README
¶
kube-keepalived-vip
Kubernetes Virtual IP address/es using keepalived
AKA "how to set up virtual IP addresses in kubernetes using IPVS - The Linux Virtual Server Project".
Overview
kubernetes v1.6 offers 3 ways to expose a service:
- L4 LoadBalancer: Available only on cloud providers such as GCE and AWS
- Service via NodePort: The NodePort directive allocates a port on every worker node, which proxy the traffic to the respective Pod.
- L7 Ingress: The Ingress is a dedicated loadbalancer (eg. nginx, HAProxy, traefik, vulcand) that redirects incoming HTTP/HTTPS traffic to the respective endpoints
If this works, 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 assume that Ingress-es are run on the 3 kubernetes worker nodes above 10.4.0.x, which are exposed to the public to load-balance incoming traffic. DNS Round Robin (RR) is applied to example.com to rotate between the 3 nodes. If 10.4.0.3 goes down, one-third of the traffic to example.com is still directed to the downed node (due to DNS RR). The sysadmin has to step in and delist the faulty node from example.com. Since there will be intermittent downtime until the sysadmin intervenes, this isn't true High Availability (HA).
Here is where IPVS can help.
The idea is to expose a Virtual IP (VIP) address per service, outside of the kubernetes cluster. keepalived then uses VRRP to sync this "mapping" in the local network. With 2 or more instance of the pod running in the cluster is possible to provide HA using a single VIP address.
keepalived should be considered a complement to, and not a replacement for HAProxy or nginx. The goal is to provide robust HA, such that no downtime is experienced if one or more nodes go offline. To be exact, keepalived ensures that the VIP, which exposes a service-loadbalancer or an Ingress, is always owned by a live node. The DNS record will simply point to this single VIP (ie. sans RR) and the failover will be handled entirely by keepalived.
___________________
| |
| VIP: 10.4.0.50 |
|-----| Host IP: 10.4.0.3 |
| | Role: Master |
| |___________________|
|
| ___________________
| | |
| | VIP: Unassigned |
Public ----(example.com = 10.4.0.50)----|-----| Host IP: 10.4.0.3 |
| | Role: Slave |
| |___________________|
|
| ___________________
| | |
| | VIP: Unassigned |
|-----| Host IP: 10.4.0.3 |
| Role: Slave |
|___________________|
In the above diagram, one node assumes the role of a Master (negotiated via VRRP), and assumes the VIP. example.com points only to the shared VIP 10.4.0.50, instead of the 3 nodes. If 10.4.0.3 is taken offline, the surviving hosts elect a new master to assume the VIP. This model of HA ensures that the VIP can be reached at all times.
Requirements
The only requirement is for DaemonSets to be enabled. Check this guide to include the kube-apiserver flags for this to work.
Configuration
To expose one or more services use the flag services-configmap. The format of the data is: external IP -> namespace/serviceName. Optionally it is possible to specify forwarding method using : after the service name. The valid options are NAT and DR. For instance external IP -> namespace/serviceName:DR. By default, if the method is not specified it will use NAT. If the service name is left blank, only the VIP will be assigned and no routing will be done. This is useful e.g. if you run HAProxy in another pod on the same machines with hostnetwork in order to forward incoming smtp requests via proxy protocol to postfix.
This IP must be routable within the LAN and must be available. By default the IP address of the pods is used to route the traffic. This means that is one pod dies or a new one is created by a scale event the keepalived configuration file will be updated and reloaded.
Example
Launch the sample app "echoheaders"
First, we create a new ReplicationController and a Service for a sample app.
$ kubectl create -f examples/echoheaders.yaml
replicationcontroller "echoheaders" created
You have exposed your service on an external port on all nodes in your
cluster. If you want to expose this service to the external internet, you may
need to set up firewall rules for the service port(s) (tcp:30302) to serve traffic.
