The nginx-loadbalancer-kubernetes runs in a Kubernetes Cluster and responds to changes in resources of interest, updating designated NGINX Plus hosts with the appropriate configuration.
The controller is deployed in a Kubernetes Cluster. Upon startup, it registers interest in changes to Service resources in the "nginx-ingress" namespace. The Handler accepts the events raised by the Cluster and calls the Translator to convert the events into event definitions that are used to update NGINX Plus hosts. Next, the Handler calls the Synchronizer with the list of translated events which are fanned-out for each NGINX host. Lastly, the Synchronizer calls the NGINX+ Configuration API using the NGINX Plus Go client to update the target NGINX Plus host(s).
stateDiagram-v2
Controller --> Watcher
Controller --> Settings
Watcher --> Handler : "nlk-handler queue"
Handler --> Translator
Translator --> Handler
Handler --> Synchronizer : "nlk-synchronizer queue"
Synchronizer --> BorderClient : "HttpBorderClient | TcpBorderClient"
BorderClient --> NGINXPlusLB1
BorderClient --> NGINXPlusLB2
BorderClient --> NGINXPlusLB...
BorderClient --> NGINXPlusLBn
The Settings module is responsible for loading the configuration settings from the "nlk-config" ConfigMap resource in the "nlk" namespace. The Settings are loaded when the controller starts and are reloaded when the "nlk-config" ConfigMap resource is updated.
The Watcher is responsible for monitoring changes to Service resources in the "nginx-ingress" namespace.
It registers methods that handle each event type. Events are handled by creating a core.Event
instance and adding it to the "nlk-handler" queue.
When adding the event to the queue, the Watcher also retrieves the list of Node IPs and adds the list to the event.
The master node ip is excluded from the list. (NOTE: This should be configurable.)
The Handler is responsible for taking the core.Event
instances from the "nlk-handler" queue and calling the Translator to convert the event into a core.ServerUpdateEvent
instance,
adding each core.ServerUpdateEvent
to the "nlk-synchronizer" queue.
The Translator is responsible for converting the core.Event
event into an nginxClient.UpstreamServer
event.
This involves filtering out the core.Event
instances that are not of interest to the controller by accepting only Port names starting with the NlkPrefix value (currently nlk-).
The event is then fanned-out based on the defined Ports, one event per defined Port. Each port is then augmented with the Ingress name (the name configured in the Port definition with the NlkPrefix value removed),
and the list of the Node's IP addresses.
The Translator passes the list of events to the Synchronizer by calling the AddEvents
method.
NOTE: It is important that the Port names match the name of the defined NGINX Plus Upstreams.
In the following example the NGINX Plus Upstreams are named "nlk-nginx-lb-http" and "nlk-nginx-lb-https". These match the name in the NGINX Plus configuration.
apiVersion: v1
kind: Service
metadata:
name: nginx-ingress
namespace: nginx-ingress
spec:
type: NodePort
ports:
- port: 80
targetPort: 80
protocol: TCP
name: nlk-nginx-lb-http
- port: 443
targetPort: 443
protocol: TCP
name: nlk-nginx-lb-https
selector:
app: nginx-ingress
The Synchronizer is responsible for fanning-out the given list of core.ServerUpdateEvent
events, one for each configured NGINX Plus host.
The NGINX Plus hosts are configured using a ConfigMap resource named "nlk-config" in the "nlk" namespace. An example of the ConfigMap is shown below.
apiVersion: v1
kind: ConfigMap
data:
nginx-hosts:
"http://10.1.1.4:9000/api,http://10.1.1.5:9000/api"
metadata:
name: nlk-config
namespace: nlk
This example includes two NGINX Plus hosts to support High Availability.
Additionally, the Synchronizer is responsible for taking the core.ServerUpdateEvent
instances from the "nlk-synchronizer" queue and updating the target NGINX Plus host.
The Synchronizer uses the NGINX Plus Go client to communicate with each NGINX Plus host.
The Synchronizer uses a retry mechanism to handle failures when updating the NGINX Plus hosts.
The retry mechanism is implemented in the workqueue using the workqueue.NewItemExponentialFailureRateLimiter
,
having defaults set to a base of 2 seconds, and a maximum of 60 seconds.
The Synchronizer uses a jitter mechanism to avoid thrashing the NGINX Plus hosts. Each core.ServerUpdateEvent
instance
is added to the "nlk-synchronizer" queue with a random jitter value between 250 and 750 milliseconds.
- Steve Wagner - Solutions Architect - Community and Alliances @ F5, Inc.
- Chris Akker - Solutions Architect - Community and Alliances @ F5, Inc.