I heard that doordash ran a cluster serving >1M qps, there maybe still a setup at square/block where the team originated from

I guess most of the larger deployments are self-managed rather than their SaaS offering... Do you think this is mostly running on top of hyperscaler infrastructure? Or in traditional enterprise data centers? I'm just surprised because I never see it

It looks like[0] DoorDash is running theirs self-hosted and just on their cloud, not in the DC. At former gig we ran our much smaller clusters (largest one was ~10T and 20k qps) on kubernetes on cloud. Deployment/updates and what not wasn't difficult - the biggest difficulty was not accidentally overloading the DB from the application layer as always.

[0] - https://www.youtube.com/watch?time_continue=97&v=jCjrfpF64Kc