Set up an HA Cluster Using Keepalived and HAproxy

A highly available Kubernetes cluster ensures your applications run without outages which is required for production. In this connection, there are plenty of ways for you to choose from to achieve high availability.

This tutorial demonstrates how to configure Keepalived and HAproxy for load balancing and achieve high availability. The steps are listed as below:

  1. Prepare hosts.
  2. Configure Keepalived and HAproxy.
  3. Use KubeKey to set up a Kubernetes cluster and install KubeSphere.

Cluster Architecture

The example cluster has three master nodes, three worker nodes, two nodes for load balancing and one virtual IP address. The virtual IP address in this example may also be called “a floating IP address”. That means in the event of node failures, the IP address can be passed between nodes allowing for failover, thus achieving high availability.

architecture-ha-k8s-cluster

Notice that in this example, Keepalived and HAproxy are not installed on any of the master nodes. Admittedly, you can do that and high availability can also be achieved. That said, configuring two specific nodes for load balancing (You can add more nodes of this kind as needed) is more secure. Only Keepalived and HAproxy will be installed on these two nodes, avoiding any potential conflicts with any Kubernetes components and services.

Prepare Hosts

IP AddressHostnameRole
172.16.0.2lb1Keepalived & HAproxy
172.16.0.3lb2Keepalived & HAproxy
172.16.0.4master1master, etcd
172.16.0.5master2master, etcd
172.16.0.6master3master, etcd
172.16.0.7worker1worker
172.16.0.8worker2worker
172.16.0.9worker3worker
172.16.0.10Virtual IP address

For more information about requirements for nodes, network, and dependencies, see Multi-node Installation.

Configure Load Balancing

Keepalived provides a VRPP implementation and allows you to configure Linux machines for load balancing, preventing single points of failure. HAProxy, providing reliable, high performance load balancing, works perfectly with Keepalived.

As Keepalived and HAproxy are installed on lb1 and lb2, if either one goes down, the virtual IP address (i.e. the floating IP address) will be automatically associated with another node so that the cluster is still functioning well, thus achieving high availability. If you want, you can add more nodes all with Keepalived and HAproxy installed for that purpose.

Run the following command to install Keepalived and HAproxy first.

  1. yum install keepalived haproxy psmisc -y

HAproxy

  1. The configuration of HAproxy is exactly the same on the two machines for load balancing. Run the following command to configure HAproxy.

    1. vi /etc/haproxy/haproxy.cfg

  2. Here is an example configuration for your reference (Pay attention to the server field. Note that 6443 is the apiserver port):

    1. global
    2.     log /dev/log  local0 warning
    3.     chroot      /var/lib/haproxy
    4.     pidfile     /var/run/haproxy.pid
    5.     maxconn     4000
    6.     user        haproxy
    7.     group       haproxy
    8.     daemon
    10.    stats socket /var/lib/haproxy/stats
    12. defaults
    13.   log global
    14.   option  httplog
    15.   option  dontlognull
    16.         timeout connect 5000
    17.         timeout client 50000
    18.         timeout server 50000
    20. frontend kube-apiserver
    21.   bind *:6443
    22.   mode tcp
    23.   option tcplog
    24.   default_backend kube-apiserver
    26. backend kube-apiserver
    27.     mode tcp
    28.     option tcplog
    29.     option tcp-check
    30.     balance roundrobin
    31.     default-server inter 10s downinter 5s rise 2 fall 2 slowstart 60s maxconn 250 maxqueue 256 weight 100
    32.     server kube-apiserver-1 172.16.0.4:6443 check # Replace the IP address with your own.
    33.     server kube-apiserver-2 172.16.0.5:6443 check # Replace the IP address with your own.
    34.     server kube-apiserver-3 172.16.0.6:6443 check # Replace the IP address with your own.

  3. Save the file and run the following command to restart HAproxy.

    1. systemctl restart haproxy

  4. Make it persist through reboots:

    1. systemctl enable haproxy

  5. Make sure you configure HAproxy on the other machine (lb2) as well.

Keepalived

Keepalived must be installed on both machines while the configuration of them is slightly different.

