Installing the Sidecar
Injection
In order to take advantage of all of Istio’s features, pods in the mesh must be running an Istio sidecar proxy.
The following sections describe two
ways of injecting the Istio sidecar into a pod: manually using the istioctl
command or by enabling automatic Istio sidecar injection in the pod’s namespace.
Manual injection directly modifies configuration, like deployments, and injects the proxy configuration into it.
When enabled in a pod’s namespace, automatic injection injects the proxy configuration at pod creation time using an admission controller.
Injection occurs by applying a template defined in the istio-sidecar-injector ConfigMap.
Manual sidecar injection
To manually inject a deployment, use istioctl kube-inject:
$ istioctl kube-inject -f @samples/sleep/sleep.yaml@ | kubectl apply -f -
serviceaccount/sleep created
service/sleep created
deployment.apps/sleep created
By default, this will use the in-cluster configuration. Alternatively, injection can be done using local copies of the configuration.
$ kubectl -n istio-system get configmap istio-sidecar-injector -o=jsonpath='{.data.config}' > inject-config.yaml
$ kubectl -n istio-system get configmap istio-sidecar-injector -o=jsonpath='{.data.values}' > inject-values.yaml
$ kubectl -n istio-system get configmap istio -o=jsonpath='{.data.mesh}' > mesh-config.yaml
Run kube-inject over the input file and deploy.
$ istioctl kube-inject \
--injectConfigFile inject-config.yaml \
--meshConfigFile mesh-config.yaml \
--valuesFile inject-values.yaml \
--filename @samples/sleep/sleep.yaml@ \
| kubectl apply -f -
serviceaccount/sleep created
service/sleep created
deployment.apps/sleep created
Verify that the sidecar has been injected into the sleep pod with 2/2 under the READY column.
$ kubectl get pod -l app=sleep
NAME READY STATUS RESTARTS AGE
sleep-64c6f57bc8-f5n4x 2/2 Running 0 24s
Automatic sidecar injection
Sidecars can be automatically added to applicable Kubernetes pods using a mutating webhook admission controller provided by Istio.
When you set the istio-injection=enabled label on a namespace and the injection webhook is enabled, any new pods that are created in that namespace will automatically have a sidecar added to them.
Note that unlike manual injection, automatic injection occurs at the pod-level. You won’t see any change to the deployment itself. Instead you’ll want to check individual pods (via kubectl describe) to see the injected proxy.
Deploying an app
Deploy sleep app. Verify both deployment and pod have a single container.
$ kubectl apply -f @samples/sleep/sleep.yaml@
$ kubectl get deployment -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
sleep 1/1 1 1 12s sleep curlimages/curl app=sleep
$ kubectl get pod
NAME READY STATUS RESTARTS AGE
sleep-8f795f47d-hdcgs 1/1 Running 0 42s
Label the default namespace with istio-injection=enabled
$ kubectl label namespace default istio-injection=enabled --overwrite
$ kubectl get namespace -L istio-injection
NAME STATUS AGE ISTIO-INJECTION
default Active 5m9s enabled
istio-system Active 4m58s disabled
kube-node-lease Active 5m10s
kube-public Active 5m10s
kube-system Active 5m10s
local-path-storage Active 5m7s
Injection occurs at pod creation time. Kill the running pod and verify a new pod is created with the injected sidecar. The original pod has 1/1 READY containers and the pod with injected sidecar has 2/2 READY containers.
$ kubectl delete pod -l app=sleep
$ kubectl get pod -l app=sleep
pod "sleep-776b7bcdcd-7hpnk" deleted
NAME READY STATUS RESTARTS AGE
sleep-776b7bcdcd-7hpnk 1/1 Terminating 0 1m
sleep-776b7bcdcd-bhn9m 2/2 Running 0 7s
View detailed state of the injected pod. You should see the injected istio-proxy container and corresponding volumes. Be sure to substitute the correct name for the Running pod below.
$ kubectl describe pod -l app=sleep
...
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
...
Normal Created 11s kubelet Created container istio-init
Normal Started 11s kubelet Started container istio-init
...
Normal Created 10s kubelet Created container sleep
Normal Started 10s kubelet Started container sleep
...
Normal Created 9s kubelet Created container istio-proxy
Normal Started 8s kubelet Started container istio-proxy
Disable injection for the default namespace and verify new pods are created without the sidecar.
$ kubectl label namespace default istio-injection-
$ kubectl delete pod -l app=sleep
$ kubectl get pod
namespace/default labeled
pod "sleep-776b7bcdcd-bhn9m" deleted
NAME READY STATUS RESTARTS AGE
sleep-776b7bcdcd-bhn9m 2/2 Terminating 0 2m
sleep-776b7bcdcd-gmvnr 1/1 Running 0 2s
Understanding what happened
When Kubernetes invokes the webhook, the admissionregistration.k8s.io/v1beta1#MutatingWebhookConfiguration
configuration is applied. The default configuration injects the sidecar into
pods in any namespace with the istio-injection=enabled label. The
istio-sidecar-injector configuration map specifies the configuration for the
injected sidecar. To change how namespaces are selected for injection, you can
edit the MutatingWebhookConfiguration with the following command:
$ kubectl edit mutatingwebhookconfiguration istio-sidecar-injector
For example, you can modify the MutatingWebhookConfiguration to always inject
the sidecar into every namespace, unless a label is set. Editing this
configuration is an advanced operation. Refer to the Kubernetes documentation
for the MutatingWebhookConfiguration API
for more information.
policy
disabled - The sidecar injector will not inject the sidecar into
pods by default. Add the sidecar.istio.io/inject annotation with
value true to the pod template spec to override the default and enable injection.
enabled - The sidecar injector will inject the sidecar into pods by
default. Add the sidecar.istio.io/inject annotation with
value false to the pod template spec to override the default and disable injection.
