Link OpenTelemetry-instrumented applications to Kubernetes

If you’re a developer running apps in Kubernetes, you can use New Relic to understand how Kubernetes infrastructure affects your OpenTelemetry-instrumented applications.

After completing the steps below, you can use the New Relic UI to correlate application-level metrics from OpenTelemetry with Kubernetes infrastructure metrics. This will help you view the entire landscape of your telemetry data and work across teams to get faster mean time to resolution (MTTR) for issues in your Kubernetes environment.

Overview: How we correlate the data

The steps in this guide enable your application to inject infrastructure-specific metadata into the telemetry data. The result is that the New Relic UI is populated with actionable information. Here are the steps you'll take to get this going:

In each application container, define an environment variable to send telemetry data to the collector
Deploy the OpenTelemetry collector as a DaemonSet in agent mode with resourcedetection, resource, batch and k8sattributes processors to inject relevant metadata (cluster, deployment, and namespace names)

Prerequisites

To be successful with the steps below, you should already be familiar with OpenTelemetry and Kubernetes and have done the following:

Created the following environment variables:
- OTEL_EXPORTER_OTLP_ENDPOINT (New Relic endpoint for your region or purpose)
- NEW_RELIC_API_KEY (ingest license key)
Installed the New Relic Kubernetes integration in your cluster
Instrumented your applications with OpenTelemetry, and successfully sent data to New Relic via OpenTelemetry Protocol (OTLP)

If you have general questions about using collectors with New Relic, see our Introduction to OpenTelemetry Collector with New Relic.

Configure your application to send telemetry data to the OpenTelemetry collector

To set this up, you need to add a custom snippet to the env stanza of your Kubernetes YAML file. We have an example below that shows the snippet for a sample front-end microservice (Frontend.yaml). The snippet includes two sections that do the following:

Section 1: Ensure that the telemetry data is sent to the collector. This sets the environment variable OTEL_EXPORTER_OTLP_ENDPOINT with the host IP. It does this by calling the downward API to pull the host IP.
Section 2: Attach infrastructure-specific metadata. To do this, we capture metadata.uid using the downward API and add it to the OTEL_RESOURCE_ATTRIBUTES environment variable. This environment variable is used by the OpenTelemetry collector’s resourcedetection and k8sattributes processors to add additional infrastructure-specific context to telemetry data.

For each microservice instrumented with OpenTelemetry, add the highlighted lines below to your manifest’s env stanza:

# Frontend.yaml
apiVersion: apps/v1
kind: Deployment
...
spec:
  containers:
  - name: yourfrontendservice
    image: yourfrontendservice-beta
  env:
  # Section 1: Ensure that telemetry data is sent to the collector
  - name: HOST_IP
        valueFrom:
          fieldRef:
            fieldPath: status.hostIP
      # This is picked up by the opentelemetry sdks
      - name: OTEL_EXPORTER_OTLP_ENDPOINT
        value: "http://$(HOST_IP):55680"
  # Section 2: Attach infrastructure-specific metadata
     # Get pod ip so that k8sattributes can tag resources
- name: POD_NAME
valueFrom:
fieldRef:
fieldPath: metadata.name
- name: POD_UID
valueFrom:
fieldRef:
fieldPath: metadata.uid
      # This is picked up by the resource detector
    - name: OTEL_RESOURCE_ATTRIBUTES
      value: service.instance.id=$(POD_NAME),k8s.pod.uid=$(POD_UID)”

Configure and deploy the OpenTelemetry collector as an agent

We recommend you deploy the collector as an agent on every node within a Kubernetes cluster. The agent can receive telemetry data as well as enrich telemetry data with metadata. For example, the collector can add custom attributes or infrastructure information through processors as well as handle batching, retry, compression and other more advanced features that are handled less efficiently at the client instrumentation level.

For help configuring the collector, see the sample collector configuration file below, along with sections about setting up these options:

