Skip to content

Instantly share code, notes, and snippets.

@borkmann
Forked from joestringer/micro-k8s-setup.md
Last active August 18, 2024 07:44
Show Gist options
  • Save borkmann/84812fcac624fb659f66f0d514faa1c8 to your computer and use it in GitHub Desktop.
Save borkmann/84812fcac624fb659f66f0d514faa1c8 to your computer and use it in GitHub Desktop.
MicroK8s setup for Cilium

Set up microk8s with Cilium for development

Microk8s is a Canonical project to provide a kubernetes environment for local development, similar to minikube but without requiring a separate VM to manage. These instructions describe setting it up for common development use cases with Cilium and may be helpful in particular for testing BPF kernel extensions with Cilium.

Microk8s will run its own version of docker for the kubernetes runtime, so if you have an existing docker installation then this may be confusing, for instance when building images the image may be stored with one of these installations and not the other. This guide assumes you will run both docker daemon instances, and use your existing docker-ce for building Cilium while using the microk8s.docker daemon instance for the runtime of your kubernetes pods.

Requirements

In this howto setup was run on packet.net c1.small.x86 node with Ubuntu 17.10. When you use a remote machine from a cloud provider for testing / development, be aware of k8s default settings.

Quick howto:

# apt-get install snapd apt-transport-https ca-certificates curl software-properties-common build-essential flex bison clang llvm libelf-dev libssl-dev libcap-dev gcc-multilib libncurses5-dev pkg-config
# curl -fsSL https://download.docker.com/linux/ubuntu/gpg | apt-key add -
# add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
# apt-get update
# apt-get install docker-ce

Install and set up golang and Cilium

# wget https://dl.google.com/go/go1.11.2.linux-amd64.tar.gz
# tar xvf go1.11.2.linux-amd64.tar.gz -C /usr/local/
# mkdir -p ~/go/src/github.com/cilium/

And add to bashrc:

export GOPATH=/<home>/go/
export GOROOT=/usr/local/go/
export PATH=/snap/bin/:/usr/local/go/bin/:/root/go/bin/:$PATH

Then follow with building Cilium itself:

# cd ~/go/src/github.com/cilium/
# git clone https://github.com/cilium/cilium.git && cd cilium/
# go get -u github.com/gordonklaus/ineffassign
# go get -u github.com/jteeuwen/go-bindata/...
# SKIP_DOCS=true make

Install and set up μK8s and Cilium

For stable version:

# snap install microk8s --classic --channel=1.12/stable                                       

For latest unstable version (e.g. having microk8s.stop / microk8s.start and other extensions):

# snap install microk8s --classic --edge

Edit configs as next step:

# vim /var/snap/microk8s/current/args/kube-apiserver                            
  • Add --allow-privileged option (To support Cilium with root privileges)
# vim /var/snap/microk8s/current/args/kubelet                                   
  • Set --network-plugin option to cni (To use Cilium for network plumbing)
  • Set --cni-bin-dir option to /opt/cni/bin (To find the Cilium CNI binary)

Set up aliases to use μK8s for bashrc

# alias kubectl=microk8s.kubectl

... or do ...

# snap alias microk8s.kubectl kubectl
  • There's probably a way to set up your local kubectl to point to the local microk8s instead, rather than using the microk8s version but this is less of a problem because microk8s tends to track latest kubernetes pretty closely. I found that aliasing kubectl to microk8s.kubectl worked well for me locally.

Set up Cilium

# kubectl create -n kube-system -f https://raw.githubusercontent.com/cilium/cilium/1.3.0/examples/kubernetes/addons/etcd/standalone-etcd.yaml
# kubectl create -f https://raw.githubusercontent.com/cilium/cilium/1.3.0/examples/kubernetes/1.12/cilium.yaml
# kubectl -n kube-system edit ds cilium                                         

If you have trouble with the above steps, check the Troubleshooting section.

  • Set docker socket hostpath to point within snap path (So that Cilium can associate container labels with endpoints)

Example configuration:

      volumes:                                                                  
      - hostPath:                                                               
          path: /var/run/cilium                                                 
          type: DirectoryOrCreate                                               
        name: cilium-run                                                        
      - hostPath:                                                               
          path: /sys/fs/bpf                                                     
          type: DirectoryOrCreate                                               
        name: bpf-maps                                                          
      - hostPath:                                                               
          path: /var/snap/microk8s/current/docker.sock       <----                          
          type: Socket                                                          
        name: docker-socket                                                     

Set up other μK8s services

# microk8s.enable dns registry
  • The registry is available on localhost:32000 (via NodePort).

Pushing new custom versions of Cilium

For these steps, you are recommended to have a recent version of docker-ce so that multi-stage builds work correctly. As of Nov 2018, the version of docker provided with microk8s is not new enough (17.03, Issue)

These steps work by:

  1. Using the docker-ce version of docker client to build and update the docker-ce daemon with the new Cilium image
  2. Using docker-ce client docker binary to push the image into the k8s-deployed docker registry
  3. Using the microk8s.docker client to pull the new image from the local registry into the docker daemon provided by microk8s
  4. Relying on the registry URI for the docker image for k8s to pull the image from the microk8s.docker daemon.

