How Figma Hacked Postgres Into Scalability 

Please stop with the clickbaity titles! PostgreSQL is probably the best database choice for most businesses. Figma reached a scale most businesses will never reach. And by then they should have the resources to deal with these kinds of problems. Stop telling people how "absurd" or "insane" it is that you have to spend time writing benchmarks to find bottlenecks in your database. There is a world outside cloud providers, you know? There are businesses who just don't want (or can) throw money at cloud providers to solve all their problems (think regulated markets). You will get VERY far with a PostgreSQL database on a dedicated host and it's gonna cost you peanuts. Enough with the hot takes and one-sidedness.

Dev Ops Engineer here and I can attest that Postgres does indeed scale very well and is probably the best relational DB out there right now which is why its so widely adopted. If you have over 4 million users and over a billion rows yeah you’re going to run into some issues regardless of what DB you choose. To say Postgres doesn’t scale though is wild given this very specific use case. Front end devs need to sit this one out.

Interesting post, but i was annoyed by Theo constantly going on about how it's "absurd" that Figma had to dive deeper into these issues, and suggesting that this is all because postgres "doesn't scale". Figma is one of the most successful tech companies of the last 10 years. They have an insanely advanced product. Ofc they're going to run into problems that very few companies have and be forced to develop tailor-made solutions. The idea that this is somehow proof that we as regular devs would be better off opting into proprietary services by SaaS companies like Planetscale is nonsense.

Dear reader, this is a problem you will not have in your side project, or probably ever. I'm starting to understand why this guy's credibility is tanking so fast. Imagine, he said large companies shouldn't need database engineers 😂

Theo is getting cooked hard in the comments

It's absurd that a 10 BILLION dollar company handling terabytes of data has to hire a database team? Lol, okay Theo. Seriously, he whines about anyone having to know how to do anything beyond writing simple front-end work, and keeps praying for a world where all developers are too incompetent to handle literally any remotely difficult problems and all real work is centralized to a few monopolistic companies that everyone just pays to do everything. Absolutely no lessons learned from cloud computing in general or the recent Vercel bullshit or the recent Cloudfare bullshit, etc. There's a lot of money to be made in being a developer that can actually deal with a problem that you can't just npm install away.

oh another daily video of Theo reading a blog post about something most developers will never ever have to worry about

For figma it's not absurd to be this deep into databases. All these real time collaboration features comments and other stuff that people use daily will make their DBs explode with data just as Discord.

"Postgres doesn't scale" is such baloney

I run multi region postgresql clusters (3 to 5 geographical regions) with intra region auto scaling (vertical and horizontal) plus a layer of geo redundant pgpool clusters on top… the smaller ones cost my clients around 30 bucks/month at minimum load. So please stop this nonsense bullshit, there is no “one size fits all” data layer out there, and Postgres scales very well until your past software design decisions come back to bite you in the ass. And that’s the case for every company that grows so fast like Figma. Nothing saves you from benchmarking, analyzing, refactoring - rinse and repeat. I will always and ever choose Postgres over this new shiny toy cloud databases out there, period.

For the average joe building side projects, this scaling doesn't matter. As they say "you'll probably have more servers than users"

There is nothing that hurts more than listening to frontend guy talk about databases

Oh damn postgres doesnt scale? I need to migrate my side project quickly. Before i have more than 5 users....

A small, 750+ employee, company such as Figma ...

"Imagine if game devs had to understand DirectX and assembly..." I don't have to imagine. I was there when they did. Games ran a lot faster, with less bugs, and without being cookie cutter bullshit cranked out with Unity.

Limited knowledge about the subject, yet spoke with complete confidence right from the get go. This is not just bad, it's misleading. Unsubbed. Edit: He's sponsored by PlanetScale. No wonder...

LMAO Theo sometimes you say some really dumb stuff

"Imagine if game devs had to understand DirectX and assembly..." they actually do if they are legit, and I think any developer building a product that need a scalable database SHOULD understand how databases work and any team the size of Figma's SHOULD have world class DBAs

2:56 Figma... just an UI app. Lmao

Figma and Canva are monster products. People dont see the complexity and scale they operate

"postgres don't scale that well " **Spits out coffee**

So many people will discover how Theo does not know what hes talking about from this video ... oh wait, they will not, they will still think he is amazing despite how much nonsense was said. No it's not absurd to explore the limits of your DB and it's not that complex - it's a lot of work, but many people are doing it, I have do it to some extent. No it's not absurd that Figma needs to do that and it's not an issue with DB design. It's a big company blasting it's DB with seemingly complex data model. What is absurd is you constantly throwing the word absurd.

