Troubleshooting

Debug

Using the flag --v=XX it is possible to increase the level of logging. In particular:

--v=3 shows details about the service, Ingress rule, and endpoint changes

Authentication to the Kubernetes API Server

A number of components are involved in the authentication process and the first step is to narrow down the source of the problem, namely whether it is a problem with service authentication or with the kubeconfig file. Both authentications must work:

+-------------+   service          +------------+
|             |   authentication   |            |
+  apiserver  +<-------------------+  ingress   |
|             |                    | controller |
+-------------+                    +------------+

Service authentication

The Ingress controller needs information from API server to configure Kong. Therefore, authentication is required, which can be achieved in two different ways:

Service Account: This is recommended because nothing has to be configured. The Ingress controller will use information provided by the system to communicate with the API server. See ‘Service Account’ section for details.
Kubeconfig file: In some Kubernetes environments service accounts are not available. In this case, a manual configuration is required. The Ingress controller binary can be started with the --kubeconfig flag. The value of the flag is a path to a file specifying how to connect to the API server. Using the --kubeconfig does not require the flag --apiserver-host. The format of the file is identical to ~/.kube/config which is used by kubectl to connect to the API server. See ‘kubeconfig’ section for details.

Discovering API server

Using this flag --apiserver-host=http://localhost:8080, it is possible to specify an unsecured API server or reach a remote Kubernetes cluster using kubectl proxy. Please do not use this approach in production.

In the diagram below you can see the full authentication flow with all options, starting with the browser on the lower left hand side.

Kubernetes                                                  Workstation
+---------------------------------------------------+     +------------------+
|                                                   |     |                  |
|  +-----------+   apiserver        +------------+  |     |  +------------+  |
|  |           |   proxy            |            |  |     |  |            |  |
|  | apiserver |                    |  ingress   |  |     |  |  ingress   |  |
|  |           |                    | controller |  |     |  | controller |  |
|  |           |                    |            |  |     |  |            |  |
|  |           |                    |            |  |     |  |            |  |
|  |           |  service account/  |            |  |     |  |            |  |
|  |           |  kubeconfig        |            |  |     |  |            |  |
|  |           +<-------------------+            |  |     |  |            |  |
|  |           |                    |            |  |     |  |            |  |
|  +------+----+      kubeconfig    +------+-----+  |     |  +------+-----+  |
|         |<--------------------------------------------------------|        |
|                                                   |     |                  |
+---------------------------------------------------+     +------------------+

Service Account

If using a service account to connect to the API server, Dashboard expects the file /var/run/secrets/kubernetes.io/serviceaccount/token to be present. It provides a secret token that is required to authenticate with the API server.

Verify with the following commands:

# start a container that contains curl
$ kubectl run test --image=tutum/curl -- sleep 10000

# check that container is running
$ kubectl get pods
NAME                   READY     STATUS    RESTARTS   AGE
test-701078429-s5kca   1/1       Running   0          16s

# check if secret exists
$ kubectl exec test-701078429-s5kca ls /var/run/secrets/kubernetes.io/serviceaccount/
ca.crt
namespace
token

# get service IP of master
$ kubectl get services
NAME         CLUSTER-IP   EXTERNAL-IP   PORT(S)   AGE
kubernetes   10.0.0.1     <none>        443/TCP   1d

# check base connectivity from cluster inside
$ kubectl exec test-701078429-s5kca -- curl -k https://10.0.0.1
Unauthorized

# connect using tokens
$ TOKEN_VALUE=$(kubectl exec test-701078429-s5kca -- cat /var/run/secrets/kubernetes.io/serviceaccount/token)
$ echo $TOKEN_VALUE
eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3Mi....9A
$ kubectl exec test-701078429-s5kca -- curl --cacert /var/run/secrets/kubernetes.io/serviceaccount/ca.crt -H  "Authorization: Bearer $TOKEN_VALUE" https://10.0.0.1
{
  "paths": [
    "/api",
    "/api/v1",
    "/apis",
    "/apis/apps",
    "/apis/apps/v1alpha1",
    "/apis/authentication.k8s.io",
    "/apis/authentication.k8s.io/v1beta1",
    "/apis/authorization.k8s.io",
    "/apis/authorization.k8s.io/v1beta1",
    "/apis/autoscaling",
    "/apis/autoscaling/v1",
    "/apis/batch",
    "/apis/batch/v1",
    "/apis/batch/v2alpha1",
    "/apis/certificates.k8s.io",
    "/apis/certificates.k8s.io/v1alpha1",
    "/apis/extensions",
    "/apis/extensions/v1beta1",
    "/apis/policy",
    "/apis/policy/v1alpha1",
    "/apis/rbac.authorization.k8s.io",
    "/apis/rbac.authorization.k8s.io/v1alpha1",
    "/apis/storage.k8s.io",
    "/apis/storage.k8s.io/v1beta1",
    "/healthz",
    "/healthz/ping",
    "/logs",
    "/metrics",
    "/swaggerapi/",
    "/ui/",
    "/version"
  ]
}

