Traffic Director features

Traffic Director helps you run microservices in a global service mesh. The mesh handles networking for your microservices so that you can write application code that doesn't need to know about underlying networking complexities. This separation of application logic from networking logic allows you to improve your development velocity, increase service availability, and introduce modern DevOps practices to your organization.

Your service mesh consists of your applications, an xDS-compatible data plane (generally the open source Envoy proxy) and Traffic Director as your mesh control plane.

You can also deploy proxyless gRPC services with Traffic Director in your service mesh using a supported version of gRPC.

This document summarizes the features available in Traffic Director. The value N/A (Not Applicable) means that a feature cannot be supported because it is not compatible with the particular Traffic Director configuration.

Fully managed control plane for service mesh

Traffic Director is a managed, highly available control plane service that runs in Google Cloud. You don't need to install or update your control plane, so you have one less component to manage in your service mesh infrastructure.

Supported xDS versions

Traffic Director uses open source xDS control plane APIs to configure Envoy and proxyless gRPC clients. These clients act on behalf of your application code to deliver Traffic Director's application networking capabilities.

Version Support
xDS v2 General Availability
xDS v3 Preview

Platforms to run mesh services

You can run applications on the following platforms and adopt them into a global service mesh that is configured by Traffic Director.

Feature Supported
Compute Engine virtual machines (VMs)
Google Kubernetes Engine container instances
Kubernetes on Compute Engine container instances

Service management

Services in a mesh configured by Traffic Director benefit from service discovery, backend autoscaling, and endpoint auto-registration:

  • When an application in your mesh wants to reach another application, it can call on that service by name. This is referred to as service discovery.

  • These services are backed by instances that run your application code. These instances scale up or down dynamically based on your needs.

  • As new instances are created or removed, they need to be associated with your service. This is referred to as endpoint registration.

Feature Supported
Automated deployment of sidecar proxies for Compute Engine VMs
Automated injection of sidecar proxies for Google Kubernetes Engine Pods
Service discovery based on hostname
Instance autoscaling based on CPU utilization
Instance autoscaling based on traffic load/serving capacity (Compute Engine VMs in MIGs only)
Instance autohealing based on configurable health checks
Automatic endpoint registration for Compute Engine VM instances
Automatic endpoint registration for GKE container instances/pods
API to programmatically add or remove endpoints

Endpoints for your data plane traffic

Microservices use the data plane to reach services in your mesh, as well as outside of your mesh. Traffic Director enables you to separate application logic from networking logic so all your application needs to do is send requests to the data plane (for example, the sidecar proxy running alongside the application). The data plane takes care of sending requests to the right endpoint.

In the table below, applications described as being in the mesh are those applications that communicate with other services using the Traffic Director-managed data plane. Those applications can send traffic to in-mesh services, as well as services outside of the mesh.

Feature Supported
VM-based applications in the mesh
Container-based applications in the mesh
VM-based applications outside of the mesh
Container-based applications outside of the mesh
Applications running in on-premises data centers
Applications in multi-cloud environments

Data plane topologies

In the service mesh model, your applications communicate using a data plane. This data plane often consists of sidecar proxies deployed alongside your applications. Traffic Director is highly flexible and supports data plane topologies that fit your service networking needs.

Feature Supported
Sidecar proxies running alongside applications
Proxyless gRPC applications
Middle proxies between two applications in a mesh
Edge proxies at the boundary of your mesh
Mesh spanning multiple GKE clusters and/or Compute Engine VMs in multiple regions

Programmatic, API-driven configuration

All configuration is exposed through our REST API and dashboard out-of-the-box, allowing you to automate changes across large teams and manage changes programmatically.

Feature Supported
Google Cloud Console
gcloud command-line interface
Deployment Manager
Terraform support

Language support with proxyless gRPC applications

You can create proxyless gRPC applications that work with Traffic Director using the following programming languages.

Language Supported

Request protocols

Applications can use the following request protocols when they communicate using the Traffic Director-configured data plane.

Feature Supported

Routing and traffic management

Traffic Director supports advanced traffic management policies that you can use to steer, split, and shape traffic as it passes through your data plane. Note that most advanced traffic management is not enabled for Traffic Director with proxyless gRPC services, and none of the advanced traffic management features are available with the target TCP proxy resource.

