This post from Amex reminded of my time at PagBank, we had a huge monolith there and we spent so much effort into converting it to a service architecture. The idea around a microservice, we all know, is about independency of the services leading to resiliency.

What makes Amex’s approach compelling is not the novelty of isolation as a concept ofc, since engineers have pursued fault containment for decades, but rather the discipline required to enforce it in practice. Cells in their design never span multiple regions, no transaction in the critical path waits on a synchronous call to another cell, and a global router sits at the edge to deterministically send each transaction to the cell that already holds the authoritative data it needs. Reference data such as currency rates and merchant category codes gets replicated into every cell well ahead of time, which means transaction processing never has to reach back to a central system of record while a customer is waiting on a response.

The trade off is honest and worth dwelling on, because cellular isolation increases operational overhead and architectural complexity, and it occasionally forces teams to duplicate services that a single shared implementation would otherwise simplify. Again, at Pagbank, we lived this, sometimes was a step back to make two upfront. For systems where the cost of a widespread outage dwarfs the cost of that added complexity, the exchange tends to be worth making, and the broader lesson generalizes well beyond the world of payments. Resiliency is rarely achieved through monitoring and retries alone, it is achieved by defining clear failure boundaries and then enforcing them relentlessly through design decisions that the rest of the organization has to live with every day.

https://americanexpress.io/cell-based-architecture-for-resilient-payment-systems