Building using docker-ce

  • Make your local changes to your Cilium repository.
# DOCKER_IMAGE_TAG="my-image" make docker-image
# docker tag cilium/cilium:my-image localhost:32000/cilium/cilium:my-image
# docker push localhost:32000/cilium/cilium:my-image

This uses your local docker-ce (and docker daemon hosted at /var/run/docker.sock) to push into the registry that was configured above.

For pushing to a docker hub account:

# docker login
[...]
# DOCKER_IMAGE_TAG="my-image" make docker-image
# docker tag cilium/cilium:my-image user/cilium:my-image
# docker push user/cilium:my-image

If you have trouble with the above steps, check the Troubleshooting section.

Pre-pulling new images into microk8s.docker

The below instructions use the microk8s.docker via microk8s, which is hosted at /var/snap/microk8s/current/docker.sock.

To roll out the new Cilium with the local registry reliably, I found that it was helpful to deploy this prepull YAML; otherwise the connection for fetching the image tends to get reset during container startup, which puts the node into a bad state:

apiVersion: apps/v1beta2
kind: DaemonSet
metadata:
  name: prepull
  namespace: container-registry
spec:
  selector:
    matchLabels:
      name: prepull
  template:
    metadata:
      labels:
        name: prepull
    spec:
      initContainers:
      - name: prepull 
        image: docker
        command: ["docker", "pull", "localhost:32000/cilium/cilium:my-image"]
        volumeMounts:
        - name: docker
          mountPath: /var/run
      volumes:
      - name: docker
        hostPath:
          path: /var/snap/microk8s/current/
      containers:
      - name: pause
        image: gcr.io/google_containers/pause

# kubectl create -f prepull.yaml

When you want to re-pull the image again:

# kubectl delete po -n container-registry -l name=prepull

Update Cilium

Then, edit your Cilium DS YAML to point to the new tag, replacing image: docker.io/cilium/cilium:v1.3.0 with image: localhost:32000/cilium/cilium:my-image:

# kubectl -n kube-system edit ds cilium

Set image and imagePullPolicy:

        image: localhost:32000/cilium/cilium:my-image
        imagePullPolicy: Never
        lifecycle:
          postStart:
            exec:
              command:
              - /cni-install.sh
          preStop:
            exec:
              command:
              - /cni-uninstall.sh

If all of local registry is not used, and images pulled from docker hub instead, set image and imagePullPolicy:

        image: docker.io/user/cilium:my-image
        imagePullPolicy: Always
        lifecycle:
          postStart:
            exec:
              command:
              - /cni-install.sh
          preStop:
            exec:
              command:
              - /cni-uninstall.sh

And rollout:

# kubectl -n kube-system rollout status ds cilium

If the tag is already pointing to your custom image, you should just need to delete the pods:

# kubectl -n kube-system delete po -l k8s-app=cilium
# kubectl -n kube-system rollout status ds cilium

In order to enable and test the ipvlan datapath, command line arguments for daemon need to be:

[...]
    spec:
      containers:
      - args:
        - --debug=$(CILIUM_DEBUG)
        - --kvstore=etcd
        - --kvstore-opt=etcd.config=/var/lib/etcd-config/etcd.config
        - --disable-ipv4=$(DISABLE_IPV4)
        - --datapath-mode=ipvlan                        <--
        - --ipvlan-master-device=bond0                  <--
        - --tunnel=disabled                             <--
        - --install-iptables-rules=false                <-- implicit L3 vs L3S depending on install switch
        command:
        - cilium-agent
[...]

Rolling back

If the rollout gets stuck it can be debugged through ...

