Production Deployment Overview


This article walks you through architectural best practices for deploying Observability Pipelines Worker (OPW) into production environments. It goes through guidelines for common deployment mistakes, the different Observability Pipelines Worker deployment roles, and an overview of the three reference architectures.


Start with one aggregator

Most users of Observability Pipelines Worker have complex production environments with many clusters, services, and observability tools. A common mistake is to plan multiple Observability Pipelines Worker deployments instead of starting with one. Datadog recommends to start by deploying a single Observability Pipelines Worker aggregator and onboarding one service at a time. This allows you to get set up quickly and gain experience operating Observability Pipelines Worker before adopting more complex deployment strategies.

Deployment roles

The Observability Pipelines Worker can be deployed anywhere in your infrastructure as an agent or an aggregator.

Agent role

As an agent, the Observability Pipelines Worker (OPW) is deployed on each individual node, shifting data collection and processing to the edge:

A diagram showing an Observability Pipelines Worker in each of the two nodes, where the Workers are both receiving data from the central control plane and sending data to various sinks

Data can be received from other agents on the same node (for example, from the Datadog Agent) or collected directly from the node itself (for example, tailing log files). The goal is to centralize egress on the node through the Observability Pipelines Worker and take advantage of underutilized edge resources. Most users can deploy Observability Pipelines Worker as an agent without increasing their provisioned resources.

Aggregator role

As an aggregator, the Observability Pipelines Worker (OPW) is deployed as a standalone service on dedicated nodes, shifting data processing and routing to the dedicated nodes:

A diagram showing an Observability Pipelines Worker aggregator containing multiple Workers that are receiving data from several network load balancers and sending data to different sinks

Agents are configured to forward data to your Observability Pipelines Worker aggregators. Datadog recommends this role for users looking to get started with Observability Pipelines Worker because it is easy to set up and integrate into any infrastructure.

Reference architectures

There are three common reference architectures that demonstrate how the two deployment roles can work in your infrastructure:

  • Aggregator architecture (recommended): Datadog recommends this architecture for new users of Observability Pipelines Worker. It deploys Observability Pipelines Worker on its own dedicated nodes for easy setup and centralized processing.
  • Agent architecture: This architecture deploys Observability Pipelines Worker along the edge as an agent for local data processing, taking advantage of underutilized edge resources.
  • Unified architecture: This architecture combines the agent and aggregator architectures. Datadog recommends evolving towards this architecture after starting with the aggregator architecture.