In this tutorial, we'll show you how to create and validate a liveness probe for a Kubernetes application. We'll walk through deploying the Gremlin Kubernetes agent, deploying an application with a liveness probe, run a Latency attack, then validate that the probe triggers.
By reading this tutorial, you'll learn how to create a liveness probe and validate that it works as expected.
This tutorial will show you how to:
In Kubernetes, a probe is an automated process that checks the state of a container at different stages in its lifecycle, or at regular intervals. Kubernetes includes three types of probes: liveness probes, startup probes, and readiness probes.
For more on the different types of probes, see the Kubernetes docs. The Google Cloud team also has a great blog post on Kubernetes probes and best practices.
Before starting this tutorial, you’ll need the following:
For this tutorial, you'll need a Kubernetes cluster with at least two worker nodes. If you already have a cluster, you can continue to step 2. If not, you can create your own cluster using a tool like K3s, minikube, or kubeadm, or use a cloud Kubernetes platform like Amazon EKS, Azure Kubernetes Service, or Google Kubernetes Engine.
Once your cluster is deployed, verify that you have at least two active worker nodes by opening a terminal and running the following command :
1kubectl get nodes
In this example, we have one control plane node and two workers:
1NAME STATUS ROLES AGE VERSION2gremlin-lab-control-plane Ready control-plane,master 7d v1.21.13gremlin-lab-worker Ready <none> 7d v1.21.14gremlin-lab-worker2 Ready <none> 7d v1.21.1
The Gremlin Kubernetes agent is the recommended method of deploying Gremlin to a Kubernetes cluster. It allows Gremlin to detect resources such as Pods, Deployments, and DaemonSets, and select them as targets for running experiments. If you've already installed and configured the agent, skip to step 3.
To install the agent, follow the installation instructions in the documentation. Make sure to replace
$GREMLIN_TEAM_ID with your actual team ID,
$GREMLIN_TEAM_SECRET with your team secret key, and
$GREMLIN_CLUSTER_ID with a unique name for this cluster. You'll use the cluster ID to identify the cluster in the Gremlin web app. You can learn more about team IDs and secret keys in the Authentication documentation.
Once the agent is installed, you can confirm that it's configured correctly by logging into your Gremlin account, opening the Agents page, then clicking the Kubernetes tab. Your cluster will be listed with the name provided by
Now we'll deploy an application and create a liveness probe. For this tutorial, we'll use the open source Online Boutique application created by the Google Cloud Platform team. Online Boutique is a demo e-commerce website consisting of 12 services, each one running in a separate Deployment. Several of these Deployments already have liveness probes configured. The one we'll focus on for this tutorial is the frontend, which has both a liveness and readiness probe.
If we look at the application's Kubernetes manifest, we see that there's a liveness probe configured to run an HTTP request over port
8080 on the container's
/_healthz endpoint. Kubernetes will wait 10 seconds before running the first check according to the line
initialDelaySeconds: 10 , and since
periodSeconds isn't defined, Kubernetes uses the default value of 10 seconds. By default, the probe will also time out after 1 second (
timeoutSeconds: 1), will retry three times before reporting a failure (
failureThreshold: 3), and will report a success for each successful check (
successThreshold: 1). According to line 238, it will also include a cookie in the HTTP headers identifying the request as coming from the liveness probe.
To summarize: the frontend Pod defines a liveness probe that will send a request to the Pod's
/_healthz URL on port
8080 every 10 seconds. It will wait 10 seconds before running the first check, and will require three consecutive failed checks before reporting a failure.
