Your App Is Alive Michelle Noorali Developer at Deis @michellenoorali

Health Checks

What?

a way to monitor your app’s health A Health Check is

Is your app… responsive?

Is your app… Performant?

Is your app… Stable?

What does it look like?

• /healthcheck or /healthz endpoint • checked by external system at intervals • On success, do nothing • On failure, notify and heal

In Kubernetes Land… •Health checks are built in! •But first…

Introducing Puffy • … a simple go web server

Let’s review! • I have an app (Puffy) and I want to deploy it onto my Kubernetes cluster so… • I write a manifest … • … which references an image • … which will contain all the instructions for my app

Key Concepts • Image • Union of layered filesystems • Immutable • Container (Docker, rkt) • runtime instance of an image

Kubernetes Concepts • Pod • Represents a logical application • One or more containers • Manifest • yaml file • serves as resource (ex. pod) definition • kubectl • CLI tool to control kubernetes cluster

Pre-requisites • Running Kubernetes cluster on GCE [Google Compute Engine] • Docker Image for app stored on DockerHub • DockerHub = place to store images • Can also use Quay.io or GCR [Google Container Registry]

Puffy Repo

Dockerfile

Makefile

Manifest file

Life of Puffy: The Skeleton • puffy starts out as an image • docker build -t mnoorali/puffy:1.0.0 . • docker push mnoorali/puffy:1.0.0

Demo: Create a Puffy Pod in Kubernetes Cluster

Life of Puffy: It’s Alive • create a pod definition in manifest • create resource in kubernetes from manifest • kubectl create -f puffy-pod.yaml • watch it come to life • kubectl get pods • Check it out • curl pod_ip:port/index on a node

Let’s make sure puffy is ok. • Here comes the health check part…

Two Probes walk into a coffeeshop… • Readiness Probe • Liveness Probe

Wait. What is this probe thing? • a probe = a diagnostic periodically performed on a container • Three ways to perform a diagnostic: • Execute a command in a container • Successful if exit status code 0 • Perform tcp check on a container’s IP address on a specific port • Successful if the port is open • Perform an HTTP Get against the container’s IP address on a specific port • Successful if response has status code >= 200 and <400

Back to the probes… • Readiness Probe • Is my app ready to serve traffic? • On failure, it stops serving requests • Liveness Probe • Is my app in a good state? • On failure, it restarts the pod

Define a Liveness Probe in the Pod Manifest

Apply the Changes in the Pod Manifest to the Cluster • kubectl delete -f puffy-pod.yaml • kubectl get pods • edit puffy-pod.yaml • kubectl create -f pod-puffy.yaml

FAIL • But, what happened?

Oops • forgot to add the /healthz endpoint • kubectl delete -f puffy-pod.yaml

Add /healthz • Define /healthz in app (common convention) • Re-build/re-tag image • Push image to registry • Edit image declaration in manifest

Bring Puffy Back! • kubectl create -f puffy-pod.yaml • kubectl get pods • kubectl describe pod puffy • On node, curl ip:port/healthz

Summary • automate monitoring apps with health checks • Use Kubernetes’s built in health check features • readinessProbe • Is your app ready to serve traffic? • livenessProbe • Is your app in a good state?

That’s it folks. Visit puffy at www.github.com/michelleN/puffy Will have a link to slides above and on twitter: @michellenoorali Thank You!