More on the topic...
Generating detailed summary...
Failed to generate summary. Please try again.
American Express rebuilt its core payments platform around independent “cells,” each packed with its own microservices, databases, DNS and support services. Every cell can process credit‐card transactions on its own, and failures stay inside the cell rather than rippling across the network. You can pull a cell out of rotation for maintenance or unexpected faults without touching the others. That containment strategy shrinks the “blast radius” when something breaks and keeps latency low by avoiding cross‐cell calls in the transaction path.
To keep data local, static reference sets—currency rates, merchant category codes and so on—get pushed into every cell well before transactions hit the system. That way, no request ever pauses for a fallback lookup to a central store. For changing data that can’t be fully preloaded, American Express uses a Global Transaction Router at the edge. It reads transaction details—partner ID, market, payment type—and sends each request to the cell holding the current authoritative state. Any asynchronous replication happens outside the critical path so the router always directs traffic to the right place without adding delay.
Inside each cell, microservices speak only to local database instances over Kubernetes’ internal network. If data needs to move between cells—for example, to propagate a state change—messaging queues do it asynchronously. All ingress and egress flows must pass through the Global Transaction Router, which doubles as a payment mesh. That enforces cell boundaries, guarantees consistency, and ensures transactions never shortcut around a cell’s failure domain.
Questions about this article
No questions yet.