I wonder why other providers don't use metal ssd sync replication technique that planetscale uses? Most of them just default to EBS.
My interest in it peaked when I heard about NVMe over Fabrics (NVMe/TCP) and SPDK from Xata[1] and apparently with that performance is as good as planetscale metal, but planetscale found their methodology flawed[2] and they Xata never responded.
[1] https://xata.io/blog/reaction-to-the-planetscale-postgresql-...
It's very hard to do. They all want to do it but can't so now it's their marketing team's jobs to lie to people about why they shouldn't want it.
@samlambert what exactly makes it hard? Isn't it as simple as setting synchronous_commit=remote_apply or does planetscale have a custom strategy or are there other operational issues?
Just asking since I find it both the planetscale's engineering and its impact on competitive landscape very interesting.
you have to make sure you will never terminate these nodes, that you have all the operations maturity to cycle them responsibly, and resize them. I am sure they will get there one day but most people are still figuring out how to run databases on k8s so it's a long road.
I'm honestly confused why local nvme disks haven't become the standard for cloud offerings a long time ago.
Aiven (not working for them, just a happy client) started offering local nvme disks for their postgres service in 2017. (https://aiven.io/blog/larger-and-faster-aiven-postgresql-pla...)
Back then I was sure it was only a matter of time for other hosted database providers to move on from EBS. But until Planetscale made a lot of noise about Metal no one seemed to bother.