Skip to main content
Version: latest

trace_dns

The trace_dns gadget is used to trace DNS queries and responses.

Getting started​

Running the gadget:

$ kubectl gadget run ghcr.io/inspektor-gadget/gadget/trace_dns:latest [flags]

Guide​

DNS is a centerpiece of communication in Kubernetes, it allows pods to discover other applications and communicate with the outside world. So, it is important to be able monitor DNS requests and responses to debug DNS issues. In this guide, we will use the trace_dns gadget to trace DNS requests and responses in Kubernetes (single node minikube cluster). The gadget can be used to trace DNS requests at different stages e.g. from application pods to kube-dns service, from kube-dns service to CoreDNS pod, from CoreDNS pod to upstream DNS server, etc. The gadget can also be used to trace DNS requests based on different filters e.g. namespace, pod name, container name, domain name, etc.

info

Check out video, Demystifying DNS: A Guide to Understanding and Debugging Request Flows in Kubernetes Clusters to learn troubleshooting DNS issues with Inspektor Gadgets.

Flow of DNS request in Kubernetes:​

Before we jump to scenarios, it is important to understand DNS in Kubernetes is based on multiple components:

  • Application Pods: These are the pods that make DNS requests.
  • kube-dns Service: The Kubernetes service to which application pods send requests (dnsPolicy=ClusterFirst).
  • CoreDNS: The DNS server that resolves DNS requests in Kubernetes.
  • Upstream DNS: The DNS server that CoreDNS forwards DNS requests to.
note

You might have other components in your DNS setup e.g. nodelocaldns, etc. but for the purpose of this guide, we will focus on the above components.

The overall flow of DNS request is as follows:

Tracing an Application Pod:​

Let's try to understand the flow at different stages. We will start by tracing DNS requests for specific pod. First, create a namespace:

kubectl create namespace demo

with expected output:

namespace/demo created

Run the gadget in a terminal, in this case we are only focused on DNS requests/response in demo namespace for pod mypod:

kubectl gadget run trace_dns:latest -n demo -p mypod

Run a pod in a different terminal:

kubectl -n demo run mypod -it --image=wbitt/network-multitool -- /bin/sh

Inside the pod, perform a DNS request:

nslookup -query=a inspektor-gadget.io.

The request/response will be logged by the DNS gadget:

K8S.NODE             K8S.NAMESPACE        K8S.PODNAME          K8S.CONTAINERNAME    SRC                         DST                           COMM              PID QR QTYPE      NAME                  RCODE    ADDRESSES
minikube-docker      demo                 mypod                mypod                p/demo/mypod:56093          s/kube-system/kube-dns:53     isc-net-0…     422204 Q  A          inspektor-gadget.io.
minikube-docker      demo                 mypod                mypod                s/kube-system/kube-dns:53   p/demo/mypod:56093            isc-net-0…     422204 R  A          inspektor-gadget.io.  Success  172.67.16…

The first line shows the DNS request from the mypod pod to kube-dns service and the second line shows the response from kube-dns service back to the mypod pod as confirmed by SRC and DST fields.

Tracing Application Pod and CoreDNS Pod:​

Let's try to expand the scope and see what happens to the DNS request at CoreDNS pod. Restart the gadget as:

kubectl gadget run trace_dns:latest -n demo,kube-system -F "k8s.podName~mypod|coredns-.*" -F "name==inspektor-gadget.io." --fields=k8s.node,k8s.namespace,k8s.podname,id,src,dst,qr,name,rcode,timestamp

This will filter the DNS request/response for pods mypod and CoreDNS pods. See how are able to choose different fields via --fields.

On performing the DNS request again, you will see the following output from the gadget:

K8S.NODE                K8S.NAMESPACE           K8S.PODNAME             ID            SRC                                        DST                                      QR NAME                   RCODE    TIMESTAMP
minikube-docker         demo                    mypod                   c3b5          p/demo/mypod:50037                         s/kube-system/kube-dns:53                Q  inspektor-gadget.io.            2024-08-30T15:15:21.13712…
minikube-docker         kube-system             coredns-7db…8ff4d-r7hwl c3b5          p/demo/mypod:50037                         p/kube-system/coredns-7db6d8ff4d-r7hwl:… Q  inspektor-gadget.io.            2024-08-30T15:15:21.13720…
minikube-docker         kube-system             coredns-7db…8ff4d-r7hwl 0e01          p/kube-system/coredns-7db6d8ff4d-r7hwl:…   192.168.49.1:53                          Q  inspektor-gadget.io.            2024-08-30T15:15:21.13748…
minikube-docker         kube-system             coredns-7db…8ff4d-r7hwl 0e01          192.168.49.1:53                            p/kube-system/coredns-7db6d8ff4d-r7hwl:… R  inspektor-gadget.io.   Success  2024-08-30T15:15:21.16513…
minikube-docker         kube-system             coredns-7db…8ff4d-r7hwl c3b5          p/kube-system/coredns-7db6d8ff4d-r7hwl:…   p/demo/mypod:50037                       R  inspektor-gadget.io.   Success  2024-08-30T15:15:21.16524…
minikube-docker         demo                    mypod                   c3b5          s/kube-system/kube-dns:53                  p/demo/mypod:50037                       R  inspektor-gadget.io.   Success  2024-08-30T15:15:21.16526…

In this trace, we can see the request/response involving both the CoreDNS pod and mypod. The first two lines show the request reaching CoreDNS pod. Since we are trying to resolve an external domain name inspektor-gadget.io., CoreDNS forwards the request to the upstream DNS server (confirmed by new ID=0e01) in our case the upstream server is 192.168.49.1:53. The remaining lines show the response (ID=0e01) of the upstream DNS server to CoreDNS pod and the response (ID=c3b5) of CoreDNS pod to the pod.

note

You might not see the upstream DNS request / response if you are using CoreDNS cache plugin since CoreDNS will cache the upstream response. Also, the output isn't sorted based on timestamp but you can use it to understand the sequence of events.

This showcases how trace_dns gadget can be used to trace DNS request at different stages. This is helpful to understand if a request/response is being dropped/altered at a specific stage. Again the output can vary based on the DNS/CNI setup in your cluster but the idea here is to provide a starting point to debug DNS issues.

Finally, clean the system:

$ kubectl delete namespace demo

Limitations​

DNS over TCP support is limited since it doesn't support TCP stream reassembly. This means that the gadget will not be able to trace DNS requests that are sent over multiple TCP packets.