The frontend also has a readiness probe that follows the same behavior. The only difference is that this probe's HTTP cookie is set to
shop_session-id=x-readiness-probe instead of
To deploy the Online Boutique, either follow the quick start guide or simply run the following command:
1kubectl apply -f https://raw.githubusercontent.com/GoogleCloudPlatform/microservices-demo/release/v0.3.6/release/kubernetes-manifests.yaml
You can check on the status of the deployment by running:
1kubectl get deployments
1NAMESPACE NAME READY UP-TO-DATE AVAILABLE AGE2default adservice 1/1 1 1 5d21h3default cartservice 1/1 1 1 5d21h4default checkoutservice 1/1 1 1 5d21h5default currencyservice 1/1 1 1 5d21h6default emailservice 1/1 1 1 5d21h7default frontend 1/1 1 1 5d21h8default loadgenerator 1/1 1 1 5d21h9default paymentservice 1/1 1 1 5d21h10default productcatalogservice 1/1 1 1 5d21h11default recommendationservice 1/1 1 1 5d21h12default redis-cart 1/1 1 1 5d21h13default shippingservice 1/1 1 1 5d21h
We can check the status of our liveness probe by running
kubectl describe on the frontend Deployment and looking for the
Liveness field. Note that the
failure parameters are commented out, meaning they're using their default values of 1 and 3 respectively:
1kubectl describe deploy/frontend
1...2Liveness: http-get http://:8080/_healthz delay=10s timeout=1s period=10s #success=1 #failure=33...
The last step is to ensure that our liveness probe works the way we expect it to. In step 3, we found that our liveness probe is configured to run every 10 seconds, will fail after 3 unsuccessful checks, and considers a check unsuccessful if it takes longer than 1 second for the endpoint to respond. To test it, we'll use a Latency attack to delay all network traffic from the Pod. We'll add enough latency to exceed the timeout period for at least 30 seconds to ensure at least three failed checks in a row.
First, log into the Gremlin web app and create a new attack by navigating to Attacks, then New Attack. Select the Infrastructure tab, then select Kubernetes. Expand Deployments, then select the frontend Deployment.
Next, click on Choose a Gremlin to expand the attack selection box, select the Network category, then select Latency. Increase MS to
1000 to add one full second of latency, then click Unleash Gremlin to start the attack.
While the attack is running, use
kubectl (or a monitoring tool) to monitor the state of your frontend Deployment. You can check the current health of the frontend Deployment by running the following command:
1kubectl get deploy/frontend
1NAME READY UP-TO-DATE AVAILABLE AGE2frontend 1/1 1 1 141m
watchcommand. For example, this command refreshes the output of kubectl every 2 seconds:
watch -n 2 'kubectl get deploy/frontend'
Shortly after the attack starts, we see the following:
1NAME READY UP-TO-DATE AVAILABLE AGE2frontend 0/1 1 0 153m
We also see an interesting development in the Gremlin web app. Instead of showing the attack stage as "Successful", it instead shows "Agent Aborted".
If we scroll down to the Executions section and click on our host, we see this in the execution log:
12022-04-19 17:15:30 - Gremlin did not exit normally: container failure: Container runtime killed Gremlin during attack:2Reason: Unknown3TargetStatus: Exited4Details: The container runtime kills the Gremlin attack when the target of the attack is terminated, either by intentional removal by an operator, or due the target container exiting from an error or a failed health check. For more information about this behavior see our Knowledge Base: https://support-site.gremlin.com/support/solutions/articles/151000015041-exit-code-137
In other words, the liveness probe worked as intended and terminated the frontend Pod when it took too long to respond. Because the Gremlin container running the latency attack no longer had a valid attack target, it also terminated. This is why the web app reports "Agent Aborted." If we look back at our kubectl output a few moments later, we'll see that the Pod has restarted and the Deployment is healthy again:
1NAME READY UP-TO-DATE AVAILABLE AGE2frontend 1/1 1 1 154m
This shows that our liveness probe is working reliably and keeping our application available and responsive.
In this tutorial, we used Gremlin to validate that our liveness probe will keep our frontend Deployment running reliably. There are additional steps we can take to improve reliability even further, such as scaling up our Deployment to two or more Pods to avoid the small amount of downtime that happened when the probe terminated the old Pod.
For more ideas on experiments to run on your Kubernetes cluster, check out our white paper on the first 5 Chaos Experiments to run on Kubernetes.
Gremlin empowers you to proactively root out failure before it causes downtime. See how you can harness chaos to build resilient systems by requesting a demo of Gremlin.Get started