If it is not working, there are two possible reasons:

The contents of the tokens are invalid. Find the secret name:

 kubectl get secrets --field-selector=type=kubernetes.io/service-account-token

Delete the secret:

 kubectl delete secret {SECRET_NAME}

It will automatically be recreated.

You have a non-standard Kubernetes installation and the file containing the token may not be present.

The API server will mount a volume containing this file, but only if the API server is configured to use the ServiceAccount admission controller. If you experience this error, verify that your API server is using the ServiceAccount admission controller. If you are configuring the API server by hand, you can set this with the --admission-control parameter. Please note that you should use other admission controllers as well. Before configuring this option, please read about admission controllers.

More information:

Kubeconfig

If you want to use a kubeconfig file for authentication, follow the deploy procedure and add the flag --kubeconfig=/etc/kubernetes/kubeconfig.yaml to the deployment.

Dumping generated Kong configuration

If the controller generates configuration that it cannot apply to Kong successfully, reviewing the generated configuration manually and/or applying it in a test environment can help locate potential causes.

Under normal operation, the controller does not store generated configuration; it is only sent to Kong’s Admin API. The --dump-config flag enables a diagnostic mode where the controller also saves generated configuration and Kong Gateway responses to memory. You can retrieve these via the web interface of the diagnostic server at host:10256/debug/config.

To use the diagnostic mode:

Set the --dump-config flag (or CONTROLLER_DUMP_CONFIG environment variable) to true. Optionally set the --dump-sensitive-config flag to true to include un-redacted TLS certificate keys and credentials.

If you’re deploying with the Helm chart, add the following to your values.yaml file:
```
ingressController:
 env:
   dump_config: "true"
   dump_sensitive_config: "true"
```
(Optional) Make a change to a Kubernetes resource that you know will reproduce the issue. If you are unsure what change caused the issue originally, you can omit this step.

Port forward to the diagnostic server:

kubectl port-forward -n CONTROLLER_NAMESPACE CONTROLLER_POD 10256:10256

Retrieve successfully- and/or unsuccessfully-applied configuration:

curl -svo last_good.json localhost:10256/debug/config/successful
curl -svo last_bad.json localhost:10256/debug/config/failed

Retrieve the last error response body received from Kong Gateway:

curl -svo raw_error_body.json localhost:10256/debug/config/raw-error

Using dumped configuration

Once you have dumped configuration, take one of the following approaches to isolate issues:

If you know of a specific Kubernetes resource change that reproduces the issue, diffing last_good.json and last_bad.json will show the change the controller is trying to apply unsuccessfully.
You can apply dumped configuration via the /config Admin API endpoint (DB-less mode) or using decK (DB-backed mode) to a test instance not managed by the ingress controller. This approach lets you review requests and responses (passing --verbose 2 to decK will show all requests) and add debug Kong Lua code when controller requests result in an unhandled error (500 response).
To run a DB-less Kong Gateway instance with Docker for testing purposes, run curl https://get.konghq.com/quickstart | bash -s -- -D.

Once this image is running, run curl http://localhost:8001/config @last_bad.json to try applying the configuration and see any errors.

You can also analyze the returned error response body from Kong Gateway to understand the issue.

Inspecting network traffic with a tcpdump sidecar

Inspecting network traffic allows you to review traffic between the ingress controller and Kong admin API and/or between the Kong proxy and upstream applications. You can use this in situations where logged information does not provide you sufficient data on the contents of requests and you wish to see exactly what was sent over the network.