See http://releases.k8s.io/HEAD/docs/user-guide/services-firewalls.md for more details.
service "echoheaders" created
(Optional) Install the RBAC policies
If you enabled RBAC in your cluster (ie. kube-apiserver runs with the --authorization-mode=RBAC flag), please follow this section so that keepalived can properly query the cluster's API endpoints.
Create a service account so that keepalived can authenticate with kube-apiserver.
kubectl create sa kube-keepalived-vip
Configure the DaemonSet in vip-daemonset.yaml to use the ServiceAccount. Add the serviceAccount to the file as shown:
spec:
hostNetwork: true
serviceAccount: kube-keepalived-vip
containers:
- image: gcr.io/google_containers/kube-keepalived-vip:0.11
Configure its ClusterRole. keepalived needs to read the pods, nodes, endpoints and services.
echo 'apiVersion: rbac.authorization.k8s.io/v1alpha1
kind: ClusterRole
metadata:
name: kube-keepalived-vip
rules:
- apiGroups: [""]
resources:
- pods
- nodes
- endpoints
- services
- configmaps
verbs: ["get", "list", "watch"]' | kubectl create -f -
Configure its ClusterRoleBinding. This binds the above ClusterRole to the kube-keepalived-vip ServiceAccount.
apiVersion: rbac.authorization.k8s.io/v1alpha1
kind: ClusterRoleBinding
metadata:
name: kube-keepalived-vip
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: kube-keepalived-vip
subjects:
- kind: ServiceAccount
name: kube-keepalived-vip
namespace: default
Load the keepalived DaemonSet
Next add the required annotation to expose the service using a local IP
$ echo "apiVersion: v1
kind: ConfigMap
metadata:
name: vip-configmap
data:
10.4.0.50: default/echoheaders" | kubectl create -f -
Now the creation of the daemonset
$ kubectl create -f vip-daemonset.yaml
daemonset "kube-keepalived-vip" created
$ kubectl get daemonset
NAME CONTAINER(S) IMAGE(S) SELECTOR NODE-SELECTOR
kube-keepalived-vip kube-keepalived-vip gcr.io/google_containers/kube-keepalived-vip:0.11 name in (kube-keepalived-vip) type=worker
Note: the DaemonSet yaml file contains a node selector. This is not required, is just an example to show how is possible to limit the nodes where keepalived can run
Debug the deployment
To verify if everything is working we should check if a kube-keepalived-vip pod is in each node of the cluster
Check the labels of the nodes.
$ kubectl get nodes
NAME LABELS STATUS AGE
10.4.0.3 kubernetes.io/hostname=10.4.0.3,type=worker Ready 1d
10.4.0.4 kubernetes.io/hostname=10.4.0.4,type=worker Ready 1d
10.4.0.5 kubernetes.io/hostname=10.4.0.5,type=worker Ready 1d
Check that there's a pod running on each node
$ kubectl get pods
NAME READY STATUS RESTARTS AGE
echoheaders-co4g4 1/1 Running 0 5m
kube-keepalived-vip-a90bt 1/1 Running 0 53s
kube-keepalived-vip-g3nku 1/1 Running 0 52s
kube-keepalived-vip-gd18l 1/1 Running 0 54s
keepalived's logs should look like this if no error was encountered.
$ kubectl logs kube-keepalived-vip-a90bt
I0410 14:24:45.860119 1 keepalived.go:161] cleaning ipvs configuration
I0410 14:24:45.873095 1 main.go:109] starting LVS configuration
I0410 14:24:45.894664 1 main.go:119] starting keepalived to announce VIPs
Starting Healthcheck child process, pid=17
Starting VRRP child process, pid=18
Initializing ipvs 2.6
Registering Kernel netlink reflector
Registering Kernel netlink reflector
Registering Kernel netlink command channel
Registering gratuitous ARP shared channel
Registering Kernel netlink command channel
Using LinkWatch kernel netlink reflector...
Using LinkWatch kernel netlink reflector...
