Connectors Performance Overview

Apollo Connectors are designed with performance in mind, providing efficient ways to incorporate REST APIs directly into your graph through the GraphOS Router.

Subgraph approach vs. direct integration

Before Connectors, integrating a REST API into your graph required spinning up a passthrough GraphQL API:

With Connectors, you eliminate the need for that additional service:

This architecture provides the following performance benefits:

• Reduced network hops with direct router-to-REST communication

• Lower operational complexity without additional services to deploy and maintain

Router performance overhead

The GraphOS Router is optimized to handle Connector requests with minimal overhead:

• Low latency processing of REST API requests and responses

• Parallel execution of independent REST API requests when possible

• Efficient response transformation through declarative mapping language that executes directly in the router runtime

When comparing direct REST calls to Connector-based integration, the additional router processing adds only milliseconds of overhead while providing the significant benefits of GraphQL's declarative data fetching.

Response caching

Connectors don't currently cache responses, and no configuration options exist to tune caching for Connectors. The router routes Connector subgraphs through a separate execution path from standard GraphQL subgraphs, so router caching plugins don't apply to Connector subgraphs:

• Enabling preview_entity_cache for a subgraph that uses Connectors has no effect. The router logs a warning that the plugin isn't supported for that subgraph.

• Cache-Control headers returned by an upstream REST API aren't inspected or honored.

If your use case depends on caching Connector responses, implement caching in the upstream REST API itself (for example, with response headers your CDN or API gateway honors) instead of relying on the router.

Next steps

First, ensure that you're using batch requests whenever possible.

Additionally, the following configurations help prevent overwhelming backend services:

• Request limits (this is also a security feature)

• Traffic shaping