# kubectl get pods --all-namespaces -o wide
NAMESPACE            NAME                                    READY   STATUS         RESTARTS   AGE    IP              NODE   NOMINATED NODE
[...]
kube-system          cilium-hbcs8                            0/1     ErrImagePull   0          75s    147.75.80.23    test   <none>
[...]
# kubectl describe pod -n kube-system cilium-hbcs8
[...]
  Warning  Failed     30s                    kubelet, test      Failed to pull image "localhost:32000/cilium/cilium:my-image": rpc error: code = Unknown desc = Error while pulling image: Get http://localhost:32000/v1/repositories/cilium/cilium/images: read tcp localhost:53302->127.0.0.1:32000: read: connection reset by peer
  Normal   BackOff    4s (x4 over 103s)      kubelet, test      Back-off pulling image "localhost:32000/cilium/cilium:my-image"
  Warning  Failed     4s (x4 over 103s)      kubelet, test      Error: ImagePullBackOff
[...]

... e.g. in this case the imagePullPolicy was probably set to Always.

The daemon set updates are undone via:

# kubectl -n kube-system rollout undo ds cilium

Test workload

Check if Cilium is up and running:

# kubectl get pods --all-namespaces -o wide
# kubectl -n kube-system logs --timestamps cilium-1234

Deploying a sample application for testing Cilium w/o policy first:

# kubectl create -f https://raw.githubusercontent.com/cilium/cilium/HEAD/examples/minikube/http-sw-app.yaml
# kubectl exec -it -n kube-system cilium-1234 -- cilium endpoint list
# kubectl exec -it tiefighter -- netperf -t TCP_STREAM -H 10.23.177.124
[...]

Force endpoint regeneration:

# kubectl delete po tiefighter

Debugging

Switch to pods netns:

# microk8s.docker inspect --format '{{ .State.Pid }}' `kubectl get pod tiefighter -o jsonpath='{.status.containerStatuses[0].containerID}' | cut -c 10-21`
8667
# nsenter -t 8667 -n bash

Dumping identity and ip cache:

# kubectl exec -it -n kube-system cilium-1234 -- cilium identity list
# kubectl exec -it -n kube-system cilium-1234 -- cilium map get cilium_ipcache

Policy

Deploy Cilium policy via k8s API:

# kubectl create -f https://raw.githubusercontent.com/cilium/cilium/1.3.0/examples/minikube/sw_l3_l4_policy.yaml

Check policy enforcement for endpoints and currently deployed rules:

# kubectl -n kube-system exec cilium-1234 -- cilium endpoint list
# kubectl -n kube-system exec cilium-l234 -- cilium policy get

Get currently installed policies:

# kubectl get ciliumnetworkpolicies.cilium.io

Delete a Cilium policy via k8s API (also clears it from active policy):

# kubectl delete ciliumnetworkpolicies.cilium.io rule1

Check L3 label-based policy verdicts:

# kubectl -n kube-system exec cilium-1234 -- cilium policy trace --src-k8s-pod default:tiefighter --dst-k8s-pod default:xwing

Minimal example:

# cat sw_l3.yaml 
apiVersion: "cilium.io/v2"
kind: CiliumNetworkPolicy
metadata:
  name: "l3-rule"
specs:
  - endpointSelector:
      matchLabels:
        org: empire
    ingress:
    - fromRequires:
      - matchLabels:
          org: empire
# kubectl create -f ./sw_l3.yaml
# kubectl exec -ti xwing -- ping -c3 <IP tiefighter>
PING 10.9.245.239 (10.9.245.239): 56 data bytes

--- 10.9.245.239 ping statistics ---
3 packets transmitted, 0 packets received, 100% packet loss
command terminated with exit code 1

Cilium monitor output for observing drops by policy:

# kubectl exec -n=kube-system cilium-1234 -ti -- cilium monitor --related-to 49565 --related-to 16476
Listening for events on 8 CPUs with 64x4096 of shared memory
Press Ctrl-C to quit
<- endpoint 16476 flow 0xb5da957f identity 14294->0 state new ifindex 0: 10.9.241.61 -> 10.9.71.76 EchoRequest
xx drop (Policy denied (L3)) flow 0xb5da957f to endpoint 49565, identity 14294->53755: 10.9.241.61 -> 10.9.71.76 EchoRequest
<- endpoint 16476 flow 0xb5da957f identity 14294->0 state new ifindex 0: 10.9.241.61 -> 10.9.71.76 EchoRequest
xx drop (Policy denied (L3)) flow 0xb5da957f to endpoint 49565, identity 14294->53755: 10.9.241.61 -> 10.9.71.76 EchoRequest
<- endpoint 16476 flow 0xb5da957f identity 14294->0 state new ifindex 0: 10.9.241.61 -> 10.9.71.76 EchoRequest
xx drop (Policy denied (L3)) flow 0xb5da957f to endpoint 49565, identity 14294->53755: 10.9.241.61 -> 10.9.71.76 EchoRequest
<- endpoint 16476 flow 0x16f8a5c8 identity 14294->0 state new ifindex 0: ca:2a:e8:1c:47:1c -> 02:7f:f3:65:1c:20 ARP
<- endpoint 49565 flow 0x59d8660b identity 53755->0 state new ifindex 0: 10.9.71.76 -> 10.9.245.239 EchoRequest
-> overlay flow 0x59d8660b identity 53755->0 state new ifindex cilium_vxlan: 10.9.71.76 -> 10.9.245.239 EchoRequest
<- endpoint 49565 flow 0x59d8660b identity 53755->0 state new ifindex 0: 10.9.71.76 -> 10.9.245.239 EchoRequest
-> overlay flow 0x59d8660b identity 53755->0 state new ifindex cilium_vxlan: 10.9.71.76 -> 10.9.245.239 EchoRequest
<- endpoint 49565 flow 0x59d8660b identity 53755->0 state new ifindex 0: 10.9.71.76 -> 10.9.245.239 EchoRequest
-> overlay flow 0x59d8660b identity 53755->0 state new ifindex cilium_vxlan: 10.9.71.76 -> 10.9.245.239 EchoRequest
^C
Received an interrupt, disconnecting from monitor...

