Neon: A serverless open-source alternative to AWS Aurora Postgres | Hacker News Reader

Neon: A serverless open-source alternative to AWS Aurora Postgres

github.com

101 pointsthunderbong50 comments

I've been using https://neon.tech in production at Devlog https://dev.log.xyz for the last few months, really quite happy with the pricing and the performance. Clearly made by and for developers.

The only surprise I had was that they don't show you any database logs! Turns out I haven't needed this so it's not a big deal, but that's the biggest departure from a "traditional hosted postgres" service.

Aside from that, the only missing feature is the ability to create an Organization rather than hosting projects in a personally-affiliated account, but this is apparently coming soon.

The branching features work really well and I've used it a few times now to debug issues and test migrations without impacting production. I haven't set up the per-pull-request-preview-environment stuff yet but I plan to and I like that Neon makes it easy.

EDIT: oh, the one other annoyance is that last time I checked, it was not possible to create a read-only replica in a different region than the primary compute node. Or something like that, I'm still not sure, and multi-region read replicas are not particularly necessary for me right now (Devlog still in early alpha, not a lot of customers) so I just moved on. Issue tracked here https://github.com/neondatabase/neon/issues/4178

Neon is a managed service versus a hosted service FWIW. Exposing logs is something that we'd like to do in the future.

Apologies, I'm not super clear on the difference between managed and hosted. Very much a happy customer already and excited to see continued improvements!

Here[0] is a breakdown.

[0]: https://www.hostreview.com/blog/210610-what-is-the-differenc...

This reads like complete SEO slop to me. What is the important difference between managed and hosted re: Neon?

An important difference is that we are monitoring your database for abnormalities like replica lag, server crashes (for whatever reason), etc. We are handling server and extension updates. We are trying to make it so that your database is not something you have to worry about in the slightest.

Oh wow. I love the idea of DevLog. I’ve spent a lot of time helping engineering teams get better at regularly communicating what we’re doing, with context, that everyone in the company can understand. Have you shared this yet?

I haven't shared this widely yet but a few friends have tried it out and we're dogfooding it internally as we develop it. You should be able to sign up and try it out self-serve right now (for free until we set up a business bank account!) but it is very much an alpha product, bugs included.

Send me an email (peter @ dev.log.xyz OR peter @ peterdowns.com) if you'd like a demo or you're willing to tell me more about the problems you're hoping Devlog can solve for you!

This is crazy I just stumbled upon this company earlier today via search when I realized that Aurora Postgres Serverless V2 no longer allows you to scale down to 0 in times of low usage. AWS makes you select 0.5 ACUs as the minimum for V2 where with V1 you could select 0... is this just a straight up money grab by AWS to make the minimum monthly cost for aurora like $50 or is there a justifiable reason behind making you run virtual servers continually when there is no usage. I may be switching soon!

I'm guessing AWS simply did not build reading/writing the current DB state to/from S3 into Aurora Serverless v2. Neon has done a ton of work to make scale to zero happen in terms of treating the actual database disk as nothing more than a cache.

The company's site https://neon.tech/ made it easier to understand what it is, for me. The Github README struggles with identifying its audience, IMO.

Please say more of what you would change in the README and we will update

This is the company vercel has bet on to power their managed DB. Stacked team behind it. One of the few managed postgres companies that will still exist years from now.

Been on neon for a while. Using it for simple queries and RAG (pgvector). The thing I like about it most is the speed where I can get query output in 200ms-500ms. Incredibly fast. Given the serverless nature, it's my go to for any POCs or tests I do personally. In production, querying on neon is faster than our inhouse postgres we provision from AWS.

One really hacky application I saw was a pg search and show results without the user having to press a button. Here: https://next-ai-news.vercel.app/search?q=

I'm curious what the largest DB hosted on Neon is currently? Can you potentially scale postgres infinitely with neon without needing to shard etc?

I can't find where I read it but my current understanding is that you can't. You can scale to the largest AWS instance available for writes. They allow for automatic read replicas which I imagine can scale horizontally?

Postgres itself has limits that Neon cannot help: https://www.postgresql.org/docs/current/limits.html. Neon currently serves TB-sized databases. Our storage team has put in a ton of work to enable that. Neon does not currently support something like Citus. Neon has cheap read replicas however, which can allow your primary to just service writes.

Here is the helm chart: https://github.com/neondatabase/helm-charts/tree/main Although documentation is sparse.

Percona has a good walkthrough on how to spin up neondatabase via plain docker containers, which is a good way of pulling the system apart into logical units (storage broker, safekeeper, pageserver), and branching from an existing tenant/timeline:

https://www.percona.com/blog/using-docker-to-deploy-neon-ser...

I’m a big fan of Neon postgres. It’s pretty much unbeatable value for the price. The documentation is good and multi-tech stack friendly and I created a repo in which I added all the awesome links about Neon Postgres - https://github.com/tyaga001/awesome-neon and wrote a tutorial https://www.freecodecamp.org/news/build-an-invoice-saas-app-...

This has got to be one of the slickest websites I've seen on the web, kudos strongly deserved.

On the tech itself, it's a shame nothing like this exists for MySQL, but not surprising.

We debated internally if we wanted to provide MySQL too. On one hand it’s not that different and our storage can be adapted to run MySQL.

