Where database blog posts get flame-broiled to perfection
Oh, this is just precious. "Making Metal's performance more accessible" by finally admitting the original $600 price tag was a fantasy only a VC-funded startup with more money than sense could afford. How magnanimous of them. I remember the all-hands where they unveiled the "Metal" roadmap. The slide deck had more rocket ships on it than a SpaceX launch, and the projections looked like they were drawn by a kid who’d just discovered the exponential growth function. We all just smiled and nodded, knowing the on-call rotation was about to become a living nightmare.
It’s cute that they’re still trotting out the same benchmark slides. You know, the ones where they tested against a competitor’s free-tier instance running on a Raspberry Pi in someone’s garage? The "drastic drops in latency" were real, I’ll give them that—mostly because we spent a month manually tuning the kernel parameters for the three customers they name-dropped, while everyone else was getting throttled by the real "secret sauce": aggressive cgroup limits.
But let’s talk about these new M-class clusters. An M-10 with 1/8th of an ARM vCPU. One-eighth! What is this, a database for ants? I can just picture the sales team trying to spin this. “It’s a fractional, paradigm-shifting, hyper-converged compute slice!” No, it’s a time-share on a single, overworked processor core. I hope you don't mind noisy neighbors, because you're about to have seven of them, all in the same microscopic apartment.
And this claim, my absolute favorite:
Unlimited I/O on every M- class means you can expect exceptional performance while your product grows.
Unlimited I/O. Bless their hearts. I still have PTSD from the "Project Unlimit" JIRA epic. That was a fun quarter. Let me translate this for you from corporatese to English: "Unlimited" means "we don't bill you for it directly." It does not mean the underlying EBS volume won't throttle you back to the stone age, or that the network card won't start dropping packets like a hot potato once you exceed the burst credits we forgot to mention. "Unlimited," in my experience there, usually meant "unlimited until the finance department sees the AWS bill, at which point it becomes very, very limited."
But the real gem, the little nugget that tells you everything you need to know about the state of the union, is buried right at the end.
"Smaller sizes are coming to Postgres first with smaller sizes for Vitess to follow. Our Vitess fleet is significantly larger than our Postgres fleet, so enabling smaller Metal sizes for Vitess will take more time."
Chef's kiss. This is magnificent. For anyone who hasn't spent years watching this particular sausage get made, let me break it down. What this actually says is:
So yes, by all means, get excited about what you can build on one-eighth of a CPU. I’m sure it’ll be great.
Anyway, thanks for the laugh. I promise you, I will not be reading the next one.