Aws-load-balancer-controller: Add support for rewrite-target

[joseppla]

When creating an ingress controller for multiple backend services all serving on '/' there's no way to rewrite the target URI.

[joshrosso]

@joseppla Can you provide a small example to make sure we're on the same page?

Thanks in advanced.

[joseppla]

Sure,

ingress ex.
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
name: monitoring-ingress
namespace: monitoring
annotations:
alb.ingress.kubernetes.io/scheme: internal
alb.ingress.kubernetes.io/subnets: subnet-1,subnet-2
alb.ingress.kubernetes.io/security-groups: sg-XXXXXXX
alb.ingress.kubernetes.io/healthcheck-path: /
kubernetes.io/ingress.class: "alb"
labels:
app: monitoring
spec:
rules:

• host: k8s-monitoring.domain.com
  http:
  paths:
  
  
  
  • path: /
    
    backend:
    
    serviceName: prometheus
    
    servicePort: 9090
  
  
  • path: /scope
    
    backend:
    
    serviceName: weave-scope-app
    
    servicePort: 80
  
  

if I apply this ingress rule, the requests to 'domain/scope' are redirected to weave-scope-app searching for the '/scope' URI on the service, this URI doesn't exists and the service answers with an error, on Kubernetes standard ingress there're few anotations for modifying this requests:

app-root | Redirect requests without a path (i.e., for /) to this location. (nginx, haproxy, trafficserver)
rewrite-target | Replace matched Ingress path with this value. (nginx, trafficserver)
add-base-url | Add tag to HTML. (nginx)
preserve-host | Whether to pass the client request host (true) or the origin hostname (false) in the HTTP Host field. (trafficserver)

ex.
kubernetes.io/rewrite-target: '/'

Maybe I'm doing something wrong but the ingress seems to redirect to the correct service but the service log's showing a request to a non existent URI.

[cloudbow]

It will hit a non-existent uri. I think we just need to make sure that we serve from the non existent uri. However I am upvoting it as I faced the same issue and its not easy to server someone else url through another hack

[cloudbow]

We can use name based virtual hosting

[vk-mattc]

This is because Application Load Balancers themselves do not support URL rewriting. As I understand it the ALB ingress controller is setting up an ALB with target groups pointing directly at the NodePorts exposed on the workers themselves, which then heads into the kube-proxy service fronting the pods. Because ALBs pass in the URL unmodified this is what leads to the behavior.