Kube-router is built around concept of watchers and controllers. Watchers use Kubernetes watch API to get notification on events related to create, update, delete of Kubernetes objects. Each watcher gets notification related to a particular API object. On receiving an event from API server, watcher broadcasts events. Controller registers to get event updates from the watchers and act up on the events.
Kube-router consists of 3 core controllers and multiple watchers as depicted in below diagram.
Each of the controller follows below structure
func Run() {
for {
Sync() // control loop that runs for ever and perfom sync at periodic interval
}
}
func OnUpdate() {
Sync() // on receiving update of a watched API object (namespace, node, pod, network policy etc)
}
Sync() {
//re-concile any state changes
}
Cleanup() {
// cleanup any changes (to iptables, ipvs, network etc) done to the system
}
Network services controller is responsible for reading the services and endpoints information from Kubernetes API server and configure IPVS on each cluster node accordingly.
Please our read blog for design details and pros and cons compared to iptables based Kube-proxy https://cloudnativelabs.github.io/post/2017-05-10-kube-network-service-proxy/
Demo of Kube-router's IPVS based Kubernetes network service proxy
Features:
- round robin load balancing
- client IP based session persistence
- source IP is preserved if service controller is used in conjuction with network routes controller (kube-router with --run-router flag)
- option to explicitly masquerade (SNAT) with --masquerade-all flag
Network policy controller is responsible for reading the namespace, network policy and pods information from Kubernetes API server and configure iptables accordingly to provide ingress filter to the pods.
Kube-router supports the networking.k8s.io/NetworkPolicy API or network policy V1/GA semantics and also network policy beta semantics.
Please read blog for design details of Network Policy controller https://cloudnativelabs.github.io/post/2017-05-1-kube-network-policies/
Demo of Kube-router's iptables based implementaton of network policies
Network routes controller is responsible for reading pod CIDR allocated by controller manager to the node, and advertises the routes to the rest of the nodes in the cluster (BGP peers). Use of BGP is transperent to user for basic pod-to-pod networking.
However BGP can be leveraged to other use cases like advertising the cluster ip, routable pod ip etc. Only in such use-cases understanding of BGP and configuration is required. Please see below demo how kube-router advertises cluster IP and pod cidrs to external BGP router
The best way to get started is to deploy Kubernetes with Kube-router is with a cluster installer.
Please see the steps to deploy Kubernetes cluster with Kube-router using Kops
Please see the steps to deploy Kubernetes cluster with Kube-router using bootkube
Please see the steps to deploy Kubernetes cluster with Kube-router using Kubeadm
Depending on what functionality of kube-router you want to use, multiple deployment options are possible. You can use the flags --run-firewall, --run-router, --run-service-proxy to selectively enable only required functionality of kube-router.
Also you can choose to run kube-router as agent running on each cluster node. Alternativley you can run kube-router as pod on each node through daemonset.
Usage of ./kube-router:
--advertise-cluster-ip Add Cluster IP of the service to the RIB so that it gets advertises to the BGP peers.
--advertise-external-ip Add External IP of service to the RIB so that it gets advertised to the BGP peers.
--cleanup-config Cleanup iptables rules, ipvs, ipset configuration and exit.
--cluster-asn uint ASN number under which cluster nodes will run iBGP.
--cluster-cidr string CIDR range of pods in the cluster. It is used to identify traffic originating from and destinated to pods.
--config-sync-period duration The delay between apiserver configuration synchronizations (e.g. '5s', '1m'). Must be greater than 0. (default 1m0s)
--enable-overlay When enable-overlay set to true, IP-in-IP tunneling is used for pod-to-pod networking across nodes in different subnets. When set to false no tunneling is used and routing infrastrcture is expected to route traffic for pod-to-pod networking across nodes in different subnets (default true)
--enable-pod-egress SNAT traffic from Pods to destinations outside the cluster. (default true)
--hairpin-mode Add iptable rules for every Service Endpoint to support hairpin traffic.
-h, --help Print usage information.
--hostname-override string Overrides the NodeName of the node. Set this if kube-router is unable to determine your NodeName automatically.
--iptables-sync-period duration The delay between iptables rule synchronizations (e.g. '5s', '1m'). Must be greater than 0. (default 1m0s)
--ipvs-sync-period duration The delay between ipvs config synchronizations (e.g. '5s', '1m', '2h22m'). Must be greater than 0. (default 1m0s)
--kubeconfig string Path to kubeconfig file with authorization information (the master location is set by the master flag).
--masquerade-all SNAT all traffic to cluster IP/node port.
--master string The address of the Kubernetes API server (overrides any value in kubeconfig).
--nodeport-bindon-all-ip For service of NodePort type create IPVS service that listens on all IP's of the node.
--nodes-full-mesh Each node in the cluster will setup BGP peering with rest of the nodes. (default true)
--peer-router-asns uintSlice ASN numbers of the BGP peer to which cluster nodes will advertise cluster ip and node's pod cidr. (default [])
--peer-router-ips ipSlice The ip address of the external router to which all nodes will peer and advertise the cluster ip and pod cidr's. (default [])
--peer-router-passwords stringSlice Password for authenticating against the BGP peer defined with "--peer-router-ips".
