Awesome! Just checked the whitepaper there was a cluster with 300 machines, so all machines had to be described in configuration file that was used by master, right?

Mostly. Binaries of that era took everything on the command line. The config system would generate that from files and keep the running binaries up to date via ssh.