Support k8gb behind a reverse proxy #1710

abaguas · 2024-08-25T19:24:04Z

Problem

K8GB reads the IP addresses it exposes from Ingress.Status.LoadBalancer.Ingress or from Service.Status.LoadBalancer.Ingress for ingresses configured with Kubernetes Ingress or Istio Virtual Service resources, respectively.
However, in some setups the clients do not route their traffic to these IP addresses because the cluster is behind a reverse proxy.

Solution

To support this setup, K8GB should expose DNS records with the IP address of the reverse proxy. Since the address is unknown to the cluster, the K8GB administrator must provide it via configuration. This PR adds to K8GB the capability to read IP address from an annotation k8gb.io/exposed-ip-addresses on the GSLB resource.

Example

apiVersion: k8gb.absa.oss/v1beta1
kind: Gslb
metadata:
  annotations:
    k8gb.io/external-ips: "185.199.110.153"

Validation

A ValidatingAdmissionPolicy was added to the helm chart to validate the value of the annotation is a valid IPv4 IP address. If that is the not the case the following error is shown:

The gslbs "failover-playground-istio" is invalid: : ValidatingAdmissionPolicy 'validate-strict-exposed-ip-annotation' with binding 'enforce-strict-exposed-ip-annotation' denied request: The annotation 'k8gb.io/exposed-ip-addresses' must contain a valid IPv4 address

If the ValidatingAdmissionPolicy is disabled the controller will also reject the resource with the following error:

2024-10-31T14:02:36Z ERR github.com/k8gb-io/k8gb/controllers/logging/logr.go:72 > events: Reconciler error {"Gslb":"aba/failover-playground-istio","controller":"gslb","controllerGroup":"k8gb.absa.oss","controllerKind":"Gslb","name":"failover-playground-istio","namespace":"aba","reconcileID":"8523077b-b9bf-4206-922b-036eb8e21a7b"} error="getting load balancer exposed IPs (invalid IP address: 127.0.0.1002)"

Note: with the current testing framework it is rather cumbersome to add a test where the errors above are verified. But if we agree to proceed with chainsaw it will be quite easy: #1758

Fixes #1275

ytsarev

I tested with the ingress local setup and set the annotation on both test-gslb1 and test-gslb2 local clusters.

k -n test-gslb get ing failover-ingress -o yaml
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  annotations:
    k8gb.io/exposed-ip-addresses: 185.199.110.153,185.199.109.153
  labels:
    app: test-gslb-failover
  name: failover-ingress
  namespace: test-gslb
spec:
  ingressClassName: nginx
  rules:
  - host: failover.cloud.example.com
    http:
      paths:
      - backend:
          service:
            name: frontend-podinfo
            port:
              name: http
        path: /
        pathType: Prefix
status:
  loadBalancer:
    ingress:
    - ip: 172.20.0.6
    - ip: 172.20.0.7

 k -n test-gslb get gslb failover-ingress -o yaml
apiVersion: k8gb.absa.oss/v1beta1
kind: Gslb
metadata:
  annotations:
    kubectl.kubernetes.io/last-applied-configuration: |
      {"apiVersion":"k8gb.absa.oss/v1beta1","kind":"Gslb","metadata":{"annotations":{},"name":"failover-ingress","namespace":"test-gslb"},"spec":{"resourceRef":{"apiVersion":"networking.k8s.io/v1","kind":"Ingress","matchLabels":{"app":"test-gslb-failover"}},"strategy":{"primaryGeoTag":"eu","type":"failover"}}}
  creationTimestamp: "2024-10-05T10:59:11Z"
  finalizers:
  - k8gb.absa.oss/finalizer
  generation: 2
  name: failover-ingress
  namespace: test-gslb
  resourceVersion: "53488"
  uid: b9237df5-8742-400b-9102-3d40e401ca4a
spec:
  ingress: {}
  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels:
      app: test-gslb-failover
  strategy:
    dnsTtlSeconds: 30
    primaryGeoTag: eu
    splitBrainThresholdSeconds: 300
    type: failover
status:
  geoTag: eu
  healthyRecords:
    failover.cloud.example.com:
    - 185.199.110.153
    - 185.199.109.153
  hosts: failover.cloud.example.com
  loadBalancer:
    exposedIps:
    - 185.199.110.153
    - 185.199.109.153

dig @localhost -p 5053 failover.cloud.example.com +short +tcp
185.199.110.153
185.199.109.153

The mechanics work as expected 👍

However, I have some design questions.

In case of Ingress, we are setting k8gb.io/exposed-ip-addresses annotation on the referenced Ingress resource - the one that is getting resourceRef in the main Gslb spec, e.g.

  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels:
      app: ingress-referenced

At the same time in case of Istio scenario, we are setting the k8gb.io/exposed-ip-addresses annotation on the Service v1 kind and it is not obvious because what is getting referenced in Gslb spec in VirtualService resource, e.g.

spec:
  resourceRef:
    apiVersion: networking.istio.io/v1
    kind: VirtualService
    matchLabels:
      app: istio

So it looks like we have some form of implementation/abstraction leak here.