OTLP exporter
batch processor
resourcedetection processor
k8sattributes processor: general
k8sattributes processor: RBAC
k8sattributes processor: filters

receivers:
  otlp:
    protocols:
      grpc:

processors:
  batch:
  resource:
    attributes:
      - key: host.id
        from_attribute: host.name
        action: upsert
  resourcedetection:
    detectors: [ gke, gce ]
  k8sattributes:
    auth_type: "serviceAccount"
    passthrough: false
    filter:
      node_from_env_var: KUBE_NODE_NAME
    extract:
      metadata:
        - k8s.pod.name
        - k8s.pod.uid
        - k8s.deployment.name
        - k8s.cluster.name
        - k8s.namespace.name
        - k8s.node.name
        - k8s.pod.start_time
    pod_association:
      - from: resource_attribute
        name: k8s.pod.uid

exporters:
  otlp:
    endpoint: $OTEL_EXPORTER_OTLP_ENDPOINT
    headers:
      api-key: $NEW_RELIC_API_KEY
  logging:
    logLevel: DEBUG

service:
  pipelines:
    metrics:
      receivers: [ otlp ]
      processors: [ resourcedetection, k8sattributes, resource, cumulativetodelta, batch ]
      exporters: [ otlp ]
    traces:
      receivers: [ otlp ]
      processors: [ resourcedetection, k8sattributes, resource, batch ]
      exporters: [ otlp ]
    logs:
      receivers: [ otlp ]
      processors: [ resourcedetection, k8sattributes, resource, batch ]
      exporters: [ otlp ]

Step 1: Configure the OTLP exporter

First, configure it by adding an OTLP exporter to your OpenTelemetry collector configuration YAML file along with your New Relic API key (ingest license key) as a header.

exporters:
  otlp:
    endpoint: $OTEL_EXPORTER_OTLP_ENDPOINT
  headers: api-key: $NEW_RELIC_API_KEY

Step 2: Configure the batch processor

The batch processor accepts spans, metrics, or logs and places them into batches to make it easier to compress the data and reduce the number of outgoing requests from the collector.

processors:
  batch:

Step 3: Configure the resource detection processor

The resourcedetection processor gets host-specific information to add additional context to the telemetry data being processed through the collector. In this example, we use Google Kubernetes Engine (GKE) and Google Compute Engine (GCE) to get Google Cloud-specific metadata, including:

cloud.provider ("gcp")
cloud.platform ("gcp_compute_engine")
cloud.account.id
cloud.region
cloud.availability_zone
host.id
host.image.id
host.type

processors:
  resourcedetection:
Detectors: [ gke, gce ]

Step 4: Configure the Kubernetes Attributes processor (general)

When we run the k8sattributes processor as part of the OpenTelemetry collector running as an agent, it detects IP addresses of pods sending telemetry data to the OpenTelemetry collector agent, using them to extract pod metadata. Below is a basic Kubernetes manifest example with only a processors section. To deploy the OpenTelemetry collector as a DaemonSet, read this comprehensive manifest example.

processors:
  k8sattributes:
    auth_type: "serviceAccount"
    passthrough: false
    filter:
      node_from_env_var: KUBE_NODE_NAME
    extract:
      metadata:
        - k8s.pod.name
        - k8s.pod.uid
        - k8s.deployment.name
        - k8s.cluster.name
        - k8s.namespace.name
        - k8s.node.name
        - k8s.pod.start_time
    pod_association:
      - from: resource_attribute
        name: k8s.pod.uid

Step 5: Configure the Kubernetes Attributes processor (RBAC)

You need to add configurations for role based access control (RBAC). The k8sattributes processor needs get, watch and list permissions for pods and namespaces resources included in the configured filters. See this example of how to configure role based access control (RBAC) for ClusterRole to give a ServiceAccount the necessary permissions for all pods and namespaces in the cluster.

Step 6: Configure the Kubernetes Attributes processor (discovery filter)

When running the collector as an agent, you should apply a discovery filter so that the processor only discovers pods from the same host that it is running on. If you don’t use a filter, resource usage can be unnecessarily high, especially on very large clusters. Once the filter is applied, each processor will only query the Kubernetes API for pods running on its own node.

To set the filter, use the downward API to inject the node name as an environment variable in the pod env section of the OpenTelemetry collector agent configuration YAML file (see GitHub for an example). This will inject a new environment variable to the OpenTelemetry collector agent’s container. The value will be the name of the node the pod was scheduled to run on.

spec:
  containers:
  - env:
    - name: KUBE_NODE_NAME
      valueFrom:
        fieldRef:
          apiVersion: v1
          fieldPath: spec.nodeName

Then, you can filter by the node with the k8sattributes:

k8sattributes:
    filter:
      node_from_env_var: KUBE_NODE_NAME

Validate that your configurations are working

If you have successfully linked your OpenTelemetry data with your Kubernetes data, you should be able to see Kubernetes attributes like k8s.cluster.name and k8s.deployment.name in your spans within the distributed tracing UI.

Click to enlarge the image:

What's next?

Now that you've connected your OpenTelemetry-instrumented apps with Kubernetes, check out our best practices guide for tips to improve your use of OpenTelemetry and New Relic.