Skip to content

Container for testing app failures in orchestrators. It aims to misbehave.

License

Notifications You must be signed in to change notification settings

BretFisher/browncoat

Repository files navigation

Browncoat 🚀🤠 - it aims to misbehave

Lint Code Base Build Images

A container image for testing various app states in orchestration. It aims to misbehave in simple ways based on its config so you can ensure your orchestration responds how you want/expect. You can set this Node.js web app to:

  1. Have a slow startup
  2. Fail startup
  3. Exit while running, with error code
  4. Enable or disable Docker healthchecks
  5. Fail healthchecks
  6. Slow healthchecks
  7. Responds with a different 20x HTTP status code for different image tags (versions)

You can configure all these things at setup/runtime via environment variables, and some of them while the container is running, via specific routes (URLs).

Image tags

The images are available on Docker Hub and GitHub Container Registry. The app code is the same in all images, but different image tags have different Dockerfile and env vars set.

Image tags marked healthcheck will have a HEALTHCHECK set in Dockerfile.

HTTP Codes are different per image tag version to make container update identification easier when using httping or similar tool.

Here's a list of images stored on Docker Hub and how they are different.

Image Version Healcheck /healthz HTTP Code Note
bretfisher/browncoat:latest v1 False 201 Identical to v1
bretfisher/browncoat:v1 v1 False 201
bretfisher/browncoat:v2 v2 False 202
bretfisher/browncoat:v3 v3 False 203 Fails on start
bretfisher/browncoat:healthcheck v1 True 201
bretfisher/browncoat:v1.healthcheck v1 True 201 Identical to healthcheck
bretfisher/browncoat:v2.healthcheck v2 True 202
bretfisher/browncoat:v3.healthcheck v3 True 203 Healthcheck returns 500

Startup config options via environment variables

  • PORT - int, port to listen on, defaults to 80
  • VERSION - string, doesn't change the functionality, but used to create different images for update testing
  • FAIL_STARTUP - true/false, exit(1) right at start
  • DELAY_STARTUP - int, set to milliseconds the app will wait before listening on PORT
  • DELAY_HEALTHCHECK - int, set to milliseconds the app will wait before responding on /healthz
  • HAPPYHEALTHCHECK - true/false, false will cause /healthz to return 500, defaults to true on v1/v2, false on v3
  • ENABLE_LOGGER - true/false, defaults to true, logs all HTTP requests to stdout

Useful routes (URLs)

  • / - Returns a random image of Serenity crew.

  • /healthz - Returns JSON of environment variables. Returns 500 if HAPPYHEALTHCHECK is false. Returns a 20x if true. Actual status code matches the version of app running. This is useful for using tools like httping to test connectivity while doing rolling or blue/green updates and being able to see which is responding.

    • 201 - v1
    • 202 - v2
    • 203 - v3
  • /fail - Returns 500 and exits the Node.js process with an error code.

  • /togglehealthcheck = By default the healthcheck URL returns 20x. Hitting this URL will cause them to start returning 500. Hitting it again will return to 20x.

Examples with Docker Run

Show slow startup

Start the v1 image with a slow startup of 5 seconds (Node.js will wait 5s before listening on the port). This is useful to simulate apps that take more time to startup, and also simulate distributed environments where not all things start in "proper" order.

docker run -p 80:80 --env DELAY_STARTUP=5000 bretfisher/browncoat:v1

You could use httping or curl to show how the first 5 seconds of that container running, it doesn't repond to HTTP requests.

Fail on app startup

Next, start the v2 image and cause it to exit with an error code. This will simulate what would happen if you didn't test proper running of a container in CI and something was set wrong when you deployed container. Do you systems catch this type of quick failure that may not show up in unit tests? Note: bretfisher/browncoat:v3 does this startup failure by default.

docker run -p 80:80 --env FAIL_STARTUP=true bretfisher/browncoat:v2

Note: docker run and docker compose don't react to failed healthchecks like orchestrators, but they do provide healthcheck status.

Examples with Swarm

Create the initial Service

Create a three-container service that doesn't have a Dockerfile healthcheck built-in.

docker service create --name browncoat -p 80:80 --replicas 3 bretfisher/browncoat

Next, install httping on your host to monitor the health endpoint /healthz with something like httping -i .1 -GsY localhost/healthz. You can also run it from a container: docker run --rm bretfisher/httping -i .1 -GsY <hostIP>/healthz. Notice you get a 201 response code. v2 will have a 202 response code to make it easier to identity how updates are distributing connections.

Now we can update the service with a new image.

docker service update --image bretfisher/browncoat:v2 browncoat

Notice some connections will fail. This is normal without a healthcheck. Docker has no awareness of if you're container is truly "ready" for connections and starts sending them to new containers before app is listening. This gets worse as your DELAY_STARTUP increases.

Now let's update to another new image. This time you should have zero connection failures. The new image has HEALTHCHECK command in it, so Swarm uses it to wait for ready state.

docker service update --image bretfisher/browncoat:healthcheck browncoat

Note: this YAML should also work with docker compose but it won't replace services, it'll only indicate the health status with docker compose ps. This is useful for using the depends_on: + condition: service_healthy to have one service startup wait for another service to be healthy.

Examples with Kubernetes

Kubernetes ignores Dockerfile HEALTHCHECK commands, so you'll need to add them to your Pod spec. In Kubernetes we have two main checks: livenessProbe and readinessProbe. A little mnemonic is livenessProbe to check if container is live, and readinessProbe to check if container is ready to serve traffic But For more info you can check the docs

Create the initial Kubernetes Deployment

The below creates a Deployment with 5 pods without livenessProbe or readinessProbe and a Service to expose the pods.

kubectl apply -f kubernetes-examples/browncoat-v1.yaml

Causing a rolling update outages due to a lack of probes

Now let's update that Deployment with the default rolling update style. We will change the container image to bretfisher/browncoat:v2 and add the DELAY_STARTUP of 5s to simulate a small delay and cause some connection failures during the rolling update. Probes would have helped prevent connections from being directed to the new Pod until it was ready. Without probes, Kubernetes is blind and will start directing Service traffic to the new Pod as soon as it started.

To follow the rollout use is command in a different window kubectl rollout status deploy/browncoat

To monitor the Service connections, run httping in a different window

kubectl run httping -it --image bretfisher/httping --rm=true -- browncoat/healthz 

Now apply the new image and env var

kubectl apply -f kubernetes-examples/browncoat-v2.yaml

Add readinessProbe and monitor a new

Now we will add a readinessProbe to check if the endpoint healthz returns a successful response and will change to the image v1 to see a different response code (201).

kubectl apply -f  kubernetes-examples/browncoat-v1-withProbe.yaml

This is a Node.js app using Express for easier http servering

and Stoppable for better graceful shutdowns.