Further info

Troubleshooting

Docker daemon not starting after kernel upgrade

In some cases when switching from an older to newer kernel for testing, docker daemon may not start up due to Error starting daemon: error initializing graphdriver: driver not supported which is visible from journalctl -fu snap.microk8s.daemon-docker.service log..

# rm -rf /var/snap/microk8s/common/var/lib/docker/aufs/
# rm -rf /var/snap/microk8s/common/var/lib/docker/image/aufs/

Getting trace_printk working

For using printk(...) debugging in the code, enable via:

# kubectl exec -ti -n kube-system cilium-1234 -- mount -t tracefs nodev /sys/kernel/tracing
# kubectl exec -ti -n kube-system cilium-1234 -- cilium config Debug=Enable
# kubectl exec -ti -n kube-system cilium-1234 -- tc exec bpf dbg

Restart μK8s

In case there is a change in host IP, you can restart kubernetes API server the following way in order to propagte the new IP to all kubernetes cluster members:

# microk8s.stop
# microk8s.start

For older microk8s versions:

# snap disable microk8s
# snap enable microk8s

Problems pushing to docker registry

The docker daemon is not listening on IPv4, so connecting to localhost:32000 for the docker registry may hit something like this depending on your local configuration:

# docker push localhost:32000/cilium/cilium:my-image
The push refers to repository [localhost:32000/cilium/cilium]
Get http://localhost:32000/v2/: net/http: request canceled (Client.Timeout exceeded while awaiting headers)

To overcome this, use 127.0.0.1 instead of localhost, or remove the IPv6 (::1) host alias from /etc/hosts.

Problems deploying Cilium daemon set

When deploying cilium daemon set for the first time, the following error may occur. Make sure to have --allow-privileged option set in kube-apiserver and apiserver reloaded:

# kubectl create -n kube-system -f https://raw.githubusercontent.com/cilium/cilium/1.3.0/examples/kubernetes/addons/etcd/standalone-etcd.yaml
[...]
The DaemonSet "cilium" is invalid: 
* spec.template.spec.containers[0].securityContext.privileged: Forbidden: disallowed by cluster policy
* spec.template.spec.initContainers[0].securityContext.privileged: Forbidden: disallowed by cluster policy

Microk8s journal logs

See also https://microk8s.io/docs/

# journalctl -fu snap.microk8s.daemon-kubelet.service
# journalctl -fu snap.microk8s.daemon-apiserver
# journalctl -fu snap.microk8s.daemon-docker.service

Misc

For testing, we block all traffic other than ssh:

# iptables -A INPUT -i bond0 -m state --state ESTABLISHED,RELATED -j ACCEPT
# iptables -A INPUT -i bond0 -p tcp --dport 22 -j ACCEPT
# iptables -A INPUT -i bond0 -j DROP

This can be made persistent and stored in /etc/iptables/rules.v4:

# apt install iptables-persistent

Some quick-links for general troubleshooting:

@SuperThunder
Copy link

SuperThunder commented Jun 9, 2024

Although this gist is 4 years old it still shows up high on google results.
I was able to get the cilium addon running on microk8s by doing this:

sudo snap install microk8s --classic --channel=latest/edge

sudo microk8s enable community

microk8s enable cilium

@ybizeul
Copy link

ybizeul commented Aug 18, 2024

Also note that, from my understanding, Cilium requires taking over kube-proxy, and microk8s which uses kubelite and doesn't have an option to disable kube-proxy.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment