Installing the Kubernetes agent enables Blue Matador to collect and analyze server metrics on your Kubernetes nodes as well as the state of many Kubernetes resources. After installation, the following events will be detected on your servers without configuring any thresholds:
Node CPU Anomalies
Server Time Drift
Kubernetes API Health
Kubernetes Component Statuses
Kubernetes DaemonSet Health
Kubernetes Deployment Health
Kubernetes Pod Failures
Kubernetes Node Capacity
Kubernetes OOM Containers & Container Restarts
Kubernetes Service Health
Kubernetes Job Failures
Installing the Blue Matador Linux Agent requires either an active trial or paid account. Contact sales to get started. You will also need to meet the following requirements:
Kubernetes version 1.10+
kubectl access to your cluster
Ability to set up RBAC (Role-Based Access Control) with your credentials
1. Log in to the app and navigate to the Integrations page via Setup > Integrations
2. Expand the Kubernetes installation by clicking on the Install button on the Kubernetes tile
3. Specify a name for your Kubernetes cluster. This will help you identify your Kubernetes resources more easily in a multi-cluster environment, and helps the agent know which Kubernetes nodes belong in the same cluster.
4. Specify the namespace you would like the agent and related resources to be created in. If the namespace does not already exist, you must first create it using kubectl create namespace <name>. If you are unsure what to use, use the default namespace.
5. Create a file named bluematador-rbac.yaml with the contents of the Role-Based Access Control yaml which has been configured with your namespace.
6. Create a file named bluematador-agent.yaml with the contents of the DaemonSet config which will run the Blue Matador agent and has been generated with your namespace, cluster name, Blue Matador account ID, and API key.
7. Create the resources to run the agent in your Kubernetes cluster using kubectl
kubectl create -f bluematador-rbac.yaml,bluematador-agent.yaml
8. Verify the installation by checking the number of connected nodes against the number of nodes in your cluster.
If you are unable to verify that the agent is connected, first make sure that the pods for the agent are running.
kubectl get pods --all-namespaces --selector name=bluematador-agent
You should see an output similar to this:
If there are no pods, then the kubelet may not have been able to schedule the pods. Ensure that there are no taints or tolerations that would affect the scheduling of the pod, or if there are, then modify the DaemonSet config to tolerate them. Also ensure that each node has sufficient CPU and memory to run the agent. In small clusters of less than 10 nodes and 100 pods, the agent should be able to run on as little as 64Mi memory and 50m CPU. Medium-sized clusters should allocate at least 128Mi memory and 200m CPU, while large clusters (more than 50 nodes or 5000 pods) should allocate at least 256Mi memory and 500m CPU.
If the pods are created but the status is not Running for some pods, describe the pod to see if there is an event that indicates what is wrong. There may have been an issue pulling the docker image for the agent, and a network policy may need to be changed to allow this image to be pulled.
kubectl describe <pod name>
Once all the containers are running, you can verify that the agent is connected by viewing the logs for a pod. Typically the logs will be sparse, but look out for any errors related to Kubernetes data collection or authentication.
Verbose logging for the agent can be enabled by modifying the DaemonSet to add the environment variable BLUEMATADOR_VERBOSE with a value of 5. This will enable debug logging in the agent and can help provide more information if resources and metrics are not being properly collected.
... containers: - name: bluematador-agent image: bluematador/agent:2.0.3 imagePullPolicy: Always env: - name: BLUEMATADOR_VERBOSE value: "5" ...
Then, update the DaemonSet in your cluster
kubectl apply bluematador-daemonset.yaml
In order for the agent to connect to Blue Matador’s servers, ensure it has outgoing network connectivity to app.bluematador.com:443 and bluematador-flint-modules.s3.amazonaws.com:443.
If you are running the agent behind a proxy, set the HTTP_PROXY and HTTPS_PROXY environment variables in the DaemonSet config to allow our agent to connect through your proxy.
... containers: - name: bluematador-agent image: bluematador/agent:2.0.3 imagePullPolicy: Always env: - name: HTTP_PROXY value: - name: HTTPS_PROXY Value: ...
If you are installing the Kubernetes agent in Google Cloud Platform, your user may not have enough privileges in the Kubernetes cluster to create the necessary resources to run the Blue Matador agent. You can run the following command to give your user privileges:
kubectl create clusterrolebinding cluster-admin-binding \ --clusterrole cluster-admin \ --user $(gcloud config get-value account)
Does the agent need to run on every master and worker node? Configuring the DaemonSet to run on every node is preferred because each agent only reports the pods running on its node to Blue Matador. This allows us to distribute resource usage in larger clusters that may have thousands of pods running, but it also means that any node that does not have the agent will not have its pods monitored. You can modify the tolerations portion of the DaemonSet configuration to allow our agent to run on every node.
Should I install the Kubernetes agent if I am using the AWS integration with EC2 Instances? While the AWS integration collects many EC2 metrics, the Kubernetes agent provides value on top of that. The Kubernetes agent is able to collect data from the Kubernetes API that is critical to monitoring the health of your Kubernetes cluster. This data is not available from the AWS integration alone.
Should I install the Linux agent directly onto a Kubernetes node? The preferred method of monitoring Kubernetes is to install the Blue Matador agent as a DaemonSet in your Kubernetes cluster using the instructions above. Installing the Linux agent directly on a node will not provide any Kubernetes events, but will still provide the Linux server events. Using both installations simultaneously can result in duplicate events, and many of the Linux events are less relevant to kubernetes workloads that are more dynamic in nature.
Where is the Dockerfile for the agent’s Docker image? You can view the Dockerfile here.
Linux Agent Install (Documentation)
Chef Agent Install (Documentation)
Windows Agent Install (Documentation)