At Stripe's scale, I doubt you could accurately forecast demand for a weekend like Black Friday. Last year's sales are just a data point; what matters this year is consumer confidence on a global macroeconomic scale. Who knows what that translates to?
> assuming their infra can scale quickly
Even if Stripe is handling spikes by scaling on a hyperscaler (e.g. AWS), this scale requires advanced coordination with the hyperscaler. They can't just spin up tens of thousands / hundreds of thousands of VM's on a whim, the underlying abstraction would leak and fail.
AWS uses predictors / models for this when you are on serverless and automatically allocates instances for you ahead of time.
No idea but ultimately this is marketing. If you wanna say we are just using as much as we need that's impressive to engineers but would it require or warrant a dedicated widget there that hovers around some sort of fixed utilization ratio?
Saying "look we are processing this much but could easily handle more" you are impressing a larger group. So my guess it's more about that.
In my experience, as much as you can distil it down to a simple number, we've aimed for 30-50 depending on capacity for risk.
Is 12% normal for stripe? Especially cautious knowing demand will spike this weekend? Feels pretty non-optimised, assuming their infra can scale quickly.