How eBay Runs One of the Most Complex GraphQL Federations with WunderGraph
A deep dive into performance, correctness, compliance, and open-source collaboration at a global scale.
Company Overview
eBay operates one of the world’s largest online marketplaces, connecting millions of buyers and sellers across a wide range of product categories. Its platform supports complex search, discovery, pricing, payments, logistics, and trust systems, all operating at a global scale.
To power its web and mobile experiences, eBay relies on a large federated GraphQL architecture spanning many subgraphs and long-lived domain boundaries. The supergraph processes hundreds of thousands of requests per second and supports deeply nested queries shaped by years of platform evolution.
At this scale, GraphQL federation is core infrastructure. Performance, correctness, and compliance are not optimization goals. They are requirements with direct impact on reliability, security, and business risk.
The Challenge
Scale, Correctness Risk, and Enterprise Constraints
eBay operates a federated GraphQL system at a scale where small inconsistencies become systemic risks. With many subgraphs, deeply nested queries, and hundreds of thousands of requests per second, the federation layer must behave predictably and correctly under extreme conditions.
As the graph evolved, limitations in existing federation tooling became more visible. Edge cases in query planning, ambiguous specification behavior, and limited observability made it difficult to reason about execution paths, validate correctness, and satisfy compliance requirements at scale.
Key challenges
- Needed predictable query planning across many subgraphs
- Encountered correctness gaps around alias handling and compound keys
- Faced ambiguity in federation specification behavior at enterprise scale
- Required auditable data access for security and compliance teams
- Needed performance to remain stable under burst traffic and extreme load
“The federation spec has some loose and unclear details, and the behavior is undetermined.”
The Solution
Migrating to the Open-Source Cosmo Router
eBay evaluated WunderGraph as a way to stabilize and simplify its federation layer without introducing new operational risk. At their scale, any solution had to sustain extreme traffic, preserve correctness across many subgraphs, and remain transparent enough to satisfy security and compliance requirements.
Rather than relying on benchmarks or synthetic tests, eBay validated the Cosmo Router under real production conditions. The router was introduced alongside existing infrastructure and exercised against live workloads to confirm query planning behavior, runtime performance, and stability under burst traffic.
“WunderGraph is currently handling hundreds of thousands of requests per second for us.”
As confidence increased, traffic was progressively shifted to Cosmo. The open-source model allowed eBay to inspect and influence federation behavior as new edge cases surfaced, reducing long-term dependency risk while maintaining full operational control.
Today, all of eBay’s federated GraphQL traffic runs on WunderGraph in production, with predictable performance and behavior under sustained global load.
The Results
Predictable Performance, Lower Operational Overhead, Enterprise-Grade Control
The migration to WunderGraph simplified how eBay operates its federation at global scale.
Running the same production traffic on fewer nodes reduced operational complexity and made failures easier to contain, while delivering immediate and measurable improvements in performance and reliability.
Key outcomes
- ~50% reduction in router memory usage while handling the same production traffic
- 80% fewer router nodes with stable performance at peak load
- Hundreds of thousands of requests per second sustained in production
- Thousands of RPS handled per individual router instance
- Significantly reduced CPU utilization across the federation layer
- More predictable and correct query planning at scale
- At global scale, 100% of eBay’s federated GraphQL traffic now runs on WunderGraph in production
- Full approval from internal security and compliance teams
- No vendor lock-in or proprietary dependencies
“Even a single instance handles a couple of thousand requests per second, easily.”
The Conclusion
Built for Scale, Correctness, and Long-Term Control
WunderGraph gives eBay a federation layer that performs reliably under extreme load, behaves predictably across a complex graph, and meets enterprise compliance requirements.
By improving query planning, clarifying federation edge cases, and making data access auditable, eBay reduced operational risk while sustaining hundreds of thousands of requests per second in production.
Today, eBay’s federated GraphQL traffic runs entirely on WunderGraph, providing a stable foundation that can scale and evolve without vendor lock-in.
Ready to scale without the chaos?
Speak with an expert to discover how Federation becomes simple and seamless.
