Circuit Breaker

The outlierDetection settings in the DestinationRule custom resource help to determine whether some endpoints do not behave as expected. Refer to the Envoy documentation for more details on the Outlier Detection algorithm.

Example:

apiVersion: networking.istio.io/v1beta1
kind: DestinationRule
metadata:
  name: reviews-cb-policy
spec:
  host: reviews.prod.svc.cluster.local
  trafficPolicy:
    connectionPool:
      tcp:
        maxConnections: 100 # The maximum number of connections to the host (cumulative for all endpoints)
      http:
        maxRequestsPerConnection: 10 # The connection will be re-established after every 10 requests
    outlierDetection:
      consecutive5xxErrors: 7 # Seven consecutive errors are allowed (including 5XX, TCP and HTTP timeouts)
      interval: 5m            # over 5 minutes.
      baseEjectionTime: 15m   # Upon reaching the error limit, the endpoint will be excluded from balancing for 15 minutes.

Additionally, the VirtualService resource is used to configure the HTTP timeouts. These timeouts are also taken into account when calculating error statistics for endpoints.

Example:

apiVersion: networking.istio.io/v1beta1
kind: VirtualService
metadata:
  name: my-productpage-rule
  namespace: myns
spec:
  hosts:
  - productpage
  http:
  - timeout: 5s
    route:
    - destination:
        host: productpage

gRPC balancing

Caution! Assign a name with the grpc prefix or value to the port in the corresponding Service to make gRPC service balancing start automatically.

Locality Failover

Read the main documentation if you need.

Istio allows you to configure a priority-based locality (geographic location) failover between endpoints. Istio uses node labels with the appropriate hierarchy to define the zone:

  • topology.istio.io/subzone
  • topology.kubernetes.io/zone
  • topology.kubernetes.io/region

This comes in handy for inter-cluster failover when used together with a multicluster.

Caution! The Locality Failover can be enabled using the DestinationRule CR. Note that you also have to configure the outlierDetection.

Example:

apiVersion: networking.istio.io/v1beta1
kind: DestinationRule
metadata:
  name: helloworld
spec:
  host: helloworld
  trafficPolicy:
    loadBalancer:
      localityLbSetting:
        enabled: true # LF is enabled
    outlierDetection: # outlierDetection must be enabled
      consecutive5xxErrors: 1
      interval: 1s
      baseEjectionTime: 1m

Retry

You can use the VirtualService resource to configure Retry for requests.

Caution! All requests (including POST ones) are retried three times by default.

Example:

apiVersion: networking.istio.io/v1beta1
kind: VirtualService
metadata:
  name: ratings-route
spec:
  hosts:
  - ratings.prod.svc.cluster.local
  http:
  - route:
    - destination:
        host: ratings.prod.svc.cluster.local
    retries:
      attempts: 3
      perTryTimeout: 2s
      retryOn: gateway-error,connect-failure,refused-stream

Canary

Caution! Istio is only responsible for flexible request routing that relies on special request headers (such as cookies) or simply randomness. The CI/CD system is responsible for customizing this routing and “switching” between canary versions.

The idea is that two Deployments with different versions of the application are deployed in the same namespace. The Pods of different versions have different labels (version: v1 and version: v2).

You have to configure two custom resources:

  • A DestinationRule – defines how to identify different versions of your application (subsets);
  • A VirtualService – defines how to balance traffic between different versions of your application.

Example:

apiVersion: networking.istio.io/v1beta1
kind: DestinationRule
metadata:
  name: productpage-canary
spec:
  host: productpage
  # subsets are only available when accessing the host via the VirtualService from a Pod managed by Istio.
  # These subsets must be defined in the routes.
  subsets:
  - name: v1
    labels:
      version: v1
  - name: v2
    labels:
      version: v2
apiVersion: networking.istio.io/v1beta1
kind: VirtualService
metadata:
  name: productpage-canary
spec:
  hosts:
  - productpage
  http:
  - match:
    - headers:
       cookie:
         regex: "^(.*;?)?(canary=yes)(;.*)?"
    route:
    - destination:
        host: productpage
        subset: v2 # The reference to the subset from the DestinationRule.
  - route:
    - destination:
        host: productpage
        subset: v1

Probability-based routing

apiVersion: networking.istio.io/v1beta1
kind: VirtualService
metadata:
  name: productpage-canary
spec:
  hosts:
  - productpage
  http:
  - route:
    - destination:
        host: productpage
        subset: v1 # The reference to the subset from the DestinationRule.
      weight: 90 # Percentage of traffic that the Pods with the version: v1 label will be getting.
  - route:
    - destination:
        host: productpage
        subset: v2
      weight: 10

