[gioazzi]

Provider iteration is a really nice one - I had a big monorepo that would deploy some baseline services in many AWS accounts, across multiple regions, generating tf.json files for each provider to match all accounts that were created.

However, what really broke this model at some point was the fact that we were running so many providers instances that our Terraform Cloud would go out of memory! Since each provider instance in tf is really launching a new process it really adds up... At some point I was thinking since the engine and the providers use gRPC to communicate, it MAY be possible to distribute providers across machines, but I never investigated it further... I'm pretty sure there was a notice in the tf plugin SDK stating that it was not possible to connect them over a network... but why not? ¯\_(ツ)_/¯

[cube2222]

Yeah, esp. the AWS provider is pretty memory-intensive.

I believe someone on the team did some investigation into this (running providers remotely) but it's not really a priority (if it is for you, feel free to voice that on the issue tracker!).

Frankly though, with pricing for cloud instances being generally linear wrt to the CPU/memory size of the instance, I don't think there's much reason to prefer many smaller machines over just using a larger single one and avoiding all this added complexity.