[athenot]

Why? Having many languages allows to pick the sweetspot for each subsystem. Each language has special strengths and weaknesses, there is no silver bullet that excels at everything. Go, C/C++, Lua, Ruby, Perl, Scala, Node.js, Python... each of these are THE best choice for certain classes of problems (and terrible for others). It may be because of language features that elegantly express a solution, or particular efficiency considerations (speed, latency or memory footprint), or integrations with specific libraries (if language X has a lib that does 90% of the requirement, use it instead of reimplementing from scratch in language Y). Also, it may be a team consideration: the talent pool for front-end engineers has different preferences than the talent pool for some back-end systems.

In a large system like Tumblr, they are likely highly distributed with many different subsystems running in different parts of their infrastructure, each with their own SLA and resilience requirements.

Their team is probably large enough that most engineers focus on only some of the subsystems, and communication between them is via defined interfaces.

Though having a diverse ecosystem fosters an openess of mind, instead of enforcing "One Metaphore To Rule Them All". You find yourself borrowing efficient patterns from other environments, as they eventually start cross-pollinating.

[pencilhappen]

> Each language has special strengths and weaknesses, there is no silver bullet that excels at everything. Go, C/C++, Lua, Ruby, Perl, Scala, Node.js, Python... each of these are THE best choice for certain classes of problems (and terrible for others).

Not quite true... not all languages have a sweet spot in a production environment. Node.js isn't particularly excellent at anything, the attraction is mainly "I know JS, and I don't want to learn anything else right now", which is a terrible attitude for someone who wants to have a career in tech. Knowing more than one tool in the toolbox is a key skill, since there is no one-size-fits-all.

Jury is still out on Scala. It's a big language, but unclear if there's a production sweet spot or not compared to other JVM hosted languages.

The rest of the ones you listed have their definite sweet spots. There are tons of languages though, and most don't have one.

[athenot]

> Node.js isn't particularly excellent at anything

That's not very kind. Yes it has its flaws but it's not bad at being a higher abstraction over async-everything, augmented with a huge module repository. Support for isomorphic javascript can come in handy for some applications.

I've never done Scala but my understanding of it is that it could be a good fit for modelling certain types of business logic while being able to take advantage of exiting java libs out there.

> Knowing more than one tool in the toolbox is a key skill, since there is no one-size-fits-all.

Yes, we agree here. That was the point I was making.

[pencilhappen]

Node is adequate at async, but not excellent. As I said in a sibling comment, things that are synchronous still, like GC, make it fall down at scale. Isomorphic JS isn't that useful in most real situations.

My point about Scala is that there doesn't seem to be a clear cut niche that it excels at. There are numerous other non-Java languages that run on the JVM that could be in the running, and they all get to take advantages of the whole Java ecosystem too.

[0xCMP]

Ruby, Python, Perl, PHP, and (maybe) Lua aren't amazing at handling async operations like Node.JS can.

Go is, in many respects, a great replacement for Python and Node in that it's simple and handles concurrency well. Scala and other JVM languages fall here too, but come with their assoc complexity.

Python, Node, and Ruby have tons of battle-tested code which often makes the language the "strongest" if for nothing else than there is code that works and it's not critical enough we rewrite it ourselves.

There are definitely sweet spots for each of these languages.

[pencilhappen]

Node isn't great at async, despite the hype. Important things are still synchronous, most notably the GC, so you can very easily wind up with service stalls at scale. The module churn in Node land means that few things are actually battle tested.

BEAM languages are the sweet spot for concurrency right now.

[erroneousfunk]

I had a manager that loved Java. He wanted to add a new helper service onto an existing service that was written in Python, but the new service would be Java (because Java is the best, obviously).

This new helper was responsible for transforming, combining, and synthesizing large API responses with somewhat variable and highly nested data structures into other large API responses with somewhat variable and highly nested data structures. It certainly was not Java I was proud of, but it would have been trivial (and FAR more readable) in Python. Ugh.

[pmcollins]

But there are so many benefits to standardizing on one language (to take the extreme case). You create the business rules, APIs, and utilities in whatever language you choose and that can be shared by everyone. That becomes the core of your business. If you need something new, you create it and submit it for review, and then everyone can use it, not just the folks who choose what was your preferred language at the time of writing.

I've been in organizations where you have the Language/Platform X people over here and the Language/Platform Y people over there, and they duplicate work, don't collaborate, and their libraries don't interoperate. They have different hiring needs, and developers can't easily move between them. Ultimately, everybody with their pet languages moves on, and new people come into a very fractured environment and ask themselves WTF is this? Over the years, developers (especially new ones) are only a fraction as effective as they would be if the company had just paid the relatively small up-front price of nudging everyone into the same ecosystem.

And BTW, I have yet to encounter a problem that is hard to solve in my preferred language, but is way easier to solve in another. I think the key to that is to choose carefully what you're going to commit to.

[efficax]

> And BTW, I have yet to encounter a problem that is hard to solve in my preferred language, but is way easier to solve in another. I think the key to that is to choose carefully what you're going to commit to.

Long-running light-weight servers are nearly impossible to write in PHP, but a breeze in Go. There are tons of problem spaces where the only available libraries are in C/C++ or Java, which means its going to be easier to solve the problem in those languages than it is to build a new library from scratch. Python has some of the best packages for a number of problems, like machine learning, data science, image manipulation. Even if Go is purpose built for web crawling, beautiful soup (Python) is still what I'd choose for getting a scraper up and running quickly. And of course since The Web Is King everybody has to use JS for something these days.

If you're a large shop doing a lot of things, it's going to be more work to stick to one or two languages than it is to pick the best language for the job.

[bluesnowmonkey]

People don't just move between teams, they also move between companies, even moreso in my experience.

There's this mindset in some companies that they're IBM, they are the world, but the it's not like that anymore. You can't expect people to sit through your company training on how to use your proprietary tools and then spend 30 years working there. You want to be able to use all types of people coming in with all types of experience because it's very diverse out there. And they want to use and build skills that they can take to their next company or leverage to start their own company. You want highly motivated individuals with entrepreneurial mindsets right? What kind of environment will attract them?

Go ahead and say, "We only use COBOL here," and see how that works out. Great everything is interoperable. But you've got a team of C players. Maybe that's fine though.

[lilactown]

IMO, we should focus more on standardizing on runtimes rather than language. Languages evolve, and new ones come out that solve problems better. We luckily live in an age where many of those languages can interop at the runtime level.

As a developer, and as a team, it sucks to be told "you must use Java and JavaScript for time immemorial." An easy compromise is that any package/library/service/etc. must target the JVM or JavaScript VM and provide interop with existing code.

[Consultant32452]

>Though having a diverse ecosystem fosters an openess of mind, instead of enforcing "One Metaphore To Rule Them All". You find yourself borrowing efficient patterns from other environments, as they eventually start cross-pollinating.

I've had the opposite experience. Having a lot of different technology stacks increases the barrier to entry for anyone who might be interested in working on another part of the ecosystem. This results in increased siloing of people and information.