3 Comments
User's avatar
Martin's avatar

This is the biggest load microservice slop I’ve ever seen. How to make system 10x more expensive, with 100x more points of failure, while still having a single point of failure.

Tejas Upadhya's avatar

why don't you suggest improvements instead ? This just seems targeted lmao.

How's it microservice slop ? You don't even have all the details.

Martin's avatar

The improvement is simple: keep the cell architecture and collapse the synchronous authorization path into one service.

This is microservices for microservices’ sake. Parse, validate, enrich, transform and route all see the same transaction volume, have to succeed together, and share the same recovery path.

If one fails, the whole transaction is thrown away and replayed in another cell. So the cell is the actual failure domain.

You haven’t built independently resilient services. You’ve built a distributed monolith inside a cell, with extra RPCs, retries, timeouts and operational overhead.