One of the things about the Parler thing is that running a system at scale in prod will have different scaling bottlenecks on different environments, so even if you could get a replacement up in a day or two (you can't) it wouldn't perform the same, and would fail in new ways

and you can't know what new ways the system will fail in advance, because the causes of those failures, the underlying bottlenecks in the environment, are opaque until you try to scale on that environment

Another important point I haven't seen talked about is that Parler is the sort of client that you don't want in your infra - it's a huge target. Once you become an upstream for an adversarial service you start getting your own attacks. Most providers won't have that resilience.

All of this is why I find it so hilarious.

Once you get to Scale, implementation details of your underlying infra matter *a whole lot*. And if your system is built on one set of implementation details (AWS, in this case) then your ability to change the infra is radically constrained.

Some examples of this could be:
- Internal network speed will be different, which will require different kinds of caching
- SAN disk speeds will be different, so database performance will be different and require different tunings
- Uplink performance will be different
- Etcetera

Notice i'm not even talking about high-level primitives like load balancing, and how hard it is to build that to scale in a new environment.

Just implementation detail of the vendor network can have huge impacts when you're at scale.