100% BS. and stop always saying this/that company isn't sponsoring THIS video. you think we're dumb? there are sponsoring your channel

This video is useless for 99.99999% of devs

Don't love the generalization that postgres is somehow not fit for purpose, when it's perfectly fine until you're large enough to throw money at the problem. Never mind the fact that other databases have similar issues in some way or another. We can all benefit from learning about databases instead of treating them as black boxes that do or don't work.

7:25 - Please realize that this philosophy only applies for low-stakes applications. If you deal with high stakes - think medical, bridge building (more fun because of the metaphor), once your safety rails fail, you end up in jail and that is that.

I ❤ postgres. I would feel so lucky if my my dbs did this. This is a great problem to have. I have had dreams where my side project went viral and i end up with a 2k vercel bill and wake up with a cold sweat, imposter syndrome, and rustled jimmies.

To everyone: almost none of you will ever encounter this kind of problems, this is the top 0.00001% of scaling(probably need to add a few more zeros but you get the point). To Theo: you are so out of your realm, and it shows: - No shit you should know how to scale your database and not just rely on a 3rd party to keep your DB good, what happens when that 3rd party's system fails? Or they go bankrupt? - wtf was that gamedev comparison?? Games are so sensitive to perf changes that you need to know how it will compile down. - Postgres is one of the base scaling DBs once you started using 3 things - Yes your sponsoring makes you biaised (I'm not saying you are not enjoying the product) Overall so much stuff is wrong and you said it with so much confidence that it is clearly an issue

Stop comparing MySql to Postgres it is another league. I don't know if Figma needs Postures functionality but if talking about planetscale is just stupid.

Would hate to have theo working on a complex app. Has some serious gaps in knowledge on scale

Theo is the type of guy who still don't know Mongo have relational references.

Good blog... You lost me today when u again mentioned planetscale. Like stop comparing different solutions that too for scalability... Its bad when u miss represent something so much that too repeatedly. Take this comments section of todays video as a lesson.

sounds like a skill issue on Figma side or terrible design decisions

There is NO escaping scalability problems when you run into them, especially at the storage layer. Everyone that scales to certain point has to deal with it. As many people also find out, if some scaling problem is being “handled” for you by a third party, it becomes a growing finance problem that follows the same curve, eventually business will be breathing down your neck to get costs down or move off because the costs are becoming unsustainable.

Theo, Postgres scales very well-MySQL too. This is an outlier, like so few others where their earlier technological decisions now require them to make changes as they scale more. And yes, as large game companies, you definitely have people that worry what a rendering API is doing under the hood and what it compiles to in assembly during performance regressions-in fact it's even more crazy than that, since you have to test for so much hardware that you have to limit it and set "system requirements" based on the reasonable testing done to make sure the game runs well across those systems.

There's zero chance a company would have set up horizontal sharding from the start. Period. It would have been a waste of money in the time for where they were at.

I think for a lot of use cases you don’t really need to do Sharding yourself. The Citus extension helps quite a bit. I would argue that Postgres is the best database for almost any business.

can't take you seriously for anything outside the realm of frontend stuff.

I am not sure what caused their problems. I have seen people on youtube building Figma clones in a week and it works perfectly without any issue. Yeah, I am joking.

"Postgres doesnt scale well by default"... What?

How insane is it that a giant company making extreme use of a technology has to have a maintenance team for the said technology? Not insane inthe slightest its the most obvious conclusion possible, the fuck are you on XD

MongoDB is Webscale

Postgres also supports table sharding. It does require small changes on the application layer making the writes more complex, but the reads could stay pretty much the same. With that you could have back filling though

If they were using SQLite, they wouldn't have these problems.

The DBProxy blew my mind: We've heard you like postgres so we put a postgres in your postgres so you can query while you query

There's still time to delete this video Theo

How does Figma, a company with an online, almost real-time product that has tens (maybe hundreds) of thousands of concurrent users at a time, don't need to think about databases? The re definitely not just a "UI company" lol. The sheer size of the product is the only reason they stumble upon this issue. 99,999999% of digital companies will never have to face anything similar to this.

fail whale Theo move to Mysql? Skill issue. Move to Postgresql? Way to go. Especially when there's Yugabyte for full sharding which begs the question with Figma's blog post. Why haven't they mentioned Yugabyte at all in their eval of solutions before they decided to go Deadpool with maximum effort?

Figma: "Postgres does not scale" Supabase: "Hold my beer"

If we allow Theo will say figma will use tailwind css + next js and drop postgres and use localstorage

Or they could've just used the CockroachDB 🤷

Your example about Atomic-ness at 34:20, is essentially what was going on in the Post Office’s Horizon IT system. Scary.