On the other hand focus and product identity are very important for startups. So it’s only Postgres

There's a project to try get a MySQL compatible extension for Postgres going, which may work.

https://nextgres.com/

Isn't this what PlanetScale is all about?

It's not open source, and whilst its built on Vitess, Vitess alone doesn't do what Neon does, just makes it easier to horizontally scale.

What makes this attractive to me is the ‘database branching’ functionality and simple setup for it.

Managing databases locally and in CI isn’t much fun when dealing with migrations, so the fact you can trivially work with a copy in a branch, or deploy preview builds in PRs with their own DB branch, is a great innovation.

I mean, sure you could slap something similar together with some scripts, but having it out of the box is good dev ex.

I thought that Postgres was an open-source alternative to AWS Aurora Postgres?

Hm. I thought that AWS Aurora Postgres is a proprietary alternative to Postgres?

If any neon product people are on here, please bring it to gcp

We just announced upcoming Azure support today[0]. Right now we are fully invested in AWS and Azure. Launching in additional regions can take quite a bit of work, but that pales in comparison to another cloud. We'd love to launch everywhere that our customers or potential customers are, but it also requires a larger team. We've been hiring[1] for Azure-specific positions recently, and I assume we would do the same for GCP, OCI, or IBM Cloud.

[0]: https://neon.tech/blog/neon-is-coming-to-azure [1]: https://neon.tech/careers

Sounds like a neat idea.

Is it compatible with postgresql extensions? (Do most extension live in the compute layer?)

Here is the list of Postgres extensions we support: https://neon.tech/docs/extensions/pg-extensions.

The extensions are bundled into the compute images.

Shouldn't the title be,

Neon: A serverless open-source alternative to Amazon Aurora PostgreSQL Compatible.

neon.tech's offering looks good - the $69 offering.

Any reviews?

I've used it, it's one of the only dbs you can use for free. It's extremely easy to get started. The only thing I'm worried about is the bait and switch, the offering is so much better than the alternative that I'm certain there will be a squeeze eventually.

I don't understand the business model. I guess the goal is to be acquired by AWS/GCP/Azure etc?

It's cheap to have a live db because the compute is detached from storage. This means they can spin up your compute on demand. While the compute is running it's around the same price as aws, gcp, azure, etc but you pay for more storage because it's replicated. One really nice thing about this is that you do not need to shut down the db to upgrade the machine and you get what are essentially free read replicas.

I think the business model makes sense, I just think they will eventually raise prices because they can.

1) That's how e.g. RDS is already implemented, right? It's an AWS cash cow.

2) Who is going to use an external (from the perspective of your cloud or internal work load) RDBMS?

RDS isn't implemented like that and if you are on AWS you can get neon to be in the same datacenter and even pay via AWS.

I don't work for neon, I've just gone pretty deep on all the managed db offerings.

(As in "compute is detached from storage", not all of the specifics.)

> if you are on AWS you can get neon to be in the same datacenter and even pay via AWS

Nice.

As the title says... Aurora is somewhat implemented like that.

RDS just runs regular database servers and provides some nice management automation around it.

The business model is to sell a Database as a Service, similar to Mongo, DataStax, etc.

Isn't the storage kinda expensive? It's $15 per 10 GiB, which is ~10x more expensive than AWS SSD storage and ~100x more than bulk storage with an S3-like service. It's $1500 per TiB, which is a lot of money. I guess it might not be designed for time series data like I am used to collecting, which requires thinking on the TiB scale at the very least.

I do appreciate it's Apache 2.0 licensed, so you can run it yourself. I can't complain too much about pricing for what looks to be my dream database that I have always wanted with branching and separated storage and compute.

Yeah that storage pricing is a killer. Given Aurora is $0.1/GB, I just wouldn't even consider this.

Neon is a lot more than just storage. It's branching, autoscaling, (cheap) read replicas, serverless computing, etc. If you only care about storage pricing, Neon is not the service for you. As you've said, storage pricing is cheaper elsewhere.

Sure, but I've got serverless computing with Aurora v2. I've got upto 15 read replicas, blue/green deployments, upto 128 cores per replica at a cheaper price... I'm just not really thinking that a bit of branching will make up for a 10x or more increase in cost.

Those increased storage costs make the decoupled compute and the associated auto-scaling benefits almost irrelevant as far as the overall cost picture goes.

We have former Aurora users come to Neon who tell us they saved money by switching. If someone like yourself were to migrate, they definitely would be an enterprise sale, which has different pricing than what is available in our current tiers.

What counts as storage? There’s a huge difference between storage vs database size as there’s a huge overhead otherwise.

RDS and Aurora usage just keeps growing even if you don’t have much real data.

To be fair to neon: you get free point in time recovery, HA, branching, automatic scaling without downtime, and read replicas (that also automatically scale).

Yea, it's not really a product for the type of massive bulk storage I'm thinking about. This seems more like the database you use for everything that isn't the bulk storage. Things that get read/written to often enough to care about versioning and HA and all those other things. Time series bulk data wouldn't be a good application for their managed service since it's usually "write once, read sometimes, store cheap".

> Isn't the storage kinda expensive?

Depends on the actual performance. Normal EBS is dead slow but you can pay for the IoPs.

Reply on news.ycombinator.com