I really admire any DB that helps with zero effort scaling. I have immense respect for engineers that solve DB scaling problems. However i have to say this: The last DBaaS we used was FoundationDB. They left us and many others hanging dry. Ever since our blind choice has been AWS RDS Multi-AZ.
What does PlanetScale use? Is it a totally new engine? FoundationDB is rock solid. The only thing that can come close is some projects based on RocksDB, like TiDB. It would be great to see some comparison.
PlanetScale is built on Vitess ( https://vitess.io ) which actually uses MySQL as the backing store, and Vitess has been deployed at Youtube, Slack, and a bunch of other companies for years ( the first commits date back to September 2010 ). It is just about as rock solid as you can get.
Hey I am on the team at PlanetScale and you are 100% correct about that. One of the great decisions we made was that our SaaS is the OSS version of Vitess so even if we went away you would still be able to continue with Vitess. Also we are working on rolling out BYOK ( bring your own Kubernetes ) as a deploy option so the databases and Vitess components all reside in your Account not in ours.
It is Apache 2 and thats the only version we will deploy, there isn't a Proprietary version that planetscale will use in the background. This isn't a fork or even a re-licensed version of Vitess. Everything we develop goes into the Apache 2 OSS project
Vitess solves a lot of the complex issues with DB scaling like horizontal partitioning and being able to use it as a complement to AWS could end up being a really nice solution.
p.s. got to meet some of the team and yeah they're super impressive.
It seems like you'd like me more if I sold proprietary software.
It is pretty disingenuous to call Open Source code an advertisement, I don't profit off of it.
An exponential amount of people from HN have used GUN than I've gotten downvotes, so yes, I keep hanging out here. Scaling databases is pretty up the alleyway.
The product is drop in compatible with MySQL so if you are using MySQL today its likely you would see some improvement. Also if you have large data sets or if you want a Highly Available database running alongside your applications in Kubernetes it would be a good fit for you. That last case is extremely powerful. You can treat your Database as a first class k8s citizen right along side your applications and have a performant scaleable, and highly available database with no more management then what you would give any k8s application.
The scaleout relational db world is definitely interesting. There's about a dozen competitors now consolidating around mysql and postgres dialects with both strategies of building on top of the native engine and building a new one to emulate it.
What does the Planetscale team see as the main differentiation with these other companies?
That was a talk sponsored by the EKS team at Amazon, we ( at planetscale ) were working with them and our partner to show off how well we could scale, but since it was part of the EKS track we wanted to highlight the details of Kubernetes.
26 comments
[ 4.2 ms ] story [ 25.0 ms ] threadI think many new databases are handling the concerns around lock-in by being Postgres or MySQL compatible. PlaneScale is MySQL compatible of course.
I'm curious what happened to make people feel left "hanging dry"? Guessing it's how the DBaaS was shut down.
https://news.ycombinator.com/item?id=16877395
Isn't Vitess already open source (Apache 2.0 from their site)? Or did you mean to say managed/hosted/DBaaS version of Vitess?
See: https://github.com/vitessio/vitess/issues/2518
Vitess solves a lot of the complex issues with DB scaling like horizontal partitioning and being able to use it as a complement to AWS could end up being a really nice solution.
p.s. got to meet some of the team and yeah they're super impressive.
Seems like that is how Silicon Valley tech startups grow these days. :/
We have an Open Source "deploy your own" DBaaS that we're testing scaling:
- Have hit 8,000 real world concurrent peers on 70MB memory.
- Backups to S3 for $0.01/month & runs on a free Heroku instance.
Current stats: https://guntest.herokuapp.com/stats.html
Going to test live in-production write load next.
The goal is that it shards across a bunch of t3 micros.
Anybody have thoughts / tips on different types of scaling tests to run?
What is the minimum level of scale/performance that people need?
How many concurrent users on a machine do you think you can squeeze before you have to pay any real/serious money? More than 100K? 1M?
Usually his self-advertisments are triggered by anything P2P/distributed related, so this one seems like a big stretch even for him.
It is pretty disingenuous to call Open Source code an advertisement, I don't profit off of it.
An exponential amount of people from HN have used GUN than I've gotten downvotes, so yes, I keep hanging out here. Scaling databases is pretty up the alleyway.
Cheers!
What does the Planetscale team see as the main differentiation with these other companies?
https://www.slideshare.net/AmazonWebServices/running-a-highp...
(Vitess is barely mentioned on slide 28 but that's what the demo was running on, not sure why)