Test in production

This is work in progress. We will add its sections in pieces. Your feedback is welcome at discuss.istio.io.

Test your microservice, in production!

Testing individual microservices

  1. Issue an HTTP request from the testing pod to one of your services:

    1. $ kubectl exec -it $(kubectl get pod -l app=sleep -o jsonpath='{.items[0].metadata.name}') -- curl http://ratings:9080/ratings/7

Chaos testing

Perform some chaos testing in production and see how your application reacts. After each chaos operation, access the application’s webpage and see if anything changed. Check the pods’ status with kubectl get pods.

  1. Terminate the details service in one pod.

    1. $ kubectl exec -it $(kubectl get pods -l app=details -o jsonpath='{.items[0].metadata.name}') -- pkill ruby

  2. Check the pods status:

    1. $ kubectl get pods
    2. NAME                            READY   STATUS    RESTARTS   AGE
    3. details-v1-6d86fd9949-fr59p     1/1     Running   1          47m
    4. details-v1-6d86fd9949-mksv7     1/1     Running   0          47m
    5. details-v1-6d86fd9949-q8rrf     1/1     Running   0          48m
    6. productpage-v1-c9965499-hwhcn   1/1     Running   0          47m
    7. productpage-v1-c9965499-nccwq   1/1     Running   0          47m
    8. productpage-v1-c9965499-tjdjx   1/1     Running   0          48m
    9. ratings-v1-7bf577cb77-cbdsg     1/1     Running   0          47m
    10. ratings-v1-7bf577cb77-cz6jm     1/1     Running   0          47m
    11. ratings-v1-7bf577cb77-pq9kg     1/1     Running   0          48m
    12. reviews-v1-77c65dc5c6-5wt8g     1/1     Running   0          47m
    13. reviews-v1-77c65dc5c6-kjvxs     1/1     Running   0          48m
    14. reviews-v1-77c65dc5c6-r55tl     1/1     Running   0          47m
    15. sleep-88ddbcfdd-l9zq4           1/1     Running   0          47m

    Note that the first pod was restarted once.

  3. Terminate the details service in all its pods:

    1. $ for pod in $(kubectl get pods -l app=details -o jsonpath='{.items[*].metadata.name}'); do echo terminating $pod; kubectl exec -it $pod -- pkill ruby; done

  4. Check the webpage of the application:

    Bookinfo Web Application, details unavailable

    Bookinfo Web Application, details unavailable

    Note that the details section contains error messages instead of book details.

  5. Check the pods status:

    1. $ kubectl get pods
    2. NAME                            READY   STATUS    RESTARTS   AGE
    3. details-v1-6d86fd9949-fr59p     1/1     Running   2          48m
    4. details-v1-6d86fd9949-mksv7     1/1     Running   1          48m
    5. details-v1-6d86fd9949-q8rrf     1/1     Running   1          49m
    6. productpage-v1-c9965499-hwhcn   1/1     Running   0          48m
    7. productpage-v1-c9965499-nccwq   1/1     Running   0          48m
    8. productpage-v1-c9965499-tjdjx   1/1     Running   0          48m
    9. ratings-v1-7bf577cb77-cbdsg     1/1     Running   0          48m
    10. ratings-v1-7bf577cb77-cz6jm     1/1     Running   0          48m
    11. ratings-v1-7bf577cb77-pq9kg     1/1     Running   0          49m
    12. reviews-v1-77c65dc5c6-5wt8g     1/1     Running   0          48m
    13. reviews-v1-77c65dc5c6-kjvxs     1/1     Running   0          49m
    14. reviews-v1-77c65dc5c6-r55tl     1/1     Running   0          48m
    15. sleep-88ddbcfdd-l9zq4           1/1     Running   0          48m

    The first pod restarted twice and two other details pods restarted once. You may experience the Error and the CrashLoopBackOff statuses until the pods reach Running status.

In both cases, the application did not crash. The crash in the details microservice did not cause other microservices to fail. This behavior means you did not have a cascading failure in this situation. Instead, you had gradual service degradation: despite one microservice crashing, the application could still provide useful functionality. It displayed the reviews and the basic information about the book.

You are ready to add a new version of the reviews application.