  1. Run the following command to configure Keepalived.

    1. vi /etc/keepalived/keepalived.conf

  2. Here is an example configuration (lb1) for your reference:

    1. global_defs {
    2.   notification_email {
    3.   }
    4.   router_id LVS_DEVEL
    5.   vrrp_skip_check_adv_addr
    6.   vrrp_garp_interval 0
    7.   vrrp_gna_interval 0
    8. }
    10. vrrp_script chk_haproxy {
    11.   script "killall -0 haproxy"
    12.   interval 2
    13.   weight 2
    14. }
    16. vrrp_instance haproxy-vip {
    17.   state BACKUP
    18.   priority 100
    19.   interface eth0                       # Network card
    20.   virtual_router_id 60
    21.   advert_int 1
    22.   authentication {
    23.     auth_type PASS
    24.     auth_pass 1111
    25.   }
    26.   unicast_src_ip 172.16.0.2      # The IP address of this machine
    27.   unicast_peer {
    28.     172.16.0.3                         # The IP address of peer machines
    29.   }
    31.   virtual_ipaddress {
    32.     172.16.0.10/24                  # The VIP address
    33.   }
    35.   track_script {
    36.     chk_haproxy
    37.   }
    38. }

    Note

    • For the interface field, you must provide your own network card information. You can run ifconfig on your machine to get the value.

    • The IP address provided for unicast_src_ip is the IP address of your current machine. For other machines where HAproxy and Keepalived are also installed for load balancing, their IP address must be input for the field unicast_peer.

  3. Save the file and run the following command to restart Keepalived.

    1. systemctl restart keepalived

  4. Make it persist through reboots:

    1. systemctl enable haproxy

  5. Make sure you configure Keepalived on the other machine (lb2) as well.

Verify High Availability

Before you start to create your Kubernetes cluster, make sure you have tested the high availability.

  1. On the machine lb1, run the following command:

    1. [[email protected] ~]# ip a s
    2. 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    3.     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    4.     inet 127.0.0.1/8 scope host lo
    5.        valid_lft forever preferred_lft forever
    6.     inet6 ::1/128 scope host
    7.        valid_lft forever preferred_lft forever
    8. 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    9.     link/ether 52:54:9e:27:38:c8 brd ff:ff:ff:ff:ff:ff
    10.     inet 172.16.0.2/24 brd 172.16.0.255 scope global noprefixroute dynamic eth0
    11.        valid_lft 73334sec preferred_lft 73334sec
    12.     inet 172.16.0.10/24 scope global secondary eth0 # The VIP address
    13.        valid_lft forever preferred_lft forever
    14.     inet6 fe80::510e:f96:98b2:af40/64 scope link noprefixroute
    15.        valid_lft forever preferred_lft forever

  2. As you can see above, the virtual IP address is successfully added. Simulate a failure on this node:

    1. systemctl stop haproxy

  3. Check the floating IP address again and you can see it disappear on lb1.

    1. [[email protected] ~]# ip a s
    2. 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    3.     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    4.     inet 127.0.0.1/8 scope host lo
    5.        valid_lft forever preferred_lft forever
    6.     inet6 ::1/128 scope host
    7.        valid_lft forever preferred_lft forever
    8. 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    9.     link/ether 52:54:9e:27:38:c8 brd ff:ff:ff:ff:ff:ff
    10.     inet 172.16.0.2/24 brd 172.16.0.255 scope global noprefixroute dynamic eth0
    11.        valid_lft 72802sec preferred_lft 72802sec
    12.     inet6 fe80::510e:f96:98b2:af40/64 scope link noprefixroute
    13.        valid_lft forever preferred_lft forever

  4. Theoretically, the virtual IP will be failed over to the other machine (lb2) if the configuration is successful. On lb2, run the following command and here is the expected output:

    1. [[email protected] ~]# ip a s
    2. 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    3.     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    4.     inet 127.0.0.1/8 scope host lo
    5.        valid_lft forever preferred_lft forever
    6.     inet6 ::1/128 scope host
    7.        valid_lft forever preferred_lft forever
    8. 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    9.     link/ether 52:54:9e:3f:51:ba brd ff:ff:ff:ff:ff:ff
    10.     inet 172.16.0.3/24 brd 172.16.0.255 scope global noprefixroute dynamic eth0
    11.        valid_lft 72690sec preferred_lft 72690sec
    12.     inet 172.16.0.10/24 scope global secondary eth0   # The VIP address
    13.        valid_lft forever preferred_lft forever
    14.     inet6 fe80::f67c:bd4f:d6d5:1d9b/64 scope link noprefixroute
    15.        valid_lft forever preferred_lft forever

  5. As you can see above, high availability is successfully configured.

Use KubeKey to Create a Kubernetes Cluster

KubeKey is an efficient and convenient tool to create a Kubernetes cluster. Follow the steps below to download KubeKey.

