Choosing Go at American Express | Hacker News Reader

Choosing Go at American Express

americanexpress.io

43 pointsmadflojo15 comments

"...we narrowed our testing languages to C++17, Java, Node.js, and Go. The first three languages were already in use at American Express, whereas Go was not."

I am curious: why was Go the only language not currently in use at AMEX to be considered? Why not Rust? Why not dotNet Core?

I'm not an expert in Rust or dotNet Core but I can give a bit of insight. At the time (2016) Rust wasn't as visibly used in production as it is today. Also from my understanding over the past few years a lot of new features have gone into Rust which makes it more viable these days.

As far as dotNet Core, it is used at AXP but generally not in our high performance environments. Not saying it couldn't as I'd have to do a lot more research to say one way or another on that. But I'd say that's the primary reason it wasn't added to the list.

My apologies: I missed the 2016 reference -- I wouldn't have recommended dotNetCore three years ago and I'm only now waking up to the power of and speed of Rust and didn't realize just how young it is. I'm still a little amused that Go was the only language not in use already that was considered but understand that the list of possible high performance alternatives was indeed short... shorter than I realized.

All good. There were a lot of factors to creating the list, Enterprise use and readiness was one big one. So we narrowed it down to what we thought were the best choices. Always the possibility we missed something, but thats true of any decision like this.

Appreciate the read and questions though!

Are you on the AMEX team(s) in question? I am always interested in knowing all of the factors that were part of the decision in why a team chose the language or platform it did and if, after a respectable period of time, everyone is still happy with the choice or if there's buyer's remorse and what greener pasture languages/platforms (or problems with the selected stack) are bringing on those second thoughts.

Yeah, thoughts and opinions are my own of course but overall we've had good experiences.

I tried to insert some of the benefits we are seeing into the article in different pieces but now that we are several years later have it running in production. We are happy with the decision. In my opinion, there are even some newer things we didn't write in Go that I kind of wish we did.

The reasons we choose Go are holding true, and I can't say enough about how the performance and low GC times out of the box is something that is very beneficial. At least for my use cases.

>Go achieved the second-best performance at 140,000 requests per second. This number wasn’t far off from the top performer, which clocked in at 167,000 requests per second.

I wonder what the top performer was... I bet it was C.

If you're doing just binary to string conversion there's no need for a features-language like Java or Node. C probably would've been a better option.

C/C++ are never better options because of the number and sheer variety of undefined behaviors, manual memory management and ways to shoot onself on the foot. It's something to use when there's no other choice (legacy code base, embedded systems - though this is also moving to higher level languages).

I promise you they are. There's a reason why over 90% of the programs you invoke or interact with on a daily basis are written in C.

Yes. Most of them are evolutions of legacy code bases and that's why they're also security nightmares.

If you're building anything new it's very hard to make a valid case for coding in C. You're almost always better off working in Rust, Java, Swift or Go

It would be nice to see a breakdown of pros/cons for the various languages, the article doesn't really explain why other languages were excluded, it just talks about the benefits of Go. Also, I'd be interested if the same decision was made today, i.e. V8 has different performance characteristics than it did in 2016, and there are interesting ways to reduce/overcome warmup time.

>These platforms were purpose built for performance, concurrency, resiliency, and availability. Given advancements in the programming space, we wanted to see what languages best suited our reliability and performance needs. We also needed something that fit modern infrastructure and design patterns.

>As such, we launched an effort to evaluate different languages.

I feel the purpose of a payment network is performance, concurrency, resiliency, and availability, and somewhere, wayyy towards the bottom of the list, at 2px font size, modernity and extensibility are there.

Very interesting blog article. Thanks for sharing. Are you able to comment on how ML might be integrated to the payments and rewards system? Is it just a matter of APIs on top of Kubernetes/Docker? Or did you have to recreate models in Go?

Seems weird to choose just one of { C++, Go, Java, Javascript }. None of these languages is really a substitute for any of the others, except perhaps for Go and Java.

Go certainly cannot cover all C++ uses, nor Javascript/Node.

They weren't choosing for all use cases, but for the ones they needed.

Reply on news.ycombinator.com