, .88)

# ipvsadm -ln
IP Virtual Server version 1.2.1 (size=4096)
Prot LocalAddress:Port Scheduler Flags
  -> RemoteAddress:Port           Forward Weight ActiveConn InActConn
TCP  10.10.1.90:5100 rr
  -> 10.10.1.87:5100              Masq    1      0          0
  -> 10.10.1.88:5100              Masq    1      0          0
UDP  10.10.1.90:5150 rr
  -> 10.10.1.87:5150              Masq    1      0          0
  -> 10.10.1.88:5150              Masq    1      0          0

• Selinux is always Permissive
• If stop firewalld, still not work either.
• sysctl difference:

# before:
net.ipv4.conf.all.accept_redirects = 1
net.ipv4.conf.all.forwarding = 0
net.ipv4.conf.all.route_localnet = 0
net.ipv4.conf.default.forwarding = 0
net.ipv4.conf.lo.forwarding = 0
net.ipv4.ip_forward = 0

# after
net.ipv4.conf.all.accept_redirects = 0
net.ipv4.conf.all.forwarding = 1
net.ipv4.conf.all.route_localnet = 1
net.ipv4.conf.default.forwarding = 1
net.ipv4.conf.lo.forwarding = 1
net.ipv4.ip_forward = 1

Not sure if any further check can do now, please advise, thank you!