The following example uses the sidecar.istio.io/inject annotation to disable sidecar injection.
$ kubectl apply -f - <<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
name: ignored
labels:
app: ignored
spec:
selector:
matchLabels:
app: ignored
template:
metadata:
labels:
app: ignored
annotations:
sidecar.istio.io/inject: "false"
spec:
containers:
- name: ignored
image: governmentpaas/curl-ssl
command: ["/bin/sleep","infinity"]
EOF
It can then be verified that no sidecar was injected.
$ kubectl get pods -l app=ignored
NAME READY STATUS RESTARTS AGE
ignored-b469d6fcf-8dngz 1/1 Running 0 57s
template
The sidecar injection template uses https://golang.org/pkg/text/template which, when parsed and executed, is decoded to the following struct containing the list of containers and volumes to inject into the pod.
type SidecarInjectionSpec struct {
RewriteAppHTTPProbe bool `yaml:"rewriteAppHTTPProbe"`
InitContainers []corev1.Container `yaml:"initContainers"`
Containers []corev1.Container `yaml:"containers"`
Volumes []corev1.Volume `yaml:"volumes"`
DNSConfig *corev1.PodDNSConfig `yaml:"dnsConfig"`
ImagePullSecrets []corev1.LocalObjectReference `yaml:"imagePullSecrets"`
}
The template is applied to the following data structure at runtime.
type SidecarTemplateData struct {
DeploymentMeta *metav1.ObjectMeta
ObjectMeta *metav1.ObjectMeta
Spec *corev1.PodSpec
ProxyConfig *meshconfig.ProxyConfig // Defined by https://istio.io/docs/reference/config/service-mesh.html#proxyconfig
MeshConfig *meshconfig.MeshConfig // Defined by https://istio.io/docs/reference/config/service-mesh.html#meshconfig
}
ObjectMeta and Spec are from the pod. ProxyConfig and MeshConfig
are from the istio ConfigMap in the istio-system namespace. Templates can conditionally
define injected containers and volumes with this data.
For example, the following template
containers:
- name: istio-proxy
image: istio.io/proxy:0.5.0
args:
- proxy
- sidecar
- --configPath
- {{ .ProxyConfig.ConfigPath }}
- --binaryPath
- {{ .ProxyConfig.BinaryPath }}
- --serviceCluster
{{ if ne "" (index .ObjectMeta.Labels "app") -}}
- {{ index .ObjectMeta.Labels "app" }}
{{ else -}}
- "istio-proxy"
{{ end -}}
expands to
containers:
- name: istio-proxy
image: istio.io/proxy:0.5.0
args:
- proxy
- sidecar
- --configPath
- /etc/istio/proxy
- --binaryPath
- /usr/local/bin/envoy
- --serviceCluster
- sleep
when applied over a pod defined by the pod template spec in samples/sleep/sleep.yaml
More control: adding exceptions
There are cases where users do not have control of the pod creation, for instance, when they are created by someone else. Therefore they are unable to add the annotation sidecar.istio.io/inject in the pod, to explicitly instruct Istio whether to install the sidecar or not.
Think of auxiliary pods that might be created as an intermediate step while deploying an application. OpenShift Source to Image Builds, for example, creates such pods for building the source code of an application. Once the binary artifact is built, the application pod is ready to run and the auxiliary pods are discarded. Those intermediate pods should not get an Istio sidecar, even if the policy is set to enabled and the namespace is properly labeled to get automatic injection.
For such cases you can instruct Istio to not inject the sidecar on those pods, based on labels that are present in those pods. You can do this by editing the istio-sidecar-injector ConfigMap and adding the entry neverInjectSelector. It is an array of Kubernetes label selectors. They are OR'd, stopping at the first match. See an example:
apiVersion: v1
kind: ConfigMap
metadata:
name: istio-sidecar-injector
data:
config: |-
policy: enabled
neverInjectSelector:
- matchExpressions:
- {key: openshift.io/build.name, operator: Exists}
- matchExpressions:
- {key: openshift.io/deployer-pod-for.name, operator: Exists}
template: |-
initContainers:
...
The above statement means: Never inject on pods that have the label openshift.io/build.name or openshift.io/deployer-pod-for.name – the values of the labels don’t matter, we are just checking if the keys exist. With this rule added, the OpenShift Builds use case illustrated above is covered, meaning auxiliary pods will not have sidecars injected (because source-to-image auxiliary pods do contain those labels).
For completeness, you can also use a field called alwaysInjectSelector, with similar syntax, which will always inject the sidecar on pods that match that label selector, regardless of the global policy.
The label selector approach gives a lot of flexibility on how to express those exceptions. Take a look at these docs to see what you can do with them!
Uninstalling the automatic sidecar injector
$ kubectl delete mutatingwebhookconfiguration istio-sidecar-injector
mutatingwebhookconfiguration.admissionregistration.k8s.io "istio-sidecar-injector" deleted
The above command will not remove the injected sidecars from Pods. A rolling update or simply deleting the pods and forcing the deployment to create them is required.
Optionally, it may also be desirable to clean-up other resources that were modified in this task.
$ kubectl label namespace default istio-injection-
namespace/default labeled