Overview

The OpenTelemetry Collector can be deployed in multiple ways. The daemonset pattern is a common deployment where a Collector instance runs on every Kubernetes node alongside the core Datadog Agent.

The gateway pattern provides an additional deployment option that uses a centralized, standalone Collector service. This gateway layer can perform actions such as tail-based sampling, aggregation, filtering, and routing before exporting the data to one or more backends such as Datadog. It acts as a central point for managing and enforcing observability policies.

When you enable the gateway:

1. A Kubernetes Deployment (<RELEASE_NAME>-datadog-otel-agent-gateway-deployment) manages the standalone gateway Collector pods.
2. A Kubernetes Service (<RELEASE_NAME>-datadog-otel-agent-gateway) exposes the gateway pods and provides load balancing.
3. The existing DaemonSet Collector pods are configured by default to send their telemetry data to the gateway service instead of directly to Datadog.

Requirements

Before you begin, ensure you have the following:

• Datadog Account:
  • A Datadog account.
  • Your Datadog API key.
• Software:
  • A Kubernetes cluster (v1.29+). EKS Fargate and GKE Autopilot are not supported.
  • Helm (v3+).
  • Datadog Helm chart version 3.137.1 or higher.
  • kubectl.

Installation and configuration

This guide uses the Datadog Helm chart to configure the DDOT Collector gateway.

Deploying the gateway with a daemonset

To get started, enable both the gateway and the DaemonSet Collector in your values.yaml file. This is the most common setup.

# values.yaml
targetSystem: "linux"
datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  # Enable the Collector in the Agent Daemonset
  otelCollector:
    enabled: true

# Enable the standalone Gateway Deployment
otelAgentGateway:
  enabled: true
  replicas: 3
  nodeSelector:
    # Example selector to place gateway pods on specific nodes
    gateway: "true"

In this case, the daemonset Collector uses a default config that sends OTLP data to the gateway’s Kubernetes service:

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317
      http:
        endpoint: 0.0.0.0:4318
exporters:
  debug:
    verbosity: detailed
  otlphttp:
    endpoint: http://<release>-datadog-otel-agent-gateway:4318
    tls:
      insecure: true
processors:
  batch:
    timeout: 10s
connectors:
  datadog/connector:
    traces:
      compute_top_level_by_span_kind: true
      peer_tags_aggregation: true
      compute_stats_by_span_kind: true
service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlphttp, datadog/connector]
    metrics:
      receivers: [otlp, datadog/connector]
      processors: [batch]
      exporters: [otlphttp]
    logs:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlphttp]

The gateway Collector uses a default config that listens on the service ports and sends data to Datadog:

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317
      http:
        endpoint: 0.0.0.0:4318
exporters:
  debug:
    verbosity: detailed
  datadog:
    api:
      key: ${env:DD_API_KEY}
processors:
  batch:
    timeout: 10s
service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [datadog]
    metrics:
      receivers: [otlp]
      processors: [batch]
      exporters: [datadog]
    logs:
      receivers: [otlp]
      processors: [batch]
      exporters: [datadog]

Deploying a standalone gateway

If you have an existing DaemonSet deployment, you can deploy the gateway independently.

# values.yaml
targetSystem: "linux"
fullnameOverride: "gw-only"
agents:
  enabled: false
clusterAgent:
  enabled: false
datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
otelAgentGateway:
  enabled: true
  replicas: 3
  nodeSelector:
    gateway: "true"

After deploying the gateway, you must update the configuration of your existing DaemonSet Collectors to send data to the new gateway service endpoint (for example, http://gw-only-otel-agent-gateway:4318).

Customizing Collector configurations

You can override the default configurations for both the DaemonSet and gateway Collectors using the datadog.otelCollector.config and otelAgentGateway.config values, respectively.

