Federated GraphQL and data ingestion

Problem

Product teams were calling fragmented backends. Clients should not care which of 30+ providers owns a field. The read path still has to stay fast while billions of events land. If this path is slow, store operations feel it.

Approach

• Built a federated GraphQL gateway (Apollo) so each domain publishes a subgraph and clients query one graph.
• Paired it with high-throughput ingestion for billions of events.
• Added a read-through cache on the hot path so repeated reads do not wait on every provider.
• This is the same class of work that started with associate clock-in status for 5000+ stores and about 3 million associates: 99.999% uptime because store operations run on it.

Impact

• 30+ providers behind one graph.
• Read-through cache for low-latency reads.
• Billions of events on the ingestion path.
• 99.999% availability for millions of users on a tier-0 path.