Architecture

An Istio service mesh is logically split into a data plane and a control plane.

  • The data plane is composed of a set of intelligent proxies (Envoy) deployed as sidecars. These proxies mediate and control all network communication between microservices along with Mixer, a general-purpose policy and telemetry hub.

  • The control plane manages and configures the proxies to route traffic. Additionally, the control plane configures Mixers to enforce policies and collect telemetry.

The following diagram shows the different components that make up each plane:

The overall architecture of an Istio-based application.
Istio Architecture

Traffic in Istio is categorized as data plane traffic and control plane traffic. Data plane traffic refers to the messages that the business logic of the workloads send and receive. Control plane traffic refers to configuration and control messages sent between Istio components to program the behavior of the mesh. Traffic management in Istio refers exclusively to data plane traffic.

Components

The following sections provide a brief overview of each of Istio’s core components.

Envoy

Istio uses an extended version of the Envoy proxy. Envoy is a high-performance proxy developed in C++ to mediate all inbound and outbound traffic for all services in the service mesh. Envoy proxies are the only Istio components that interact with data plane traffic.

Envoy proxies are deployed as sidecars to services, logically augmenting the services with Envoy’s many built-in features, for example:

  • Dynamic service discovery
  • Load balancing
  • TLS termination
  • HTTP/2 and gRPC proxies
  • Circuit breakers
  • Health checks
  • Staged rollouts with %-based traffic split
  • Fault injection
  • Rich metrics

This sidecar deployment allows Istio to extract a wealth of signals about traffic behavior as attributes. Istio can, in turn, use these attributes in Mixer to enforce policy decisions, and send them to monitoring systems to provide information about the behavior of the entire mesh.

The sidecar proxy model also allows you to add Istio capabilities to an existing deployment with no need to rearchitect or rewrite code. You can read more about why we chose this approach in our Design Goals.

Some of the Istio features and tasks enabled by Envoy proxies include:

  • Traffic control features: enforce fine-grained traffic control with rich routing rules for HTTP, gRPC, WebSocket, and TCP traffic.

  • Network resiliency features: setup retries, failovers, circuit breakers, and fault injection.

  • Security and authentication features: enforce security policies and enforce access control and rate limiting defined through the configuration API.

Mixer

Mixer is a platform-independent component. Mixer enforces access control and usage policies across the service mesh, and collects telemetry data from the Envoy proxy and other services. The proxy extracts request level attributes, and sends them to Mixer for evaluation. You can find more information on this attribute extraction and policy evaluation in our Mixer Configuration documentation.

Mixer includes a flexible plugin model. This model enables Istio to interface with a variety of host environments and infrastructure backends. Thus, Istio abstracts the Envoy proxy and Istio-managed services from these details.

Pilot

Pilot provides service discovery for the Envoy sidecars, traffic management capabilities for intelligent routing (e.g., A/B tests, canary rollouts, etc.), and resiliency (timeouts, retries, circuit breakers, etc.).

Pilot converts high level routing rules that control traffic behavior into Envoy-specific configurations, and propagates them to the sidecars at runtime. Pilot abstracts platform-specific service discovery mechanisms and synthesizes them into a standard format that any sidecar conforming with the Envoy API can consume.

The following diagram shows how the platform adapters and Envoy proxies interact.

Service discovery
Service discovery
  1. The platform starts a new instance of a service which notifies its platform adapter.

  2. The platform adapter registers the instance with the Pilot abstract model.

  3. Pilot distributes traffic rules and configurations to the Envoy proxies to account for the change.

This loose coupling allows Istio to run on multiple environments such as Kubernetes, Consul, or Nomad, while maintaining the same operator interface for traffic management.

You can use Istio’s Traffic Management API to instruct Pilot to refine the Envoy configuration to exercise more granular control over the traffic in your service mesh.

Citadel

Citadel enables strong service-to-service and end-user authentication with built-in identity and credential management. You can use Citadel to upgrade unencrypted traffic in the service mesh. Using Citadel, operators can enforce policies based on service identity rather than on relatively unstable layer 3 or layer 4 network identifiers. Starting from release 0.5, you can use Istio’s authorization feature to control who can access your services.

Galley

Galley is Istio’s configuration validation, ingestion, processing and distribution component. It is responsible for insulating the rest of the Istio components from the details of obtaining user configuration from the underlying platform (e.g. Kubernetes).

Design goals

A few key design goals informed Istio’s architecture. These goals are essential to making the system capable of dealing with services at scale and with high performance.

  • Maximize Transparency: To adopt Istio, an operator or developer is required to do the minimum amount of work possible to get real value from the system. To this end, Istio can automatically inject itself into all the network paths between services. Istio uses sidecar proxies to capture traffic and, where possible, automatically program the networking layer to route traffic through those proxies without any changes to the deployed application code. In Kubernetes, the proxies are injected into pods and traffic is captured by programming iptables rules. Once the sidecar proxies are injected and traffic routing is programmed, Istio can mediate all traffic. This principle also applies to performance. When applying Istio to a deployment, operators see a minimal increase in resource costs for the functionality being provided. Components and APIs must all be designed with performance and scale in mind.

  • Extensibility: As operators and developers become more dependent on the functionality that Istio provides, the system must grow with their needs. While we continue to add new features, the greatest need is the ability to extend the policy system, to integrate with other sources of policy and control, and to propagate signals about mesh behavior to other systems for analysis. The policy runtime supports a standard extension mechanism for plugging in other services. In addition, it allows for the extension of its vocabulary to allow policies to be enforced based on new signals that the mesh produces.

  • Portability: The ecosystem in which Istio is used varies along many dimensions. Istio must run on any cloud or on-premises environment with minimal effort. The task of porting Istio-based services to new environments must be trivial. Using Istio, you are able to operate a single service deployed into multiple environments. For example, you can deploy on multiple clouds for redundancy.

  • Policy Uniformity: The application of policy to API calls between services provides a great deal of control over mesh behavior. However, it can be equally important to apply policies to resources which are not necessarily expressed at the API level. For example, applying a quota to the amount of CPU consumed by an ML training task is more useful than applying a quota to the call which initiated the work. To this end, Istio maintains the policy system as a distinct service with its own API rather than the policy system being baked into the proxy sidecar, allowing services to directly integrate with it as needed.

Was this information useful?
Do you have any suggestions for improvement?

Thanks for your feedback!