I have a slightly modified design proposal: what if we set k8gb.io/exposed-ip-addresses annotation on the main Gslb resource spec itself, this way we just override any referenced network resource and make it independent from the specific underlying referenced resource

apiVersion: k8gb.absa.oss/v1beta1
kind: Gslb
metadata:
  name: test-gslb-failover
  namespace: test-gslb
  annotations:
     k8gb.io/exposed-ip-addresses: 185.199.110.153,185.199.109.153
spec:
  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels: # ingresses.networking.k8s.io resource selector
      app: test-gslb-failover
  strategy:
    type: failover 
    primaryGeoTag:

@abaguas What do you think?

abaguas · 2024-10-08T20:43:06Z

I tested with the ingress local setup and set the annotation on both test-gslb1 and test-gslb2 local clusters.

k -n test-gslb get ing failover-ingress -o yaml
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  annotations:
    k8gb.io/exposed-ip-addresses: 185.199.110.153,185.199.109.153
  labels:
    app: test-gslb-failover
  name: failover-ingress
  namespace: test-gslb
spec:
  ingressClassName: nginx
  rules:
  - host: failover.cloud.example.com
    http:
      paths:
      - backend:
          service:
            name: frontend-podinfo
            port:
              name: http
        path: /
        pathType: Prefix
status:
  loadBalancer:
    ingress:
    - ip: 172.20.0.6
    - ip: 172.20.0.7

 k -n test-gslb get gslb failover-ingress -o yaml
apiVersion: k8gb.absa.oss/v1beta1
kind: Gslb
metadata:
  annotations:
    kubectl.kubernetes.io/last-applied-configuration: |
      {"apiVersion":"k8gb.absa.oss/v1beta1","kind":"Gslb","metadata":{"annotations":{},"name":"failover-ingress","namespace":"test-gslb"},"spec":{"resourceRef":{"apiVersion":"networking.k8s.io/v1","kind":"Ingress","matchLabels":{"app":"test-gslb-failover"}},"strategy":{"primaryGeoTag":"eu","type":"failover"}}}
  creationTimestamp: "2024-10-05T10:59:11Z"
  finalizers:
  - k8gb.absa.oss/finalizer
  generation: 2
  name: failover-ingress
  namespace: test-gslb
  resourceVersion: "53488"
  uid: b9237df5-8742-400b-9102-3d40e401ca4a
spec:
  ingress: {}
  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels:
      app: test-gslb-failover
  strategy:
    dnsTtlSeconds: 30
    primaryGeoTag: eu
    splitBrainThresholdSeconds: 300
    type: failover
status:
  geoTag: eu
  healthyRecords:
    failover.cloud.example.com:
    - 185.199.110.153
    - 185.199.109.153
  hosts: failover.cloud.example.com
  loadBalancer:
    exposedIps:
    - 185.199.110.153
    - 185.199.109.153

dig @localhost -p 5053 failover.cloud.example.com +short +tcp
185.199.110.153
185.199.109.153

The mechanics work as expected 👍

However, I have some design questions.

In case of Ingress, we are setting k8gb.io/exposed-ip-addresses annotation on the referenced Ingress resource - the one that is getting resourceRef in the main Gslb spec, e.g.

  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels:
      app: ingress-referenced

At the same time in case of Istio scenario, we are setting the k8gb.io/exposed-ip-addresses annotation on the Service v1 kind and it is not obvious because what is getting referenced in Gslb spec in VirtualService resource, e.g.

spec:
  resourceRef:
    apiVersion: networking.istio.io/v1
    kind: VirtualService
    matchLabels:
      app: istio

So it looks like we have some form of implementation/abstraction leak here.

I have a slightly modified design proposal: what if we set k8gb.io/exposed-ip-addresses annotation on the main Gslb resource spec itself, this way we just override any referenced network resource and make it independent from the specific underlying referenced resource

apiVersion: k8gb.absa.oss/v1beta1
kind: Gslb
metadata:
  name: test-gslb-failover
  namespace: test-gslb
  annotations:
     k8gb.io/exposed-ip-addresses: 185.199.110.153,185.199.109.153
spec:
  resourceRef:
    apiVersion: networking.k8s.io/v1
    kind: Ingress
    matchLabels: # ingresses.networking.k8s.io resource selector
      app: test-gslb-failover
  strategy:
    type: failover 
    primaryGeoTag:

@abaguas What do you think?

I added it to the Ingress and Service resources because those are the ones carrying the IP address information. So we would override it there.

I agree with your point, in many cases the user won't know the technical details of the ingress or k8gb to know where to place the annotation. Setting it on the GSLB also avoid annotating a resource that we do not control, which may save us from unexpected corner cases. Let's proceed with your proposal

ytsarev

I tested the happy path scenario of setting k8gb.io/exposed-ip-addresses to valid IP values. Everything worked perfectly with the valid values 👍

Concern:

I can set annotation to something like k8gb.io/exposed-ip-addresses: 192.169.0.test propagating invalid value.

