Installation on AWS EKS Fargate

New Relic supports monitoring Kubernetes workloads on EKS Fargate by automatically injecting a sidecar containing the infrastructure agent and the nri-kubernetes integration in each pod that needs to be monitored.

If the same Kubernetes cluster also contains EC2 nodes, our solution will also be deployed as a DaemonSet in all of them. No sidecar will be injected into pods scheduled in EC2 nodes, and no DaemonSet will be deployed to Fargate nodes. Here's an example of a hybrid instance with both Fargate and EC2 nodes:

New Relic collects all the supported metrics for all Kubernetes objects regardless of where they are scheduled, whether it's Fargate or EC2 nodes. Please note that, due to the limitations imposed by Fargate, the New Relic integration is limited to running in unprivileged mode on Fargate nodes. This means that metrics that are usually fetched from the host directly, like running processes, will not be available for Fargate nodes.

The agent in both scenarios will scrape data from Kube State Metrics (KSM), Kubelet, and cAdvisor and send data in the same format.

Other components of the New Relic solution for Kubernetes, such as nri-prometheus, nri-metadata-injection, and nri-kube-events, do not have any particularities and will be deployed by our Helm Chart normally as they would in non-Fargate environments.

You can choose between two alternatives for installing New Relic full observability in your EKS Fargate cluster:

• Automatic injection (recommended)
• Manual injection

Regardless of the approach you choose, the experience is exactly the same after it's installed. The only difference is how the container is injected. We do recommend setting up automatic injection with the New Relic infrastructure monitoring operator because it will eliminate the need to manually edit each deployment you want to monitor.

Automatic injection (recommended)

By default, when Fargate support is enabled, New Relic will deploy an operator to the cluster (newrelic-infra-operator). Once deployed, this operator will automatically inject the monitoring sidecar to pods that are scheduled into Fargate nodes, while also managing the creation and the update of Secrets, ClusterRoleBindings, and any other related resources.

This operator accepts a variety of advanced configuration options that can be used to narrow or widen the scope of the injection, through the use of label selectors for both pods and namespaces.

What the operator does

Behind the scenes, the operator sets up a MutatingWebhookConfiguration, which allows it to modify the pod objects that are about to be created in the cluster. On this event, and when the pod being created matches the user’s configuration, the operator will:

1. Add a sidecar container to the pod containing the New Relic Kubernetes integration.
2. If a secret doesn't exist, create one in the same namespace as the pod containing the New Relic license key, which is needed for the sidecar to report data.
3. Add the pod's service account to a ClusterRoleBinding previously created by the operator chart, which will grant this sidecar the required permissions to hit the Kubernetes metrics endpoints.

The ClusterRoleBinding grants the following permissions to the pod being injected:

- apiGroups: [""]
  resources:
    - "nodes"
    - "nodes/metrics"
    - "nodes/stats"
    - "nodes/proxy"
    - "pods"
    - "services"
    - "namespaces"
  verbs: ["get", "list"]
- nonResourceURLs: ["/metrics"]
  verbs: ["get"]

Here's the injection workflow:

$
helm repo add newrelic https://helm-charts.newrelic.com

## Global values

global:
# -- The cluster name for the Kubernetes cluster.
cluster: "_YOUR_K8S_CLUSTER_NAME_"

# -- The license key for your New Relic Account. This will be preferred configuration option if both `licenseKey` and `customSecret` are specified.
licenseKey: "_YOUR_NEW_RELIC_LICENSE_KEY_"

# -- (bool) In each integration it has different behavior. Enables operating system metric collection on each EC2 K8s node. Not applicable to Fargate nodes.
# @default -- false
privileged: true

# -- (bool) Must be set to `true` when deploying in an EKS Fargate environment
# @default -- false
fargate: true

## Enable nri-bundle sub-charts

newrelic-infra-operator:
# Deploys the infrastructure operator, which injects the monitoring sidecar into Fargate pods
enabled: true
tolerations: 
    - key: "eks.amazonaws.com/compute-type"
    operator: "Equal"
    value: "fargate"
    effect: "NoSchedule"
config:
    ignoreMutationErrors: true
    infraAgentInjection:
    # Injection policies can be defined here.  See [values file](https://github.com/newrelic/newrelic-infra-operator/blob/main/charts/newrelic-infra-operator/values.yaml#L114-L125) for more detail.
    policies:
    - namespaceName: namespace-a
    - namespaceName: namespace-b

newrelic-infrastructure:
# Deploys the Infrastructure Daemonset to EC2 nodes.  Disable for Fargate-only clusters.
enabled: true

nri-metadata-injection:
# Deploy our mutating admission webhook to link APM and Kubernetes entities
enabled: true

kube-state-metrics:
# Deploys Kube State Metrics.  Disable if you are already running KSM in your cluster.
enabled: true

nri-kube-events:
# Deploy the Kubernetes events integration.
enabled: true

newrelic-logging:
# Deploys the New Relic's Fluent Bit daemonset to EC2 nodes.  Disable for Fargate-only clusters.
enabled: true

newrelic-prometheus-agent:
# Deploys the Prometheus agent for scraping Prometheus endpoints.
enabled: true
config:
    kubernetes:
    integrations_filter:
        enabled: true
        source_labels: ["app.kubernetes.io/name", "app.newrelic.io/name", "k8s-app"]
        app_values: ["redis", "traefik", "calico", "nginx", "coredns", "kube-dns", "etcd", "cockroachdb", "velero", "harbor", "argocd", "istio"]