Ingress

To use Ingress, you need to:

  • Configure the Ingress controller by adding Istio sidecar to it. In our case, you need to enable the enableIstioSidecar parameter in the ingress-nginx module’s IngressNginxController custom resource.
  • Set up an Ingress that refers to the Service. The following annotations are mandatory for Ingress:
    • nginx.ingress.kubernetes.io/service-upstream: "true" — using this annotation, the Ingress controller sends requests to a single ClusterIP (from Service CIDR) while envoy load balances them. Ingress controller’s sidecar is only catching traffic directed to Service CIDR.
    • nginx.ingress.kubernetes.io/upstream-vhost: myservice.myns.svc — using this annotation, the sidecar container can identify the application service that serves requests.

Examples:

apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: productpage
  namespace: bookinfo
  annotations:
    # Nginx proxies traffic to the ClusterIP instead of pods' own IPs.
    nginx.ingress.kubernetes.io/service-upstream: "true"
    # In Istio, all routing is carried out based on the `Host:` headers.
    # Instead of letting Istio know about the `productpage.example.com` external domain, 
    # we use the internal domain of which Istio is aware.
    nginx.ingress.kubernetes.io/upstream-vhost: productpage.bookinfo.svc
spec:
  rules:
    - host: productpage.example.com
      http:
        paths:
        - path: /
          pathType: Prefix
          backend:
            service:
              name: productpage
              port:
                number: 9080
apiVersion: v1
kind: Service
metadata:
  name: productpage
  namespace: bookinfo
spec:
  ports:
  - name: http
    port: 9080
  selector:
    app: productpage
  type: ClusterIP

Authorization configuration examples

Decision-making algorithm

Caution! The following algorithm for deciding the fate of a request becomes active after AuthorizationPolicy is created for the application:

  • The request is denied if it falls under the DENY policy;
  • The request is allowed if there are no ALLOW policies for the application;
  • The request is allowed if it falls under the ALLOW policy.
  • All other requests are denied.

In other words, if you explicitly deny something, then only this restrictive rule will work. If you explicitly allow something, only explicitly authorized requests will be allowed (however, restrictions will stay in force and have precedence).

Caution! The policies based on high-level parameters like namespace or principal require enabling Istio for all involved applications. Also, there must be organized Mutual TLS between applications.

Examples:

  • Let’s deny POST requests for the myapp application. Since a policy is defined, only POST requests to the application are denied (as per the algorithm above).

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: deny-post-requests
      namespace: foo
    spec:
      selector:
        matchLabels:
          app: myapp
      action: DENY
      rules:
      - to:
        - operation:
            methods: ["POST"]
    
  • Below, the ALLOW policy is defined for the application. It only allows requests from the bar namespace (other requests are denied).

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: deny-all
      namespace: foo
    spec:
      selector:
        matchLabels:
          app: myapp
      action: ALLOW # The default value, can be skipped.
      rules:
      - from:
        - source:
            namespaces: ["bar"]
    
  • Below, the ALLOW policy is defined for the application. Note that it does not have any rules, so not a single request matches it (still, the policy exists). Thus, our decision-making algorithm suggests that if something is allowed, then everything else is denied. In this case, “everything else” includes all the requests.

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: deny-all
      namespace: foo
    spec:
      selector:
        matchLabels:
          app: myapp
      action: ALLOW # The default value, can be skipped.
      rules: []
    
  • Below, the (default) ALLOW policy is defined for the application. Note that it has an empty rule. Any request matches this rule, so it is naturally approved.

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: allow-all
      namespace: foo
    spec:
      selector:
        matchLabels:
          app: myapp
      rules:
      - {}
    

Deny all for the foo namespace

There are two ways you can do that:

  • Explicitly. Here, the DENY policy is created. It has a single {} rule that covers all the requests:

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: deny-all
      namespace: foo
    spec:
      action: DENY
      rules:
      - {}
    
  • Implicitly. Here, the (default) ALLOW policy is created that does not have any rules. Thus, no requests will match it, and the policy will deny all of them.

    apiVersion: security.istio.io/v1beta1
    kind: AuthorizationPolicy
    metadata:
      name: deny-all
      namespace: foo
    spec: {}
    

Deny requests from the foo NS only

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: deny-from-ns-foo
 namespace: myns
spec:
 action: DENY
 rules:
 - from:
   - source:
       namespaces: ["foo"]

Allow requests for the foo NS only

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-intra-namespace-only
 namespace: foo
spec:
 action: ALLOW
 rules:
 - from:
   - source:
       namespaces: ["foo"]