Download KubeKey from its GitHub Release Page or use the following command directly.

  1. curl -sfL https://get-kk.kubesphere.io | VERSION=v1.1.0 sh -

Run the following command first to make sure you download KubeKey from the correct zone.

  1. export KKZONE=cn

Run the following command to download KubeKey:

  1. curl -sfL https://get-kk.kubesphere.io | VERSION=v1.1.0 sh -

Note

After you download KubeKey, if you transfer it to a new machine also with poor network connections to Googleapis, you must run export KKZONE=cn again before you proceed with the steps below.

Note

The commands above download the latest release (v1.1.0) of KubeKey. You can change the version number in the command to download a specific version.

Make kk executable:

  1. chmod +x kk

Create an example configuration file with default configurations. Here Kubernetes v1.20.4 is used as an example.

  1. ./kk create config --with-kubesphere v3.1.0 --with-kubernetes v1.20.4

Note

  • Recommended Kubernetes versions for KubeSphere v3.1.0: v1.17.9, v1.18.8, v1.19.8 and v1.20.4. If you do not specify a Kubernetes version, KubeKey will install Kubernetes v1.19.8 by default. For more information about supported Kubernetes versions, see Support Matrix.

  • If you do not add the flag --with-kubesphere in the command in this step, KubeSphere will not be deployed unless you install it using the addons field in the configuration file or add this flag again when you use ./kk create cluster later.

  • If you add the flag --with-kubesphere without specifying a KubeSphere version, the latest version of KubeSphere will be installed.

Deploy KubeSphere and Kubernetes

After you run the commands above, a configuration file config-sample.yaml will be created. Edit the file to add machine information, configure the load balancer and more.

Note

The file name may be different if you customize it.

config-sample.yaml example

  1. ...
  2. spec:
  3.   hosts:
  4.   - {name: master1, address: 172.16.0.4, internalAddress: 172.16.0.4, user: root, password: Testing123}
  5.   - {name: master2, address: 172.16.0.5, internalAddress: 172.16.0.5, user: root, password: Testing123}
  6.   - {name: master3, address: 172.16.0.6, internalAddress: 172.16.0.6, user: root, password: Testing123}
  7.   - {name: worker1, address: 172.16.0.7, internalAddress: 172.16.0.7, user: root, password: Testing123}
  8.   - {name: worker2, address: 172.16.0.8, internalAddress: 172.16.0.8, user: root, password: Testing123}
  9.   - {name: worker3, address: 172.16.0.9, internalAddress: 172.16.0.9, user: root, password: Testing123}
  10.   roleGroups:
  11.     etcd:
  12.     - master1
  13.     - master2
  14.     - master3
  15.     master:
  16.     - master1
  17.     - master2
  18.     - master3
  19.     worker:
  20.     - worker1
  21.     - worker2
  22.     - worker3
  23.   controlPlaneEndpoint:
  24.     domain: lb.kubesphere.local
  25.     address: 172.16.0.10   # The VIP address
  26.     port: 6443
  27. ...

Note

  • Replace the value of controlPlaneEndpoint.address with your own VIP address.
  • For more information about different parameters in this configuration file, see Multi-node Installation.

Start installation

After you complete the configuration, you can execute the following command to start the installation:

  1. ./kk create cluster -f config-sample.yaml

Verify installation

  1. Run the following command to inspect the logs of installation.

    1. kubectl logs -n kubesphere-system $(kubectl get pod -n kubesphere-system -l app=ks-install -o jsonpath='{.items[0].metadata.name}') -f

  2. When you see the following message, it means your HA cluster is successfully created.

    1. #####################################################
    2. ###              Welcome to KubeSphere!           ###
    3. #####################################################
    5. Console: http://172.16.0.4:30880
    6. Account: admin
    7. Password: [email protected]
    9. NOTES：
    10.   1. After logging into the console, please check the
    11.      monitoring status of service components in
    12.      the "Cluster Management". If any service is not
    13.      ready, please wait patiently until all components
    14.      are ready.
    15.   2. Please modify the default password after login.
    17. #####################################################
    18. https://kubesphere.io             2020-xx-xx xx:xx:xx
    19. #####################################################