$
helm upgrade --install newrelic-bundle newrelic/nri-bundle -n newrelic --create-namespace -f values.yaml

Automatic injection: Known limitations

Here are some issues to be aware of when using automatic injection:

1. Currently there is no controller that watches the whole cluster to make sure that secrets that are no longer needed are garbage collected. However, all objects share the same label that you can use to remove all resources, if needed. We inject the label newrelic/infra-operator-created: true, which you can use to delete resources with a single command.

2. At the moment, it's not possible to use the injected sidecar to monitor services running in the pod. The sidecar will only monitor Kubernetes itself. However, advanced users might want to exclude these pods from automatic injection and manually inject a customized version of the sidecar with on-host integrations enabled by configuring them and mounting their configurations in the proper place. For help, see this tutorial.

Automatic injection: Configuration

You can configure different aspects of the automatic injection. By default, the operator will inject the monitoring sidecar to all pods deployed in Fargate nodes which are not part of a Job or a BatchJob.

This behavior can be changed through configuration options. For example, you can define selectors to narrow or widen the selection of pods that are injected, assign resources to the operator, and tune the sidecar. Also, you can add other attributes, labels, and environment variables. Please refer to the chart README.md and values.yaml.

Update to the latest version or to a new configuration

To update to the latest version of the EKS Fargate integration, upgrade the Helm repository using helm repo update newrelic and reinstall the bundle by simply running again the command above.

To update the configuration of the infrastructure agent injected or the operator itself, modify the values-newrelic.yaml and upgrade the Helm release with the new configuration. The operator is updated immediately, and your workloads will be instrumented with the new version on their next restart. If you wish to upgrade them immediately, you can force a restart of your workloads by running:

$
kubectl rollout restart deployment YOUR_APP

Automatic injection: Uninstall

In order to uninstall the sidecar performing the automatic injection but keep the rest of the New Relic solution, using Helm, disable the infra-operator by setting infra-operator.enabled to false, either in the values.yaml file or in the command line (--set), and re-run the installation command above.

We strongly recommend keeping the --set global.fargate=true flag, since it does not enable automatic injection but makes other components of the installation Fargate-aware, preventing unwanted behavior.

To uninstall the whole solution:

1. Completely uninstall the Helm release.

2. Rollout the pods in order to remove the sidecar:

   bash

   Copy

   $
   kubectl rollout restart deployment YOUR_APP

3. Garbage collect the secrets:

   bash

   Copy

   $
   kubectl delete secrets -n YOUR_NAMESPACE -l newrelic/infra-operator-created=true

Manual injection

If you've any concerns about the automatic injection, you can inject the sidecar manually directly by modifying the manifests of the workloads scheduled that are going to be scheduled on Fargate nodes. Please note that adding the sidecar into deployments scheduled into EC2 nodes may lead into incorrect or duplicate data, especially if those nodes are already being monitored with the DaemonSet.

These objects are required for the sidecar to successfully report data:

• The ClusterRole providing the permission needed by the nri-kubernetes integration.
• A ClusterRoleBinding linking the ClusterRole and the service account of the pod.
• The secret storing the New Relic licenseKey in each Fargate namespace.
• The sidecar container in the specific template of the monitored workload.

Complete these steps for manual injection:

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
    labels:
    app: newrelic-infrastructure
    name: newrelic-newrelic-infrastructure-infra-agent
rules:
    - apiGroups:
        - ""
    resources:
        - nodes
        - nodes/metrics
        - nodes/stats
        - nodes/proxy
        - pods
        - services
    verbs:
        - get
        - list
    - nonResourceURLs:
        - /metrics
    verbs:
        - get

