This target is sending the same query to both a backend exposing all the types in a schema and to a router started with a supergraph schema pointing to that same backend for each subgraph. The target then checks that both answers are the same. The invariant tested here is that router will return the expected response regardless of how the schema is cut up, or the number of subgraph requests it takes to get a response. Before launching the fuzz target, you have to spawn a subgraph server and start a router.

Start the subgraph server with this command:

cd fuzz/subgraph && cargo run --release

Start the router with a supergraph in fuzz directory and any router configuration you need to test. Below, the command uses router.yaml configuration in the fuzz directory:

 cargo run -- --config fuzz/router.yaml --supergraph fuzz/subgraph/supergraph.graphql --dev --log=trace

Run the fuzzer with this command:

# Only works on Linux and MacOS
cargo +nightly fuzz run router

This target fuzzes the Connector's Mapping Language and ensures that it continues to compose and that it successfully handles requests to a running router instance with the fuzzed schema.

cargo +nightly fuzz run connectors