For application pods in the Istio service mesh, all traffic to/from the pods needs to go through the sidecar proxies (istio-proxy containers). This istio-cni CNI plugin will set up the pods' networking to fulfill this requirement in place of the current Istio injected pod initContainers istio-init approach.

This is currently accomplished (for IPv4) via configuring the iptables rules in the netns for the pods.

The CNI handling the netns setup replaces the current Istio approach using a NET_ADMIN privileged initContainers container, istio-init, injected in the pods along with istio-proxy sidecars. This removes the need for a privileged, NET_ADMIN container in the Istio users' application pods.

A complete set of instructions on how to use and install the Istio CNI is available on the Istio documentation site under Install Istio with the Istio CNI plugin. Only a summary is provided here. The steps are:

1. Install Kubernetes and kubelet in a manner that can support the CNI

2. Install Kubernetes with the ServiceAccount admission controller enabled

3. Install the Istio CNI components. A specific example assuming locally built CNI images would be:

$ CNI_HUB=docker.io/my_userid
$ CNI_TAG=mytag
# run from the ${GOPATH}/src/istio.io/cni dir (repo where istio/cni was cloned)
$ helm template --name=istio-cni --namespace=kube-system --set "excludeNamespaces={}" --set hub=${CNI_HUB} --set tag=${CNI_TAG} --set pullPolicy=IfNotPresent --set logLevel=debug  deployments/kubernetes/install/helm/istio-cni > istio-cni_install.yaml
$ kubectl apply -f istio-cni_install.yaml

4. Create and apply Istio manifests with the Istio CNI plugin enabled using the --set istio_cni.enabled=true Helm variable

For most Kubernetes environments the istio-cni helm parameters' defaults will configure the Istio CNI plugin in a manner compatible with the Kubernetes installation. Refer to the Hosted Kubernetes Usage section for Kubernetes environment specific procedures.

Helm chart params

Not all hosted Kubernetes clusters are created with the kubelet configured to use the CNI plugin so compatibility with this istio-cni solution is not ubiquitous. The istio-cni plugin is expected to work with any hosted kubernetes leveraging CNI plugins. The below table indicates the known CNI status of hosted Kubernetes environments and whether istio-cni has been trialed in the cluster type.

1. Enable network-policy in your cluster. NOTE: for existing clusters this redeploys the nodes.

2. Make sure your kubectl user (service-account) has a ClusterRoleBinding to the cluster-admin role. This is also a typical pre-requisite for installing Istio on GKE.

   1. kubectl create clusterrolebinding cni-cluster-admin-binding --clusterrole=cluster-admin [email protected]
      1. User [email protected] is an admin user associated with the gcloud GKE cluster

3. Create the Istio CNI manifests with this Helm chart option --set cniBinDir=/home/kubernetes/bin

No special set up is required for IKS, as it currently uses the default cni-conf-dir and cni-bin-dir.

Add the following section into istio-cni.yaml to run the install-cni DaemonSet container as privileged so that it has proper write permission in the host filesystem:

securityContext:
  privileged: true

2. Grant privileged permission to istio-cni service account:

$ oc adm policy add-scc-to-user privileged -z istio-cni -n kube-system

First, clone this repository under $GOPATH/src/istio.io/.

For linux targets:

$ GOOS=linux make build

You can also build the project from a non-standard location like so:

$ ISTIO_CNI_RELPATH=github.com/some/cni GOOS=linux make build

To push the Docker image:

$ export HUB=docker.io/myuser
$ export TAG=dev
$ GOOS=linux make docker.push

NOTE: Set HUB and TAG per your docker registry.

The Helm package tarfile can be created via

$ helm package $GOPATH/src/istio.io/cni/deployments/kubernetes/install/helm/istio-cni

An example for hosting a test repo for the Helm istio-cni package:

1. Create package tarfile with helm package $GOPATH/src/istio.io/cni/deployments/kubernetes/install/helm/istio-cni
2. Copy tarfile to dir to serve the repo from
3. Run helm serve --repo-path <dir where helm tarfile is> &
   1. The repo URL will be output (http://127.0.0.1:8879)
   2. (optional) Use the --address <IP>:<port> option to bind the server to a specific address/port

To use this repo via helm install:

$ helm repo add local_istio http://127.0.0.1:8879
$ helm repo update

At this point the istio-cni chart is ready for use by helm install.

To make use of the istio-cni chart from another chart:

1. Add the following to the other chart's requirements.yaml:

   - name: istio-cni
     version: ">=0.0.1"
     repository: http://127.0.0.1:8879/
     condition: istio-cni.enabled

2. Run helm dependency update <chart> on the chart that needs to depend on istio-cni.

   1. NOTE: for istio/istio the charts need to be reorganized to make helm dependency update work. The child charts (pilot, galley, etc) need to be made independent charts in the directory at the same level as the main istio chart (istio/istio#9306).

The Istio CNI testing strategy and execution details are explained here.

