Envoy v2 JSON REST and gRPC APIs

Goals

This repository contains the draft v2 JSON REST and gRPC Envoy APIs. Envoy today has a number of JSON REST APIs through which it may discover and have updated its runtime configuration from some management server. These are:

• Cluster Discovery Service (CDS)
• Rate Limit Service (RLS)
• Route Discovery Service (RDS)
• Service Discovery Service (SDS)

Version 2 of the Envoy API will evolve existing APIs and introduce new APIs to:

• Allow for more advanced load balancing through load and resource utilization reporting to management servers.
• Improve N^2 health check scalability issues by optionally offloading health checking to other Envoy instances.
• Support Envoy deployment in edge, sidecar and middle proxy deployment models via changes to the listener model and CDS/SDS APIs.
• Allow streaming updates from the management server on change, instead of polling APIs from Envoy. gRPC APIs will be supported alongside JSON REST APIs to provide for this.
• Ensure all Envoy runtime configuration is dynamically discoverable via API calls, including listener configuration, certificates and runtime settings, which are today sourced from the filesystem. There will still remain a static bootstrap configuration file that will specify items unlikely to change during runtime, including the Envoy node identity, xDS management server addresses, administration interface and tracing configuration.
• Revisit and where appropriate cleanup any v1 technical debt.

Status

Draft work-in-progress. Input is welcome via issue filing. Small, localized PRs are also welcome, but any major changes or suggestions should be coordinated in a tracking issue with the authors.

Principles

• Proto3 will be used to specify the canonical API. This will provide directly the gRPC API and via gRPC-JSON transcoding the JSON REST API.

• xDS APIs should support eventual consistency. For example, if RDS references a cluster that has not yet been supplied by CDS, it should be silently ignored and traffic not forwarded until the CDS update occurs.

• The API is primarily intended for machine generation and consumption. It is expected that the management server is responsible for mapping higher level configuration concepts to API responses. Similarly, static configuration fragments may be generated by templating tools, etc. The APIs and tools used to generate xDS configuration are beyond the scope of the definitions in this repository.

• Listeners will be immutable. Any updates to a listener via LDS will require the draining of existing connections for the specific bound IP/port. As a result, new requests will only be guaranteed to observe the new configuration after existing connections have drained or the drain timeout.

• Versioning will be expressed via proto3 package namespaces, i.e. package envoy.api.v2;.

• Custom components (e.g. filters, resolvers, loggers) will use a reverse DNS naming scheme, e.g. com.google.widget, com.lyft.widget.

• Wrapped protobuf fields should be used for all non-string scalar types, to support non-zero default values. While only some fields require wrapping, for consistency we prefer to have all non-string scalar fields wrapped.

APIs

Unless otherwise stated, the APIs with the same names as v1 APIs have a similar role.

• Cluster Discovery Service (CDS).
• Endpoint Discovery Service (EDS). This has the same role as SDS in the v1 API, the new name better describes what the API does in practice. Advanced global load balancing capable of utilizing N-dimensional upstream metrics is now supported.
• Listener Discovery Service (LDS). This new API supports dynamic discovery of the listener configuration (which ports to bind to, TLS details, filter chains, etc.).
• Health Discovery Service (HDS). This new API supports efficient endpoint health discovery by the management server via the Envoy instances it manages. Individual Envoy instances will typically receive HDS instructions to health check a subset of all endpoints. The health check subset may not be a subset of the Envoy instance's EDS endpoints.
• Rate Limit Service.
• Route Discovery Service (RDS).

Terminology

Some relevant existing terminology is repeated below and some new v2 terms introduced.

• Cluster: A cluster is a group of logically similar endpoints that Envoy connects to. In v2, RDS routes points to clusters, CDS provides cluster configuration and Envoy discovers the cluster members via EDS.

• Downstream: A downstream host connects to Envoy, sends requests, and receives responses.

• Endpoint: An endpoint is an upstream host that is a member of one or more clusters. Endpoints are discovered via EDS.

• Listener: A listener is a named network location (e.g., port, unix domain socket, etc.) that can be connected to by downstream clients. Envoy exposes one or more listeners that downstream hosts connect to.

• Locality: A location where an Envoy instance or an endpoint runs. This includes region, zone and sub-zone identification.

• Management server: A logical server implementing the v2 Envoy APIs. This is not necessarily a single physical machine since it may be replicated/sharded and API serving for different xDS APIs may be implemented on different physical machines.

• Region: Geographic region where a zone is located.

• Sub-zone: Location within a zone where an Envoy instance or an endpoint runs. This allows for multiple load balancing targets within a zone.

• Upstream: An upstream host receives connections and requests from Envoy and returns responses.

• xDS: CDS/EDS/HDS/LDS/RDS.

• Zone: Availability Zone (AZ) in AWS, Zone in GCP.