Hacker News Reader

Comment by kccqzy | Hacker News Reader

> A key feature of these environments is that state is not generally persisted from one request to the next. That means we can’t use standard client-side database connection pooling.

So we introduced so many optimizations just because we can't persist state. I can't help but think this is being penny wise pound foolish; the problems this article is solving wouldn't have been problems in the first place if you choose a boring architecture.

View on HN

Serverless environments are a matter of fact. They exist, they are super convenient, and people want to use Postgres in them. This is a bridge for such people. Yes, it requires some sophisticated engineering, but it also enables to have less boilerplate code for a lot of people.

I think if this particular logic is hidden away behind a server, it doesn't really matter - people want convenience, and for someone else to hide the complexity.

Serverless postgres-as-a-service isn't a stretch at all.

This thread does remind me of a previous thread, I believe it was something like optimizing a basic string operation (concatenation?) by going down to assembly and SIMD. Few people have a need to know the details of that particular implementation, and mere mortals cannot comprehend the wizardry going on there either, but that's just fine because if it gets incorporated into standard libraries or the underlying implementations that said standard libraries use, everybody benefits without ever having to understand the inner workings.

Unless it breaks one day and no one knows why

Also known as layers of abstractions. The entirety of (software) technology is built on that concept.

Sure, you want the easiest thing possible that fulfills the common use cases. But right now serverless isn't there quite yet. Having persistent Postgres connections is a really basic requirement that the serverless platforms should've implemented already.

Heroku and similar things are the sweet spot right now for me. I don't have to deal with much, and I get exactly the amount of flexibility that I need. Sure it'd be a little nicer not to deal with Express on my own in a NodeJS environment, but right now it's easier to deal with than it is to avoid. Some day I'll happily go serverless.

Things are getting more serverlesss too. With Deno Deploy, Cloudflare Workers, etc I see myself eventually wanting to run most of my application on platforms like this, and a serverless & distributed database makes that much easier and I expect it’ll probably perform better in many cases too.

100%

I've been doing "serverless" on app engine for more than a decade and it handles postgres connections fine without any heroics.

There's always instances with state somewhere; it's just a question of how smart your serverless infrastructure is about pooling the instances.

The “serverless” being referred to in the article is run using V8 isolates, whereas app engine runs full Node.js processes. Although they have limitations like not supporting raw TCP, isolates can start within a couple milliseconds and are much more CPU efficient than standalone processes.

Bound to be controversial hot take: serverless is the boring architecture you were looking for in the first place

I think serverless is fine as long as you don't lock yourself in to proprietary, vendor-specific APIs. For example, with AWS Lambda and Python, you could use Mangum (see https://mangum.io/ ) so you're deploying an ASGI app, based on say FastAPI or Starlette. For databases, stick with Postgres.

Almost. It'd be boring if it'd handle these Postgres connections efficiently for me instead of making me rely on workarounds.

(As cool as Neon is. If you're already serverless for other reasons, they're providing a nice solution for Postgres.)

As if boring architectures don't have their own problems.

Such as? Besides of course not being able to find people that want to work on it.

Boring architectures generally don’t scale well

Being boring has nothing to do with scale. Some of the biggest applications run on incredibly simple architectures, and often are able to scale so well because of that (boring) simplicity.

Cassandra is probably in the realm 'boring and scalable'

Definitely not. I'd much rather scale Postgres manually through layers of sharding and proxies than handle Cassandra.

There is ScyllaDB for a much better reimplementation of Cassandra/Dynamo, but wide-column databases are still best for niche scenarios, especially as RDBMS are rapidly evolving into natively distributed architectures.

What are the biggest apps on the simplest architectures and what are the architectures?

Stack Overflow is probably one of the more public examples. That is mostly a C# monolith with a relational database behind it.

Monolithic apps running on big servers talking to databases running on big servers. Very little orchestration overhead or complexity with serverless and the rest.

Everything from StackOverflow to ad serving systems to high-frequency trading exchanges are run this way.

Stack overflow is unique in that it’s most read only. Link to high frequency trading “boring” architecture?

What is your definition of non-boring architecture? I'm curious as to what you consider the exciting architectures necessary for scaling to make sure we're talking about the same thing.

Quora, Pinterest, Twitter, etc are all just big app instances talking to DB instances, with separate systems for background processing, queues, and caching. Are you suggesting that they would scale better with serverless functions instead?

Here's a list of architectures: http://highscalability.com/all-time-favorites/

Ironic that this page is down? (also not https? heh)

I usually find that "interesting" architectures scale even worse, at least for the 99% of projects and organizations that'd never actually outgrow the "boring" solution. All the pain of the big fancy solution, but without the actual need that makes it truly better than the alternative.

... of course, the best-paying jobs are at the places that really do need that scale, so of course we all want an excuse to play with those tools, even when it's not the right business choice. And it's not as if software folks are the only ones with a principal-agent problem, in business. Oh well.

I'm with you. I understand the critique, but experience has shown me "boring" is more maintainable, and at some point even "clever" tools stop being worth the effort. Besides, eventually scale pushes tech to roll their own, or rewrite/augment tools.

I dislike using boring as a measure of sufficiency. It makes me wonder if those drawn to the clever solutions carry a "I know better than you" perspective, the kind of dogmatic optimism that tanked SVB.

Define "boring" and "scale", because in my experience you can vertically scale hardware to an almost unbelievable number of concurrently logged-in users.

With most services, until you go over a few million users concurrent users, you really don't need to horizontally scale.

IDK, I'm assuming "boring" here means stateful monolith?

So you have:

a) No enforceable isolation, reliance entirely on good behaviors and module systems

b) No independent scaling

c) No isolation of security components so instead of a minting service and an edge service you have one service that does both

d) Single language lock-in

I could go on and on an don

Boring architectures already exist. Why reimplement what's already done?

Reply on news.ycombinator.com