Testing is difficult at scale
I remember an article about WhatsApp (before it was acquired).
They said they can't really do any serious load-testing. What they ended up doing was to fade-in chances on a server-by-server basis and see how it played out.
WhatsApp had very few servers back then, even though it had 900 million users.
That said, because of the lack of a FB account, I couldn't care less about their outages. The sooner they go away, the better for humanity.