--routes-sync-period duration The delay between route updates and advertisements (e.g. '5s', '1m', '2h22m'). Must be greater than 0. (default 1m0s)
--run-firewall Enables Network Policy -- sets up iptables to provide ingress firewall for pods. (default true)
--run-router Enables Pod Networking -- Advertises and learns the routes to Pods via iBGP. (default true)
--run-service-proxy Enables Service Proxy -- sets up IPVS for Kubernetes Services. (default true)```
-
Kube-router need to access kubernetes API server to get information on pods, services, endpoints, network policies etc. The very minimum information it requires is the details on where to access the kubernetes API server. This information can be passed as
kube-router --master=http://192.168.1.99:8080/orkube-router --kubeconfig=<path to kubeconfig file>. -
If you run kube-router as agent on the node, ipset package must be installed on each of the nodes (when run as daemonset, container image is prepackaged with ipset)
-
If you choose to use kube-router for pod-to-pod network connectivity then Kubernetes controller manager need to be configured to allocate pod CIDRs by passing
--allocate-node-cidrs=trueflag and providing acluster-cidr(i.e. by passing --cluster-cidr=10.1.0.0/16 for e.g.) -
If you choose to run kube-router as daemonset, then both kube-apiserver and kubelet must be run with
--allow-privileged=trueoption -
If you choose to use kube-router for pod-to-pod network connecitvity then Kubernetes cluster must be configured to use CNI network plugins. On each node CNI conf file is expected to be present as /etc/cni/net.d/10-kuberouter.conf .
bridgeCNI plugin andhost-localfor IPAM should be used. A sample conf file that can be downloaded aswget -O /etc/cni/net.d/10-kuberouter.conf https://raw.githubusercontent.com/cloudnativelabs/kube-router/master/cni/10-kuberouter.conf
This is quickest way to deploy kube-router (dont forget to ensure the requirements). Just run
kubectl apply -f https://raw.githubusercontent.com/cloudnativelabs/kube-router/master/daemonset/kube-router-all-service-daemonset.yaml
Above will run kube-router as pod on each node automatically. You can change the arguments in the daemonset definition as required to suit your needs. Some samples can be found at https://github.com/cloudnativelabs/kube-router/tree/master/daemonset with different argument to select set of the services kube-router should run.
You can choose to run kube-router as agent runnng on each node. For e.g if you just want kube-router to provide ingress firewall for the pods then you can start kube-router as
kube-router --master=http://192.168.1.99:8080/ --run-firewall=true --run-service-proxy=false --run-router=false
You can clean up all the configurations done (to ipvs, iptables, ip routes) by kube-router on the node by running
kube-router --cleanup-config
If you have a kube-proxy in use, and want to try kube-router just for service proxy you can do
kube-proxy --cleanup-iptables
followed by
kube-router --master=http://192.168.1.99:8080/ --run-service-proxy=true --run-firewall=false --run-router=false
and if you want to move back to kube-proxy then clean up config done by kube-router by running
kube-router --cleanup-config
and run kube-proxy with the configuration you have.
Communication from a Pod that is behind a Service to its own ClusterIP:Port is
not supported by default. However, It can be enabled per-service by adding the
kube-router.io/service.hairpin= annotation, or for all Services in a cluster by
passing the flag --hairpin-mode=true to kube-router.
Additionally, the hairpin_mode sysctl option must be set to 1 for all veth
interfaces on each node. This can be done by adding the "hairpinMode": true
option to your CNI configuration and rebooting all cluster nodes if they are
already running kubernetes.
Hairpin traffic will be seen by the pod it originated from as coming from the Service ClusterIP if it is logging the source IP.
10-kuberouter.conf
{
"name":"mynet",
"type":"bridge",
"bridge":"kube-bridge",
"isDefaultGateway":true,
"hairpinMode":true,
"ipam": {
"type":"host-local"
}
}To enable hairpin traffic for Service my-service:
kubectl annotate service my-service "kube-router.io/service.hairpin="
Please read below blog on how to user DSR in combination with --advertise-external-ip to build highly scalable and available ingress.
https://cloudnativelabs.github.io/post/2017-11-01-kube-high-available-ingress/
You can enable DSR(Direct Server Return) functionality per service. When enabled service endpoint will directly respond to the client by passign the service proxy. When DSR is enabled Kube-router will uses LVS's tunneling mode to achieve this.
To enable DSR you need to annotate service with kube-router.io/service.dsr=tunnel annotation. For e.g.
kubectl annotate service my-service "kube-router.io/service.dsr=tunnel"
In the current implementation when annotation is applied on the service, DSR will be applicable only to the external IP's.
Also when DSR is used, current implementation does not support port remapping. So you need to use same port and target port for the service
You will need to enable hostIPC: true and hostPID: true in kube-router daemonset manifest.
Also host path /var/run/docker.sock must be made a volumemount to kube-router.
Above changes are required for kube-router to enter pod namespeace and create ipip tunnel in the pod and to assign the external IP to the VIP.
For an e.g manifest please look at manifest with DSR requirements enabled.
Kube-router uses LVS for service proxy. LVS support rich set of scheduling alogirthms. You can annotate
the service to choose one of the scheduling alogirthms. When a service is not annotated round-robin scheduler is selected by default
For least connection scheduling use:
kubectl annotate service my-service "kube-router.io/service.scheduler=lc"
For round-robin scheduling use:
kubectl annotate service my-service "kube-router.io/service.scheduler=rr"
For source hashing scheduling use:
kubectl annotate service my-service "kube-router.io/service.scheduler=sh"
For destination hashing scheduling use:
kubectl annotate service my-service "kube-router.io/service.scheduler=dh"