tink runs the bootstrapping / downloading of dependencies in the running node
process. This is cool, because you can clone any repo and just hit "npm start"
and it can immediately run the application. But it will need to write to disk
from the running process ... >

Something which might not be possible in serverless environments where you might
not even have a r/w disk. And if you scale an app using tink horizontally, all
instances have to re-fetch dependencies. >

That's why I like yarns approach, you run it once and can then even save the
cache folder to you application image. The task of fetching dependencies is a
one-time only "compile" step, and does not need to be executed for every
instance.