TCP Traffic Shifting

5 minute read page test

This task shows you how to shift TCP traffic from one version of a microservice to another.

A common use case is to migrate TCP traffic gradually from an older version of a microservice to a new one. In Istio, you accomplish this goal by configuring a sequence of routing rules that redirect a percentage of TCP traffic from one destination to another.

In this task, you will send 100% of the TCP traffic to tcp-echo:v1. Then, you will route 20% of the TCP traffic to tcp-echo:v2 using Istio’s weighted routing feature.

This document configures internal mesh (east-west) traffic that requires Gateway API features that are either experimental or Istio specific. Before using the Gateway API instructions, make sure to:

Install the experimental version of the Gateway API CRDs:

$ kubectl kustomize "github.com/kubernetes-sigs/gateway-api/config/crd/experimental?ref=444631bfe06f3bcca5d0eadf1857eac1d369421d" | kubectl apply -f -

Configure Istio to read the alpha Gateway API resources by setting the PILOT_ENABLE_ALPHA_GATEWAY_API environment variable to true when installing Istio:
```
$ istioctl install --set values.pilot.env.PILOT_ENABLE_ALPHA_GATEWAY_API=true --set profile=minimal -y
```

Before you begin

Setup Istio by following the instructions in the Installation guide.
Review the Traffic Management concepts doc.

Set up the test environment

To get started, create a namespace for testing TCP traffic shifting.
```
$ kubectl create namespace istio-io-tcp-traffic-shifting
```

Deploy the sleep sample app to use as a test source for sending requests.

$ kubectl apply -f @samples/sleep/sleep.yaml@ -n istio-io-tcp-traffic-shifting

Deploy the v1 and v2 versions of the tcp-echo microservice.

$ kubectl apply -f @samples/tcp-echo/tcp-echo-services.yaml@ -n istio-io-tcp-traffic-shifting

Apply weight-based TCP routing

Route all TCP traffic to the v1 version of the tcp-echo microservice.

$ kubectl apply -f @samples/tcp-echo/tcp-echo-all-v1.yaml@ -n istio-io-tcp-traffic-shifting

$ kubectl apply -f @samples/tcp-echo/gateway-api/tcp-echo-all-v1.yaml@ -n istio-io-tcp-traffic-shifting

Determine the ingress IP and port:

Follow the instructions in Determining the ingress IP and ports to set the TCP_INGRESS_PORT and INGRESS_HOST environment variables.

Use the following commands to set the SECURE_INGRESS_PORT and INGRESS_HOST environment variables:

$ kubectl wait --for=condition=programmed gtw tcp-echo-gateway -n istio-io-tcp-traffic-shifting
$ export INGRESS_HOST=$(kubectl get gtw tcp-echo-gateway -n istio-io-tcp-traffic-shifting -o jsonpath='{.status.addresses[0].value}')
$ export TCP_INGRESS_PORT=$(kubectl get gtw tcp-echo-gateway -n istio-io-tcp-traffic-shifting -o jsonpath='{.spec.listeners[?(@.name=="tcp-31400")].port}')

Confirm that the tcp-echo service is up and running by sending some TCP traffic.

$ export SLEEP=$(kubectl get pod -l app=sleep -n istio-io-tcp-traffic-shifting -o jsonpath={.items..metadata.name})
$ for i in {1..20}; do \
kubectl exec "$SLEEP" -c sleep -n istio-io-tcp-traffic-shifting -- sh -c "(date; sleep 1) | nc $INGRESS_HOST $TCP_INGRESS_PORT"; \
done
one Mon Nov 12 23:24:57 UTC 2022
one Mon Nov 12 23:25:00 UTC 2022
one Mon Nov 12 23:25:02 UTC 2022
one Mon Nov 12 23:25:05 UTC 2022
one Mon Nov 12 23:25:07 UTC 2022
one Mon Nov 12 23:25:10 UTC 2022
one Mon Nov 12 23:25:12 UTC 2022
one Mon Nov 12 23:25:15 UTC 2022
one Mon Nov 12 23:25:17 UTC 2022
one Mon Nov 12 23:25:19 UTC 2022
...