1. Collect your pod's container id using kubectl.

$ ns=test-istio
$ podnm=reviews-v1-6b7f6db5c5-59jhf
$ container_id=$(kubectl get pod -n ${ns} ${podnm} -o jsonpath="{.status.containerStatuses[?(@.name=='istio-proxy')].containerID}" | sed -n 's/docker:\/\/\(.*\)/\1/p')

2. SSH into the Kubernetes worker node that runs your pod.

3. Use nsenter to view the iptables.

$ cpid=$(docker inspect --format '{{ .State.Pid }}' $container_id)
$ nsenter -t $cpid -n iptables -L -t nat -n -v --line-numbers -x

The CNI plugins are executed by threads in the kubelet process. The CNI plugins logs end up the syslog under the kubelet process. On systems with journalctl the following is an example command line to view the last 1000 kubelet logs via the less utility to allow for vi-style searching:

$ journalctl -t kubelet -n 1000 | less

Each GKE cluster's will have many categories of logs collected by Stackdriver. Logs can be monitored via the project's log viewer and/or the gcloud logging read capability.

The following example grabs the last 10 kubelet logs containing the string "cmdAdd" in the log message.

$ gcloud logging read "resource.type=gce_instance AND jsonPayload.SYSLOG_IDENTIFIER=kubelet AND jsonPayload.MESSAGE:cmdAdd" --limit 10 --format json

• istio-cni.yaml

  • Helm chart manifest for deploying install-cni container as daemonset
  • istio-cni-config configmap with CNI plugin config to add to CNI plugin chained config
  • creates service-account istio-cni with ClusterRoleBinding to allow gets on pods' info

• install-cni container

  • copies istio-cni binary and istio-iptables.sh to /opt/cni/bin
  • creates kubeconfig for the service account the pod is run under
  • injects the CNI plugin config to the config file pointed to by CNI_CONF_NAME env var
    • example: CNI_CONF_NAME: 10-calico.conflist
    • jq is used to insert CNI_NETWORK_CONFIG into the plugins list in /etc/cni/net.d/${CNI_CONF_NAME}

• istio-cni

  • CNI plugin executable copied to /opt/cni/bin
  • currently implemented for k8s only
  • on pod add, determines whether pod should have netns setup to redirect to Istio proxy
    • if so, calls istio-iptables.sh with params to setup pod netns

• istio-iptables.sh

  • fork of Istio's istio-iptables.sh
  • sets up iptables to redirect a list of ports to the port envoy will listen

The framework for this implementation of the CNI plugin is based on the containernetworking sample plugin.

The Istio makefiles and container build logic was leveraged heavily/lifted for this repo.

Specifically:

• golang build logic
• multi-arch target logic
• k8s lib versions (Gopkg.toml)
• docker container build logic
  • setup staging dir for docker build
  • grab built executables from target dir and cp to staging dir for docker build
  • tagging and push logic

The details for the deployment & installation of this plugin were pretty much lifted directly from the Calico CNI plugin.

Specifically:

• CNI installation script
  • This does the following
    • sets up CNI conf in /host/etc/cni/net.d/*
    • copies calico CNI binaries to /host/opt/cni/bin
    • builds kubeconfig for CNI plugin from service-account info mounted in the pod: https://github.com/projectcalico/cni-plugin/blob/master/k8s-install/scripts/install-cni.sh#L142
    • reference: https://kubernetes.io/docs/reference/access-authn-authz/service-accounts-admin/
• The CNI installation script is containerized and deployed as a daemonset in k8s. The relevant calico k8s manifests were used as the model for the istio-cni plugin's manifest:
  • daemonset and configmap
    • search for the calico-node Daemonset and its install-cni container deployment
  • RBAC
    • this creates the service account the CNI plugin is configured to use to access the kube-api-server

The installation script install-cni.sh injects the istio-cni plugin config at the end of the CNI plugin chain config. It creates or modifies the file from the configmap created by the Kubernetes manifest.

Workflow:

1. Check k8s pod namespace against exclusion list (plugin config)

   1. Config must exclude namespace that Istio control-plane is installed in
   2. If excluded, ignore the pod and return prevResult

2. Setup redirect rules for the pods:

   1. Get the port list from from pods definition
   2. Setup iptables with required port list: nsenter --net=<k8s pod netns> /opt/cni/bin/istio-iptables.sh ...

   Following conditions will prevent the redirect rules to be setup in the pods:

   1. Pods only have 1 container(no sidecar proxy injected)
   2. Pods have annotation `sidecar.istio.io/inject` set to `false` or has no key `sidecar.istio.io/status` in annotations
   

3. Return prevResult

TBD istioctl / auto-sidecar-inject logic for handling things like specific include/exclude IPs and any other features.

• Watch configmaps or CRDs and update the istio-cni plugin's config with these options.

Anything needed? The netns is destroyed by kubelet so ideally this is a NOOP.

The plugin leverages logrus & directly utilizes some Calico logging lib util functions.

The proposed Istio pod network controller has the problem of synchronizing the netns setup with the rest of the pod init. This approach requires implementing custom synchronization between the controller and pod initialization.

Kubernetes has already solved this problem by not starting any containers in new pods until the full CNI plugin chain has completed successfully. Also, architecturally, the CNI plugins are the components responsible for network setup for container runtimes.