perf/

perf/ — competitor comparison harness

Containerised head-to-head perf: gwag vs Apollo Router vs graphql-mesh vs WunderGraph against the same hello-proto, hello-openapi, hello-graphql backends.

docs/perf.md is the on-your-hardware sweep. This directory is the side-by-side. Each gateway runs its own sweep in turn (serial, single-process). Result lands at perf/comparison.md.

Layout

perf/
  Dockerfile               ubuntu:latest + Go + Node + each gateway binary
  competitors.yaml         declarative list of gateways to test
  cmd/compare/             Go orchestrator (boots backends, sweeps each gateway)
  configs/
    mesh/                  graphql-mesh config
    apollo/                Apollo Router config
    wundergraph/           WunderGraph config (TODO)
  run.sh                   container entrypoint

Status: gwag and graphql-mesh wired; Apollo Router in single-subgraph mode against hello-graphql; WunderGraph not yet integrated. Current numbers: comparison.md.

How to run

Docker (hermetic):

docker build -t gwag-perf -f perf/Dockerfile .
docker run --rm \
  -v $(pwd)/perf/.out:/out \
  -e HOST_UID=$(id -u) -e HOST_GID=$(id -g) \
  gwag-perf

Host-local (faster iteration; needs the bench stack down):

bin/bench down                       # frees :18080 + :50090
perf/run.sh local
perf/run.sh local --only gwag        # debug: one gateway

Caveats

• Apollo Router and WunderGraph are GraphQL-federation specialists. They appear here in single-subgraph mode (forwarding to hello-graphql) — the closest apples-to-apples row for a GraphQL-in / GraphQL-out cost.
• graphql-mesh is the closest peer on multi-format ingest.
• Numbers reflect this host's hardware, not absolute speed. Same driver hits every gateway, so the relative ordering carries even if the absolute RPS doesn't.
• First Docker build pulls ~1 GB (Go + Node + Apollo binary + npm tree); allow 5–10 minutes. If graphql-mesh or Apollo Router versions drift, pin in configs/mesh/package.json or bump APOLLO_VERSION in scripts/start-apollo.sh + Dockerfile.