Feature Supported with Envoy proxy configured to handle HTTP or gRPC traffic Supported with Envoy proxy configured to handle TCP traffic Supported with proxyless gRPC
HTTP/Layer 7 request routing based on suffix/prefix/full/regex match on:
• Host name N/A
• Path N/A
• Headers N/A
• Method N/A N/A
• Cookies N/A
• Request parameters N/A N/A
Fault injection N/A
Configurable timeouts N/A
Retries N/A
Redirects N/A
URI rewrites N/A
Request/response header transformations N/A
Traffic splitting
Traffic mirroring
Outlier detection

Load balancing

You can configure advanced load balancing methods and algorithms to load balance at the service, backend group (instance groups or network endpoint groups), and individual backend or endpoint levels. For more information, see Backend services overview.

Feature Supported with Envoy proxy configured to handle HTTP or gRPC traffic Supported with Envoy proxy configured to handle TCP traffic Supported with proxyless gRPC
Service selection based on weight-based traffic splits
Backend (instance group or network endpoint group) selection based on region (prefer nearest region with healthy backend capacity)
Backend selection using rate-based (requests per second) balancing mode N/A
Backend selection based on utilization-based balancing mode (VMs in Compute Engine instance groups only)
Configurable maximum capacity per backend (Compute Engine and GKE only)
Circuit breaking
Backend selection based on configurable load balancing policies*:
  • Round robin
  • Least request
  • Ring hash
  • Random
  • Original destination
  • Maglev
Round robin only

*See localityLbPolicy for additional details.

Service and backend capacity management

Traffic Director takes service and backend capacity into account to ensure optimal distribution of traffic across your services' backends. Traffic Director is integrated with Google Cloud infrastructure so that it automatically collects capacity data. You can also set and configure capacity manually.

Feature Supported
Automatically tracks backend capacity and utilization, based on CPU, for VM instances in a Managed Instance Group
Manual capacity and overrides for VM and container instances in MIGs and NEGs based on request rate
Manual capacity draining


Enterprise workloads generally rely on high-availability deployments to ensure service uptime. Traffic Director supports these types of deployments by enabling multi-zone/multi-region redundancy.

Feature Supported
Automatic failover to another zone within the same region that has healthy backend capacity
Automatic failover to nearest region with healthy backend capacity

Health checks

Centralized health checking to determine backend health. For reference information, see Health checks overview.

Feature Supported
gRPC health checks
HTTP health checks
HTTPS health checks
HTTP/2 health checks
TCP health checks
Configurable health checks:
  • Port
  • Check intervals
  • Timeouts
  • Healthy and unhealthy thresholds
Configurable request path (HTTP, HTTPS, HTTP/2)
Configurable request string or path (TCP or SSL)
Configurable expected response string


Observability tools provide monitoring, debugging, and performance information to help you understand your service mesh. The following capabilities are either provided out-of-the-box or configured in your data plane. Your application code doesn't need to do anything special to generate this observability data.

The service health dashboard is available with proxyless gRPC services, but you cannot configure data plane logging and tracing. Traffic Director cannot configure a gRPC application's logging and tracing. You can enable this by following the instructions in the troubleshooting sections or gRPC guides available on open source sites. For example, you can use Opencensus to enable metrics collection and tracing in your proxyless gRPC services.

Feature Supported with proxies Supported with proxyless gRPC services
Service health dashboard
Data plane logging
Data plane tracing

Session affinity

Client-server communications often involve multiple successive requests. In such a case, it's helpful to route successive client requests to the same backend or server. Traffic Director provides configurable options to send requests from a particular client, on a best effort basis, to the same backend as long as the backend is healthy and has capacity. For more information, see Backend services overview

Feature Supported with HTTP(S) proxies Supported with TCP proxies Supported with proxyless gRPC services
Client IP address
HTTP cookie N/A
HTTP header N/A
Generated cookie (sets client cookie on first request) N/A

Network topologies

Traffic Director supports common Google Cloud network topologies.

Feature Supported
Single network in a Google Cloud project
Shared VPC (single network shared across multiple Google Cloud projects)

See Limitations for a detailed explanation of how Shared VPC is supported with Traffic Director.


Traffic Director is compliant with the following standards.

Compliance certification
ISO 27001, ISO 27017, ISO 27018