Keepalived-VIP

Kubernetes presents keepalived as a tool to create a Virtual IP address (VIP).

In this discussion, we will unravel how to effectively use IPVS - The Linux Virtual Server Project to configure a VIP for Kubernetes.

Prelude

The v1.6 version of Kubernetes provides three modes to expose a Service:

L4 LoadBalancer : This can only be utilized on cloud providers like GCE or AWS.
NodePort : NodePort allows the opening of a port on each node. This port then routes the request to a randomly selected pod.
L7 Ingress : Ingress serves as a LoadBalancer (for instance, nginx, HAProxy, traefik, vulcand) which directs HTTP/HTTPS requests to the corresponding service endpoint.

So if we've got all these ways, why do we need keepalived?


                                             ___________________
                                            |                   |
                                          |-----| Host IP: 10.4.0.3 |
                                          |     |___________________|
                                          |
                                          |     ___________________
                                          |    |                   |
Public ----(example.com = 10.4.0.3/4/5)----|-----| Host IP: 10.4.0.4 |
                                          |    |___________________|
                                          |
                                          |     ___________________
                                          |    |                   |
                                          |-----| Host IP: 10.4.0.5 |
                                                |___________________|

Let's suppose that the Ingress operates on 3 Kubernetes nodes, exposing the 10.4.0.x IP for load balancing purposes.

If the DNS Round Robin (RR) cycles the requests corresponding to example.com to these 3 nodes and 10.4.0.3 crashes, a third of the traffic will still be directed towards 10.4.0.3. This causes a downtime, until the DNS identifies the failure and corrects the direction.

Strictly speaking, this doesn't truly offer High Availability (HA).

Here, IPVS comes to our rescue by associating each service with a VIP and exposing the VIP outside the Kubernetes cluster.

The difference with service-loadbalancer and ingress-nginx

Looking at the diagram below,

                                              ___________________
                                             |                   |
                                             | VIP: 10.4.0.50    |
                                       |-----| Host IP: 10.4.0.3 |
                                       |    | Role: Master      |
                                       |    |___________________|
                                       |
                                       |     ___________________
                                       |    |                   |
                                       |    | VIP: Unassigned   |
Public ----(example.com = 10.4.0.50)----|-----| Host IP: 10.4.0.4 |
                                       |    | Role: Slave       |
                                       |    |___________________|
                                       |
                                       |     ___________________
                                       |    |                   |
                                       |    | VIP: Unassigned   |
                                       |-----| Host IP: 10.4.0.5 |
                                             | Role: Slave       |
                                             |___________________|

Only one node (selected by VRRP) is chosen as the Master, and the VIP is 10.4.0.50. If 10.4.0.3 fails, another node from the remaining ones becomes Master and takes on the VIP, ensuring the true implementation of HA.

Environment Requirements

All that's needed is to ensure that the Kubernetes cluster running keepalived-vip has a normal DaemonSets feature.

RBAC

As Kubernetes introduced the concept of RBAC post version 1.6, we first need to set the rule. For detailed information regarding RBAC, please refer to the guide.

vip-rbac.yaml

... (Please refer to original text for better code understanding)

clusterrolebinding.yaml

... (Please refer to original text for better code understanding)

$ kubectl create -f vip-rbac.yaml
$ kubectl create -f clusterrolebinding.yaml

Example

Firstly, create a simple service.

nginx-deployment.yaml

... (Please refer to original text for better code understanding)

The primary task is to get the pod to listen to port 80, then open the service NodePort monitoring 30320.

$ kubecrl create -f nginx-deployment.yaml

Next, we focus on the config map.

... (Please refer to original text for better code understanding)

Do make a note, 10.87.2.50 must be replaced with an unused IP from your own network segment, e.g., 10.87.2.X. nginx is the service name, and this can be changed accordingly.

Following the confirmation,

... (Please refer to original text for better code understanding)

The next step is to set up keepalived-vip.

... (Please refer to original text for better code understanding)

Establish the daemonset

... (Please refer to original text for better code understanding)

Check the configuration status

... (Please refer to original text for better code understanding)

You can randomly select a pod to inspect its configuration

... (Please refer to original text for better code understanding)

Finally, test the feature

... (Please refer to original text for better code understanding)

10.87.2.50:80 (our hypothetical VIP, as no node actually uses this IP) can now help us direct this service.

All the codes mentioned above can be found here.

Reference Documents

上一页Traefik Ingress 下一页Cloud Provider

最后更新于11个月前