apiVersion: apps/v1
kind: Deployment
spec:
    template:
    spec:
        containers:
        - name: newrelic-infrastructure
        env:
        - name: NRIA_LICENSE_KEY
            valueFrom:
            secretKeyRef:
                key: license
                name: newrelic-newrelic-infrastructure-config
        - name: NRIA_VERBOSE
            value: "1"
        - name: DISABLE_KUBE_STATE_METRICS
            value: "true"
        - name: CLUSTER_NAME
            value: testing-injection
        - name: COMPUTE_TYPE
            value: serverless
        - name: NRK8S_NODE_NAME
            valueFrom:
            fieldRef:
                apiVersion: v1
                fieldPath: spec.nodeName
        - name: NRIA_DISPLAY_NAME
            valueFrom:
            fieldRef:
                apiVersion: v1
                fieldPath: spec.nodeName
        - name: NRIA_CUSTOM_ATTRIBUTES
            value: '{"clusterName":"$(CLUSTER_NAME)", "computeType":"$(COMPUTE_TYPE)", "fargateProfile":"[YOUR FARGATE PROFILE]"}'
        - name: NRIA_PASSTHROUGH_ENVIRONMENT
            value: KUBERNETES_SERVICE_HOST,KUBERNETES_SERVICE_PORT,CLUSTER_NAME,CADVISOR_PORT,NRK8S_NODE_NAME,KUBE_STATE_METRICS_URL,KUBE_STATE_METRICS_POD_LABEL,TIMEOUT,ETCD_TLS_SECRET_NAME,ETCD_TLS_SECRET_NAMESPACE,API_SERVER_SECURE_PORT,KUBE_STATE_METRICS_SCHEME,KUBE_STATE_METRICS_PORT,SCHEDULER_ENDPOINT_URL,ETCD_ENDPOINT_URL,CONTROLLER_MANAGER_ENDPOINT_URL,API_SERVER_ENDPOINT_URL,DISABLE_KUBE_STATE_METRICS,DISCOVERY_CACHE_TTL
        image: newrelic/infrastructure-k8s:2.4.0-unprivileged
        imagePullPolicy: IfNotPresent
        resources:
            limits:
            memory: 100M
            cpu: 200m
            requests:
            cpu: 100m
            memory: 50M
        securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            runAsUser: 1000
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /var/db/newrelic-infra/data
            name: tmpfs-data
        - mountPath: /var/db/newrelic-infra/user_data
            name: tmpfs-user-data
        - mountPath: /tmp
            name: tmpfs-tmp
        - mountPath: /var/cache/nr-kubernetes
            name: tmpfs-cache
[...]

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
    name: newrelic-newrelic-infrastructure-infra-agent
roleRef:
    apiGroup: rbac.authorization.k8s.io
    kind: ClusterRole
    name: newrelic-newrelic-infrastructure-infra-agent
subjects:
    - kind: ServiceAccount
    name: [INSERT_SERVICE_ACCOUNT_NAME_OF_WORKLOAD]
    namespace: [INSERT_SERVICE_ACCOUNT_NAMESPACE_OF_WORKLOAD]

apiVersion: v1
data:
    license: INSERT_YOUR_NEW_RELIC_LICENSE_ENCODED_IN_BASE64
kind: Secret
metadata:
    name: newrelic-newrelic-infrastructure-config
    namespace: [INSERT_NAMESPACE_OF_WORKLOAD]
type: Opaque

Manual injection: Update to the latest version

To update any of the components, you just need to modify the deployed yaml. Updating any of the fields of the injected container will cause the pod to be re-created.

Manual injection: Uninstall the Fargate integration

To remove the injected container and the related resources, you just have to remove the following:

• The sidecar from the workloads that should be no longer monitored.
• All the secrets containing the newrelic license.
• ClusterRole and ClusterRoleBinding objects.

Notice that removing the sidecar container will cause the pod to be re-created.

Logging

New Relic logging isn't available on Fargate nodes because of security constraints imposed by AWS, but here are some logging options:

• If you're using Fluentbit for logging, see Kubernetes plugin for log forwarding.
• If your log data is already being monitored by AWS FireLens, see AWS FireLens plugin for log forwarding.
• If your log data is already being monitored by Amazon CloudWatch Logs, see Stream logs using Kinesis Data Firehose.
• See AWS Lambda for sending CloudWatch logs.
• See Three ways to forward logs from Amazon ECS to New Relic.

Troubleshooting

LAST SEEN | TYPE | REASON | OBJECT | MESSAGE
:--|:--|:--|:--|:--
3m9s (x2 over 8m10s) | Warning | FailedScheduling | Pod/no-fargate-deploy-cbddd6ccf-8f9x4 | 0/2 nodes are available: 2 node(s) had untolerated taint {eks.amazonaws.com/compute-type: fargate}. preemption: 0/2 nodes are available: 2 Preemption is not helpful for scheduling..

LAST SEEN | TYPE | REASON | OBJECT | MESSAGE
:--|:--|:--|:--|:--
61s | Warning | FailedScheduling | Pod/newrelic-bundle-newrelic-infra-operator-admission-create-d8ggt | 0/2 nodes are available: 2 Too many pods. preemption: 0/2 nodes are available: 2 No preemption victims found for incoming pod..

View your EKS data

Here's an example of what a Fargate node looks like in the New Relic UI:

To view your AWS data:

1. Go to one.newrelic.com > All capabilities > Infrastructure > Kubernetes and do one of the following:

   • Select an integration name to view data.
   • Select the Explore data icon to view AWS data.

2. Filter your data using two Fargate tags:

   • computeType=serverless
   • fargateProfile=[name of the Fargate profile to which the workload belongs]