You should notice that all the timestamps have a prefix of one, which means that all traffic was routed to the v1 version of the tcp-echo service.

Transfer 20% of the traffic from tcp-echo:v1 to tcp-echo:v2 with the following command:

$ kubectl apply -f @samples/tcp-echo/tcp-echo-20-v2.yaml@ -n istio-io-tcp-traffic-shifting

$ kubectl apply -f @samples/tcp-echo/gateway-api/tcp-echo-20-v2.yaml@ -n istio-io-tcp-traffic-shifting

Wait a few seconds for the new rules to propagate and then confirm that the rule was replaced:

$ kubectl get virtualservice tcp-echo -o yaml -n istio-io-tcp-traffic-shifting
apiVersion: networking.istio.io/v1beta1
kind: VirtualService
  ...
spec:
  ...
  tcp:
  - match:
    - port: 31400
    route:
    - destination:
        host: tcp-echo
        port:
          number: 9000
        subset: v1
      weight: 80
    - destination:
        host: tcp-echo
        port:
          number: 9000
        subset: v2
      weight: 20

$ kubectl get tcproute tcp-echo -o yaml -n istio-io-tcp-traffic-shifting
apiVersion: gateway.networking.k8s.io/v1alpha2
kind: TCPRoute
  ...
spec:
  parentRefs:
  - group: gateway.networking.k8s.io
    kind: Gateway
    name: tcp-echo-gateway
    sectionName: tcp-31400
  rules:
  - backendRefs:
    - group: ""
      kind: Service
      name: tcp-echo-v1
      port: 9000
      weight: 80
    - group: ""
      kind: Service
      name: tcp-echo-v2
      port: 9000
      weight: 20
...

Send some more TCP traffic to the tcp-echo microservice.

$ export SLEEP=$(kubectl get pod -l app=sleep -n istio-io-tcp-traffic-shifting -o jsonpath={.items..metadata.name})
$ for i in {1..20}; do \
kubectl exec "$SLEEP" -c sleep -n istio-io-tcp-traffic-shifting -- sh -c "(date; sleep 1) | nc $INGRESS_HOST $TCP_INGRESS_PORT"; \
done
one Mon Nov 12 23:38:45 UTC 2022
two Mon Nov 12 23:38:47 UTC 2022
one Mon Nov 12 23:38:50 UTC 2022
one Mon Nov 12 23:38:52 UTC 2022
one Mon Nov 12 23:38:55 UTC 2022
two Mon Nov 12 23:38:57 UTC 2022
one Mon Nov 12 23:39:00 UTC 2022
one Mon Nov 12 23:39:02 UTC 2022
one Mon Nov 12 23:39:05 UTC 2022
one Mon Nov 12 23:39:07 UTC 2022
...

You should now notice that about 20% of the timestamps have a prefix of two, which means that 80% of the TCP traffic was routed to the v1 version of the tcp-echo service, while 20% was routed to v2.

Understanding what happened

In this task you partially migrated TCP traffic from an old to new version of the tcp-echo service using Istio’s weighted routing feature. Note that this is very different than doing version migration using the deployment features of container orchestration platforms, which use instance scaling to manage the traffic.

With Istio, you can allow the two versions of the tcp-echo service to scale up and down independently, without affecting the traffic distribution between them.

For more information about version routing with autoscaling, check out the blog article Canary Deployments using Istio.

Cleanup

Remove the routing rules:

$ kubectl delete -f @samples/tcp-echo/tcp-echo-all-v1.yaml@ -n istio-io-tcp-traffic-shifting

$ kubectl delete -f @samples/tcp-echo/gateway-api/tcp-echo-all-v1.yaml@ -n istio-io-tcp-traffic-shifting

Remove the sleep sample, tcp-echo application and test namespace:

$ kubectl delete -f @samples/sleep/sleep.yaml@ -n istio-io-tcp-traffic-shifting
$ kubectl delete -f @samples/tcp-echo/tcp-echo-services.yaml@ -n istio-io-tcp-traffic-shifting
$ kubectl delete namespace istio-io-tcp-traffic-shifting