Allow requests from anywhere in the cluster

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-all-from-my-cluster
 namespace: myns
spec:
 action: ALLOW
 rules:
 - from:
   - source:
       principals: ["mycluster.local/*"]

Allow any requests for foo or bar clusters

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-all-from-foo-or-bar-clusters-to-ns-baz
 namespace: baz
spec:
 action: ALLOW
 rules:
 - from:
   - source:
       principals: ["foo.local/*", "bar.local/*"]

Allow any requests from foo or bar clusters where the namespace is baz

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-all-from-foo-or-bar-clusters-to-ns-baz
 namespace: baz
spec:
 action: ALLOW
 rules:
 - from:
   - source: # Logical conjunction is used for the rules below.
       namespaces: ["baz"]
       principals: ["foo.local/*", "bar.local/*"]

Allow from any cluster (via mtls)

Caution! The denying rules (if they exist) have priority over any other rules. See the algorithm.

Example:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-all-from-any-cluster-with-mtls
 namespace: myns
spec:
 action: ALLOW
 rules:
 - from:
   - source:
       principals: ["*"] # To force the MTLS usage.

Allow all requests from anywhere (including no mTLS - plain text traffic)

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-all-from-any
 namespace: myns
spec:
 action: ALLOW
 rules: [{}]

Setting up federation for two clusters using the IstioFederation CR

Cluster A:

apiVersion: deckhouse.io/v1alpha1
kind: IstioFederation
metadata:
  name: cluster-b
spec:
  metadataEndpoint: https://istio.k8s-b.example.com/metadata/
  trustDomain: cluster-b.local

Cluster B:

apiVersion: deckhouse.io/v1alpha1
kind: IstioFederation
metadata:
  name: cluster-a
spec:
  metadataEndpoint: https://istio.k8s-a.example.com/metadata/
  trustDomain: cluster-a.local

Setting up multicluster for two clusters using the IstioMulticluster CR

Cluster A:

apiVersion: deckhouse.io/v1alpha1
kind: IstioMulticluster
metadata:
  name: cluster-b
spec:
  metadataEndpoint: https://istio.k8s-b.example.com/metadata/

Cluster B:

apiVersion: deckhouse.io/v1alpha1
kind: IstioMulticluster
metadata:
  name: cluster-a
spec:
  metadataEndpoint: https://istio.k8s-a.example.com/metadata/

Control the data-plane behavior

[experimental feature] Prevent istio-proxy from terminating before the main application’s connections are closed

By default, during termination, all containers in a Pod, including istio-proxy one, receive SIGTERM signal simultanuesly. But some applications need time to properly handle the termination and sometimes they need to do some network requests. It isn’t possible when the istio-proxy stops before the application do. The solution is to add a preStop hook which evaluates the application’s activity via discovering application’s network sockets and let the sidecar stop when they aren’t in network namespace. The annotation below adds the preStop hook to istio-proxy container in application’s Pod: inject.istio.io/templates: sidecar,d8-hold-istio-proxy-termination-until-application-stops.

Upgrading Istio control-plane

  • Deckhouse allows you to install different control-plane versions simultaneously:
    • A single global version to handle namespaces or Pods with indifferent version (namespace label istio-injection: enabled). It is configured by istio.globalVersion argument in the deckhouse ConfigMap.
    • The other ones are additional, they handle namespaces or Pods with explicitly configured versions (istio.io/rev: v1x13 label for namespace or Pod). They are configured by istio.additionalVersions argument in the deckhouse ConfigMap.
  • Istio declares backward compatibility between data-plane and control-plane in the range of two minor versions: Istio data-plane and control-plane compatibility
  • Upgrade algorithm (i.e. to 1.13):
    • Confugure additional version in CM deckhouse (additionalVersions: ["1.13"]).
    • Wait for the corresponding pod istiod-v1x13-xxx-yyy to appear in d8-istiod namespace.
    • For every application Namespase with istio enabled:
      • Change istio-injection: enabled lable to istio.io/rev: v1x13.
      • Recreate the Pods in namespace (one at a time), simultaneously monitoring the application workability.
    • Reconfigure istio.globalVersion to 1.13 and remove the additionalVersions configuration.
    • Make sure, the old istiod Pod has gone.
    • Change application namespace labels to istio-injection: enabled.

To find all Pods with old Istio revision, execute the following command:

kubectl get pods -A -o json | jq --arg revision "v1x12" \
  '.items[] | select(.metadata.annotations."sidecar.istio.io/status" // "{}" | fromjson | 
   .revision == $revision) | .metadata.namespace + "/" + .metadata.name'