# values.yaml
targetSystem: "linux"
fullnameOverride: "my-gw"
datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY>
  # Enable and configure the DaemonSet Collector
  otelCollector:
    enabled: true
    config: |
      receivers:
        otlp:
          protocols:
            grpc:
              endpoint: "localhost:4317"
      exporters:
        otlp:
          endpoint: http://my-gw-otel-agent-gateway:4317
          tls:
            insecure: true
      service:
        pipelines:
          traces:
            receivers: [otlp]
            exporters: [otlp]
          metrics:
            receivers: [otlp]
            exporters: [otlp]
          logs:
            receivers: [otlp]
            exporters: [otlp]

# Enable and configure the gateway Collector
otelAgentGateway:
  enabled: true
  replicas: 3
  nodeSelector:
    gateway: "true"
  ports:
    - containerPort: 4317
      name: "otel-grpc"
  config: |
    receivers:
      otlp:
        protocols:
          grpc:
            endpoint: "0.0.0.0:4317"
    exporters:
      datadog:
        api:
          key: ${env:DD_API_KEY}
    service:
      pipelines:
        traces:
          receivers: [otlp]
          exporters: [datadog]
        metrics:
          receivers: [otlp]
          exporters: [datadog]
        logs:
          receivers: [otlp]
          exporters: [datadog]

Advanced use cases

Tail sampling with the load balancing exporter

A primary use case for the gateway is tail-based sampling. To ensure that all spans for a given trace are processed by the same gateway pod, use the load balancing exporter in your DaemonSet Collectors. This exporter consistently routes spans based on a key, such as traceID.

In the values.yaml below:

1. The daemonset Collector (datadog.otelCollector) is configured with the loadbalancing exporter, which uses the Kubernetes service resolver to discover and route data to the gateway pods.
2. The gateway Collector (otelAgentGateway) uses the tail_sampling processor to sample traces based on defined policies before exporting them to Datadog.

# values.yaml
targetSystem: "linux"
fullnameOverride: "my-gw"
datadog:
  apiKey: <DATADOG_API_KEY>
  appKey: <DATADOG_APP_KEY> 
  otelCollector:
    enabled: true
    # RBAC permissions are required for the k8s resolver in the loadbalancing exporter
    rbac:
      create: true
      rules:
        - apiGroups: [""]
          resources: ["endpoints"]
          verbs: ["get", "watch", "list"]
    config: |
      receivers:
        otlp:
          protocols:
            grpc:
              endpoint: "localhost:4317"
      exporters:
        loadbalancing:
          routing_key: "traceID"
          protocol:
            otlp:
              tls:
                insecure: true
          resolver:
            k8s:
              service: my-gw-otel-agent-gateway
              ports:
                - 4317
      service:
        pipelines:
          traces:
            receivers: [otlp]
            exporters: [loadbalancing]

otelAgentGateway:
  enabled: true
  replicas: 3
  ports:
    - containerPort: 4317
      name: "otel-grpc"
  config: |
    receivers:
      otlp:
        protocols:
          grpc:
            endpoint: "0.0.0.0:4317"
    processors:
      tail_sampling:
        decision_wait: 10s
        policies: <Add your sampling policies here>
    connectors:
      datadog/connector:
    exporters:
      datadog:
        api:
          key: ${env:DD_API_KEY}
    service:
      pipelines:
        traces/sample:
          receivers: [otlp]
          processors: [tail_sampling]
          exporters: [datadog]
        traces:
          receivers: [otlp]
          exporters: [datadog/connector]
        metrics:
          receivers: [datadog/connector]
          exporters: [datadog]

Using a custom Collector image

To use a custom-built Collector image for your gateway, specify the image repository and tag under agents.image. This follows the same process as the DaemonSet deployment. For more details, see Use Custom OpenTelemetry Components.

# values.yaml
targetSystem: "linux"
agents:
  enabled: false
  image:
    repository: <YOUR REPO>
    tag: <IMAGE TAG>
    doNotCheckTag: true
clusterAgent:
  enabled: false
otelAgentGateway:
  enabled: true
  ports:
    - containerPort: "4317"
      name: "otel-grpc"
  config: | <YOUR CONFIG>

Deploying a multi-layer gateway

For advanced scenarios, you can deploy multiple gateway layers to create a processing chain. To do this, deploy each layer as a separate Helm release, starting from the final layer and working backward.