"imagine if the average game dev had to care about assembly" - we used to, in 8-bit and 16-bit era :)

Damn that is impressive. Here we are at work just using sqlite 😂 if we had 1/10 of the scale of Figma I would just use CockroachDB as their replication and scaling story is pretty good. But at a certain scale it's more database engineering than product development and unless that's your jam I'm sure most engineers wouldn't want to do that.

This video was not sponsored by Citus.

I would love an analysis between what figma did and what Uber did which primogen talked about seems like they were tackling similar things would be great if we could get a deep dive on a compare and contrast

is putting years into scalling postgres really worth the effort instead of migrating to a different database type?

Never disagreed with you quite as much as this video. DBA is a real job for a real reason. Game devs DO need to care about machine level stuff because games are incredibly performance sensitive. Your devs should have to know their tools or have a team to ask about them if they're too complex. The role of DBA has not been automated yet. Also strikes me as weird that Figma chose to shard at the table level when one user is likely concerned with their own projects and nothing else, so they'll be doing a lot of joins across tables, but basically no aggregation over projects. Splitting by project, user, or group of users seems way more sensible. (Á la noSQL databases). Honestly I'd be curious about a relational solution for the user/org data, and have projects be in a noSQL database. E: re-read the article myself, they didn't simply put one table in one database and one table in another like you explained, my speculation is irrelevant.

There's a 100% chance that there's a video coming out next from Theo based on these comments 😄

another cloud service promotion video

Okay kid, that's dangerous misinformation. I'm a "newbie" database engineer but even I know PostgreSQL does scale very well. Worrying for my fellow "newbie" developers who are watching this, believing his BS, and being misled.

I wonder whether Figma would have been better suited with CQRS at this scale, and have more flexibility over how the query side is built. It doesn't eliminate the need to scale out, but it adds much more flexibility.

Gotta love haters, they know which fights to pickup..

Dang of all the takes it seems going after the elephant did you in 😂 Nikita from Neon DB would be a cool talk of how they scaled postgres

hello Citus. why isn't that mentioned in the first 2 minutes

{"Diskile" : 1, "notifications" 0, "follow": 0}

Will Figma become a DB company now as well?

Have they discovered... normalisation? I refuse to believe they had a single table with everything in it.

prolly going the face the same issue at my org soon. citus, a postgres extension is an attractive option does what figma did but better. not sure why it was not considered

I wonder why they didn't use Presto/Trino for the DbProxy.

Great video! Also really nice article by Figma.

*most* companies don't hit these problems.

The game Dev point is kinda off... The same way that figma is NOT the average scale of design app, most teams at studios pushing consoles are have a few people who always care about assembly, AND a few who care about DX / Opengl translation.

Postgres scales great, but i use Crunchy Data to scale it even more

looks like someone needs to run some benchmarks

IDK why but every time you say "sharding" my brain hears "sharting".

azure cosmos db for postgresql exists....

Theo is hyper fixated on the need for scale when in reality most companies just need customers

Postge´res my beloved

Isn't this basically an in-house DynamoDB that's more in tune with their business constraints? Basically it doesn't matter what db you use (Dynamo uses Mysql) but you need this key/sharding layer on top of it that have varying consistency guarantees depending on your use case.

Figma scaled their business by 100x but expect our free and opensource DB to have all of xyz requirements 😂 instead they could have contributed to the project, but no, just complain and post a blog. Well done self-entitled heroes of the industry.

05:58 cursor theme changed!?

I be horizontal sharting

It's overly complicated, ScyllaDB and it's resolved

Postgres / Citus / Spock DB wars ended ages ago 🤷‍♂️

Sponsored by...

FigJam > Excalidraw I'm right...

Not gonna get baited to comment on the "wHy ARe DbS noT bEtTer" angle. Nice try, though. Interesting insights into the frontend perspective though!

They‘re building a design tool so why do they need to think about databases… I expected more.

FE guys just don't want to get hands dirty with databases

They would scale much better if they used Nextjs for backend.

something something cassandra db something something

Havent watched the video and probably wont. I just come here for the comments seeing as Theo has the worst takes when it comes to anything outside JS. But after reading the blog I'm wondering if it would have been beneficial to have tried to also migrate the data to another database at the same time they're doing this. I'm particularly fond of cockroach DB and know for a fact it would have done what they needed to achieve. They've spent over 1.5 years and not even got to the stage of supporting cross transactions and built this big complex bit of software. But hey atleast its bought them job security for the next 5 years aha

Dear youtuber. Please highlight words as you read, or move ur move. You did a bit, but often not 🙏🏻

Just give every project a new database, should solve everything… then use Kafka or something for reporting rather than queries. You’re welcome 😂

Postgres my beloved

It's just absurd to listen to frontend guy talk such rot on postgres

Even figma? Bro....

Amazing article by Figma and an amazingly big L take by this clown

Wait so figma a design platform can't do contrast good 🥲🥲