In this case the invalid value is getting propagated as is to both Gslb

k -n test-gslb get gslb failover-ingress -o yaml
...
status:
  geoTag: eu
  healthyRecords:
    failover.cloud.example.com:
    - 192.169.0.test
  hosts: failover.cloud.example.com
  loadBalancer:
    exposedIps:
    - 192.169.0.test

And underlying DNSEndpoint

k -n test-gslb get dnsendpoints failover-ingress -o yaml
...
spec:
  endpoints:
  - dnsName: localtargets-failover.cloud.example.com
    recordTTL: 30
    recordType: A
    targets:
    - 192.169.0.test
  - dnsName: failover.cloud.example.com
    labels:
      strategy: failover
    recordTTL: 30
    recordType: A
    targets:
    - 192.169.0.test

Apart from undesired misconfiguration, it could be used as a security attack vector.
We are missing some validation here. It's probably better to catch it asap at annotation evaluation, maybe by emitting some Failed event

ytsarev · 2024-11-07T13:19:12Z

@abaguas could you please resolve the conflict? It looks like it happened after merging another PR :)

Problem K8GB reads IP addresses from `Ingress.Status.LoadBalancer.Ingress` or from `Service.Status.LoadBalancer.Ingress` for ingress configured with Kubernetes Ingress and Istio Virtual Service, respectively. The IP addresses exposed by these resources are the IP addresses exposed by the Kubernetes Cluster. However, in some setups the clients do not route their traffic to these IP addresses because the cluster is behind a reverse proxy. Solution To support this setup, K8GB should expose DNS records with the IP address of the reverse proxy. Since the address is unknown to the cluster the K8GB administrator must provide it via configuration. This PR adds to K8GB the capability to read IP address from an annotation `k8gb.io/external-ips` on the GSLB resource. Example ``` apiVersion: k8gb.absa.oss/v1beta1 kind: Gslb metadata: labels: app: ingress annotations: k8gb.io/external-ips: "185.199.110.153" ``` Fixes k8gb-io#1275 Signed-off-by: Andre Baptista Aguas <[email protected]>

Signed-off-by: Andre Aguas <[email protected]>

abaguas · 2024-11-07T15:04:45Z

@abaguas could you please resolve the conflict? It looks like it happened after merging another PR :)

@ytsarev done

abaguas requested review from donovanmuller, k0da, kuritka, ytsarev and jkremser as code owners August 25, 2024 19:24

abaguas force-pushed the reverse/proxy branch from 3dc180d to 8274656 Compare August 25, 2024 19:25

abaguas marked this pull request as draft August 25, 2024 19:26

abaguas force-pushed the reverse/proxy branch from 8274656 to d5d76c0 Compare September 24, 2024 18:02

abaguas marked this pull request as ready for review September 24, 2024 18:26

ytsarev reviewed Oct 6, 2024

View reviewed changes

abaguas force-pushed the reverse/proxy branch 2 times, most recently from af3314d to 64327c1 Compare October 13, 2024 19:56

abaguas requested a review from ytsarev October 17, 2024 19:37

ytsarev requested changes Oct 26, 2024

View reviewed changes

abaguas force-pushed the reverse/proxy branch 4 times, most recently from 78af2f6 to 95d5f02 Compare October 31, 2024 14:44

abaguas requested a review from ytsarev October 31, 2024 15:19

abaguas added 2 commits November 7, 2024 15:37

Validate k8gb.io/exposed-ip-addresses annotation

960e7e9

Signed-off-by: Andre Aguas <[email protected]>

abaguas force-pushed the reverse/proxy branch from 95d5f02 to 960e7e9 Compare November 7, 2024 14:38

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support k8gb behind a reverse proxy #1710

Support k8gb behind a reverse proxy #1710

abaguas commented Aug 25, 2024 •

edited

Loading

ytsarev left a comment •

edited

Loading

abaguas commented Oct 8, 2024

ytsarev left a comment •

edited

Loading

ytsarev commented Nov 7, 2024

abaguas commented Nov 7, 2024

Support k8gb behind a reverse proxy #1710

Are you sure you want to change the base?

Support k8gb behind a reverse proxy #1710

Conversation

abaguas commented Aug 25, 2024 • edited Loading

Problem

Solution

Example

Validation

ytsarev left a comment • edited Loading

Choose a reason for hiding this comment

abaguas commented Oct 8, 2024

ytsarev left a comment • edited Loading

Choose a reason for hiding this comment

ytsarev commented Nov 7, 2024

abaguas commented Nov 7, 2024

abaguas commented Aug 25, 2024 •

edited

Loading

ytsarev left a comment •

edited

Loading

ytsarev left a comment •

edited

Loading