TCP Traffic Shifting

This task shows you how to gradually migrate TCP traffic from one version of a microservice to another. For example, you might migrate TCP traffic from an older version to a new version.

A common use case is to migrate TCP traffic gradually from one version of a microservice to another. In Istio, you accomplish this goal by configuring a sequence of rules that route a percentage of TCP traffic to one service or 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.

Before you begin

Set up the test environment

  1. To get started, create a namespace for testing TCP traffic shifting and label it to enable automatic sidecar injection.

    1. $ kubectl create namespace istio-io-tcp-traffic-shifting
    2. $ kubectl label namespace istio-io-tcp-traffic-shifting istio-injection=enabled

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

    Zip

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

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

    Zip

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

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

Apply weight-based TCP routing

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

    Zip

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

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

    1. $ for i in {1..20}; do \
    2. kubectl exec "$(kubectl get pod -l app=sleep -n istio-io-tcp-traffic-shifting -o jsonpath={.items..metadata.name})" \
    3. -c sleep -n istio-io-tcp-traffic-shifting -- sh -c "(date; sleep 1) | nc $INGRESS_HOST $TCP_INGRESS_PORT"; \
    4. done
    5. one Mon Nov 12 23:24:57 UTC 2018
    6. one Mon Nov 12 23:25:00 UTC 2018
    7. one Mon Nov 12 23:25:02 UTC 2018
    8. one Mon Nov 12 23:25:05 UTC 2018
    9. one Mon Nov 12 23:25:07 UTC 2018
    10. one Mon Nov 12 23:25:10 UTC 2018
    11. one Mon Nov 12 23:25:12 UTC 2018
    12. one Mon Nov 12 23:25:15 UTC 2018
    13. one Mon Nov 12 23:25:17 UTC 2018
    14. one Mon Nov 12 23:25:19 UTC 2018
    15. ...

    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.

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

    Zip

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

    Wait a few seconds for the new rules to propagate.

  4. Confirm that the rule was replaced:

    1. $ kubectl get virtualservice tcp-echo -o yaml -n istio-io-tcp-traffic-shifting
    2. apiVersion: networking.istio.io/v1beta1
    3. kind: VirtualService
    4.   ...
    5. spec:
    6.   ...
    7.   tcp:
    8.   - match:
    9.     - port: 31400
    10.     route:
    11.     - destination:
    12.         host: tcp-echo
    13.         port:
    14.           number: 9000
    15.         subset: v1
    16.       weight: 80
    17.     - destination:
    18.         host: tcp-echo
    19.         port:
    20.           number: 9000
    21.         subset: v2
    22.       weight: 20

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

    1. $ for i in {1..20}; do \
    2. kubectl exec "$(kubectl get pod -l app=sleep -n istio-io-tcp-traffic-shifting -o jsonpath={.items..metadata.name})" \
    3. -c sleep -n istio-io-tcp-traffic-shifting -- sh -c "(date; sleep 1) | nc $INGRESS_HOST $TCP_INGRESS_PORT"; \
    4. done
    5. one Mon Nov 12 23:38:45 UTC 2018
    6. two Mon Nov 12 23:38:47 UTC 2018
    7. one Mon Nov 12 23:38:50 UTC 2018
    8. one Mon Nov 12 23:38:52 UTC 2018
    9. one Mon Nov 12 23:38:55 UTC 2018
    10. two Mon Nov 12 23:38:57 UTC 2018
    11. one Mon Nov 12 23:39:00 UTC 2018
    12. one Mon Nov 12 23:39:02 UTC 2018
    13. one Mon Nov 12 23:39:05 UTC 2018
    14. one Mon Nov 12 23:39:07 UTC 2018
    15. ...

    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

  1. Remove the sleep sample, tcp-echo application, and routing rules:

    ZipZipZip

    1. $ kubectl delete -f @samples/tcp-echo/tcp-echo-all-v1.yaml@ -n istio-io-tcp-traffic-shifting
    2. $ kubectl delete -f @samples/tcp-echo/tcp-echo-services.yaml@ -n istio-io-tcp-traffic-shifting
    3. $ kubectl delete -f @samples/sleep/sleep.yaml@ -n istio-io-tcp-traffic-shifting
    4. $ kubectl delete namespace istio-io-tcp-traffic-shifting