I0410 14:24:56.017590 1 keepalived.go:151] reloading keepalived
Got SIGHUP, reloading checker configuration
Registering Kernel netlink reflector
Initializing ipvs 2.6
Registering Kernel netlink command channel
Registering gratuitous ARP shared channel
Registering Kernel netlink reflector
Opening file '/etc/keepalived/keepalived.conf'.
Registering Kernel netlink command channel
Opening file '/etc/keepalived/keepalived.conf'.
Using LinkWatch kernel netlink reflector...
VRRP_Instance(vips) Entering BACKUP STATE
Using LinkWatch kernel netlink reflector...
Activating healthchecker for service [10.2.68.5]:8080
VRRP_Instance(vips) Transition to MASTER STATE
VRRP_Instance(vips) Entering MASTER STATE
VRRP_Instance(vips) using locally configured advertisement interval (1000 milli-sec)
keepalived's configuration is empty at the start. It should automatically be updated to reflect the current setup.
$ kubectl exec kube-keepalived-vip-a90bt cat /etc/keepalived/keepalived.conf
global_defs {
vrrp_version 3
vrrp_iptables KUBE-KEEPALIVED-VIP
}
vrrp_instance vips {
state BACKUP
interface eth1
virtual_router_id 50
priority 100
nopreempt
advert_int 1
track_interface {
eth1
}
virtual_ipaddress {
172.17.4.90
}
}
# Service: default/echoheaders
virtual_server 10.4.0.50 80 {
delay_loop 5
lvs_sched wlc
lvs_method NAT
persistence_timeout 1800
protocol TCP
real_server 10.2.68.5 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
}
Test that the app can be reached via the VIP 10.4.0.50.
$ curl -v 10.4.0.50
* Rebuilt URL to: 10.4.0.50/
* Trying 10.4.0.50...
* Connected to 10.4.0.50 (10.4.0.50) port 80 (#0)
> GET / HTTP/1.1
> Host: 10.4.0.50
> User-Agent: curl/7.43.0
> Accept: */*
>
* HTTP 1.0, assume close after body
< HTTP/1.0 200 OK
< Server: BaseHTTP/0.6 Python/3.5.0
< Date: Wed, 30 Dec 2015 19:52:39 GMT
<
CLIENT VALUES:
client_address=('10.4.0.148', 52178) (10.4.0.148)
command=GET
path=/
real path=/
query=
request_version=HTTP/1.1
SERVER VALUES:
server_version=BaseHTTP/0.6
sys_version=Python/3.5.0
protocol_version=HTTP/1.0
HEADERS RECEIVED:
Accept=*/*
Host=10.4.0.50
User-Agent=curl/7.43.0
* Closing connection 0
Scaling the replication controller should automatically update and reload keepalived.
$ kubectl scale --replicas=5 replicationcontroller echoheaders
replicationcontroller "echoheaders" scaled
The latest config should reflect something similar to this after scaling up the app.
$ kubectl exec kube-keepalived-vip-a90bt cat /etc/keepalived/keepalived.conf
global_defs {
vrrp_version 3
vrrp_iptables KUBE-KEEPALIVED-VIP
}
vrrp_instance vips {
state BACKUP
interface eth1
virtual_router_id 50
priority 100
nopreempt
advert_int 1
track_interface {
eth1
}
virtual_ipaddress {
172.17.4.90
}
}
# Service: default/echoheaders
virtual_server 10.4.0.50 80 {
delay_loop 5
lvs_sched wlc
lvs_method NAT
persistence_timeout 1800
protocol TCP
real_server 10.2.68.5 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
real_server 10.2.68.6 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
real_server 10.2.68.7 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
real_server 10.2.68.8 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
real_server 10.2.68.9 8080 {
weight 1
TCP_CHECK {
connect_port 8080
connect_timeout 3
}
}
}
Related projects
Documentation
¶
There is no documentation for this package.
Source Files