Managing Systems in an Age of Dynamic Complexity Or: Why Does My Single 2U Server Have Better Uptime than GCP?

A decade ago most systems administrators were running relatively static systems. We had servers which were provisioned and updated by hand, or maybe via a human invoking Puppet, Capistrano, or CFEngine. Instances were typically added to load balancer pools manually. New instances would be provisioned when administrators decided that more capacity was needed, and systems were sized for peak loads (plus a margin). Networks were configured by hand.