Although you cannot install and use tcpdump within the controller or Kong containers, you can add a tcpdump sidecar to your Pod’s containers. The sidecar will be able to sniff traffic from other containers in the Pod. You can edit your Deployment (to add the sidecar to all managed Pods) or a single Pod and add the following under the containers section of the Pod spec:

- name: tcpdump
  securityContext:
    runAsUser: 0
  image: corfr/tcpdump
  command:
    - /bin/sleep
    - infinity

kubectl patch --type=json -n kong deployments.apps ingress-kong -p='[{
  "op":"add",
  "path":"/spec/template/spec/containers/-",
  "value":{
    "name":"tcpdump",
    "securityContext":{
        "runAsUser":0
    },
    "image":"corfr/tcpdump",
    "command":["/bin/sleep","infinity"]
  }
}]'

If you are using the Kong Helm chart, you can alternately add this to the sidecarContainers section of values.yaml.

Once the sidecar is running, you can use kubectl exec -it POD_NAME -c tcpdump and run a capture. For example, to capture traffic between the controller and Kong admin API:

tcpdump -npi any -s0 -w /tmp/capture.pcap host 127.0.0.1 and port 8001

or between Kong and an upstream application with endpoints 10.0.0.50 and 10.0.0.51:

tcpdump -npi any -s0 -w /tmp/capture.pcap host 10.0.0.50 or host 10.0.0.51

Once you’ve replicated the issue, you can stop the capture, exit the container, and use kubectl cp to download the capture from the tcpdump container to a local system for review with Wireshark.

Note that you will typically need to temporarily disable TLS to inspect application-layer traffic. If you have access to the server’s private keys you can decrypt TLS, though this does not work if the session uses an ephemeral cipher (neither the controller nor Kong proxy have support for dumping session secrets).

Gathering profiling data

The controller provides access to the Golang profiler, which provides diagnostic information on memory and CPU consumption within the program.

To enable profiling and access it, set CONTROLLER_PROFILING=true in the controller container environment (ingressController.env.profiling: true using the Helm chart), wait for the Deployment to restart, run kubectl port-forward <POD_NAME> 10256:10256, and visit http://localhost:10256/debug/pprof/.

Identify and fix an invalid configuration

Kubernetes resources can request configuration that Kong Ingress Controller can’t translate into a valid Kong Gateway configuration. While the admission webhook can reject some invalid configurations during creation and the controller can fix some invalid configurations on its own, some configuration issues require you to review and fix them. When such issues arise, Kong Ingress Controller creates Kubernetes Events to help you identify problem resources and understand how to fix them.

To determine if there are any translation failures that you might want to fix, you can monitor the ingress_controller_translation_count Prometheus metric.

Monitor for issues that require manual fixes

Kong Ingress Controller’s Prometheus metrics include ingress_controller_translation_count and ingress_controller_configuration_push_count counters. Issues that require human intervention add success=false tallies to these counters.

Kong Ingress Controller also generates error logs with a could not update kong admin for configuration push failures.

Finding problem resource Events

Kong Ingress Controller provides Kubernetes Events to help understand the state of your system. Events occur when an invalid configuration is rejected by Kong Gateway (KongConfigurationApplyFailed) or when an invalid configuration such as an upstream service that does not exist is detected (KongConfigurationTranslationFailed)..

For more information, see Events.

Debugging Konnect integration

Kong Ingress Controller needs to communicate with the Konnect cloud APIs to provide the integration. If you encounter issues with KIC in Konnect, you should first inspect logs from Kong Ingress Controller to identify the root cause.

By default, KIC logs the details for every failed request and response (method, URL, status code) it receives from Konnect. If you set the LOG_LEVEL to trace, Kong Ingress Controller will log the details for every request and response it receives from Konnect.

Here is an example of a failed request/response log entry:

Request failed  {"x_b3_traceid": "66c731200000000034ce3297e8e64544", "x_b3_spanid": "4e6955874299011d", "x_datadog_trace_id": "3805034363203503428", "x_datadog_parent_id": "5650141246939267357", "v": 0, "method": "GET", "url": "https://us.kic.api.konghq.tech/kic/api/control-planes/81bc4af5-ed3c-40b4-bb88-b5a05fbe34a1/oauth2?size=1000", "status_code": 404}

If your issue requires further investigation on the Konnect side, attach logs with tracing information to your support ticket.