1. Deploy Layer 1 (Final Layer): This layer receives from Layer 2 and exports to Datadog.

   # layer-1-values.yaml
   targetSystem: "linux"
   fullnameOverride: "gw-layer-1"
   agents:
     enabled: false
   clusterAgent:
     enabled: false
   otelAgentGateway:
     enabled: true
     replicas: 3
     nodeSelector:
       gateway: "gw-node-1"
     ports:
       - containerPort: "4317"
         hostPort: "4317"
         name: "otel-grpc"
     config: |
       receivers:
         otlp:
           protocols:
             grpc:
               endpoint: "0.0.0.0:4317"
       exporters:
         datadog:
           api:
             key: <API Key>
       service:
         pipelines:
           traces:
             receivers: [otlp]
             exporters: [datadog]
           metrics:
             receivers: [otlp]
             exporters: [datadog]
           logs:
             receivers: [otlp]
             exporters: [datadog]
   

2. Deploy Layer 2 (Intermediate Layer): This layer receives from the DaemonSet and exports to Layer 1.

   # layer-2-values.yaml
   targetSystem: "linux"
   fullnameOverride: "gw-layer-2"
   agents:
     enabled: false
   clusterAgent:
     enabled: false
   otelAgentGateway:
     enabled: true
     replicas: 3
     nodeSelector:
       gateway: "gw-node-2"
     ports:
       - containerPort: "4317"
         hostPort: "4317"
         name: "otel-grpc"
     config: |
       receivers:
         otlp:
           protocols:
             grpc:
               endpoint: "0.0.0.0:4317"
       exporters:
         otlp:
           endpoint: http://gw-layer-1-otel-agent-gateway:4317
           tls:
             insecure: true
       service:
         pipelines:
           traces:
             receivers: [otlp]
             exporters: [otlp]
           metrics:
             receivers: [otlp]
             exporters: [otlp]
           logs:
             receivers: [otlp]
             exporters: [otlp]
   

3. Deploy DaemonSet: Configure the DaemonSet to export to Layer 2.

   # daemonset-values.yaml
   targetSystem: "linux"
   datadog:
     apiKey: <DATADOG_API_KEY>
     appKey: <DATADOG_APP_KEY>
     otelCollector:
       enabled: true
       config: |
         receivers:
           otlp:
             protocols:
               grpc:
                 endpoint: "localhost:4317"
         exporters:
           otlp:
             endpoint: http://gw-layer-2-otel-agent-gateway:4317
             tls:
               insecure: true
         service:
           pipelines:
             traces:
               receivers: [otlp]
               exporters: [otlp]
             metrics:
               receivers: [otlp]
               exporters: [otlp]
             logs:
               receivers: [otlp]
               exporters: [otlp]
   

Known limitations

• Gateway pods on Fleet Automation: Standalone gateway pods are not yet visible on the Fleet Automation page. Only DaemonSet Collectors are displayed. This is being actively addressed.
• Startup race condition: When deploying the DaemonSet and gateway in the same release, DaemonSet pods might start before the gateway service is ready, causing initial connection error logs. The OTLP exporter automatically retries, so these logs can be safely ignored. Alternatively, deploy the gateway first and wait for it to become ready before deploying the DaemonSet.
• infraattributes processor requirement: The infraattributes processor requires a datadog exporter to be defined in the same Collector configuration, even if it’s not used in a pipeline. The Collector will fail to start if the exporter is missing. To resolve this, add a datadog exporter to your configuration, even if you do not reference it in a service pipeline.
• Ignorable Core Agent Connection Logs: Gateway pods might generate warning logs about failing to connect to a core Datadog Agent (for example, grpc: addrConn.createTransport failed to connect). This occurs because the gateway deployment does not include a core agent in the same pod. These logs are expected and can be safely ignored. This is being actively addressed.

Further reading

Documentation, liens et articles supplémentaires utiles:

Use Custom OpenTelemetry Components with Datadog AgentDOCUMENTATION Collector Deployment: GatewayOPENTELEMETRY Load Balancing ExporterOPENTELEMETRY Tail Sampling ProcessorOPENTELEMETRY