Init System

An init system is the first process the kernel launches after boot, conventionally given process ID 1. Every other process on the machine is its descendant, directly or through reparenting when a parent dies. Its two irreducible jobs are starting the rest of userspace and reaping orphaned processes — nothing else can do either, because nothing else exists yet when it starts, and by default only PID 1 inherits a process whose parent has exited (Linux's subreaper flag lets a process claim that role for its own subtree, which is how container inits work).

The classic Unix init, sysvinit, ran numbered shell scripts in sequence, organized by runlevel 0 through 6 — halt, single-user, multi-user, and reboot. It was simple and legible but strictly serial: each script waited for the last, so boot time scaled with the number of services regardless of how many CPUs sat idle. It also had no native concept of a service depending on another beyond script ordering by filename.

Modern init systems — Systemd on most Linux distributions, launchd on macOS — instead describe services as unit declarations with explicit dependencies, and start independent ones in parallel. This trades the transparency of a shell script anyone can read top to bottom for faster boots and features like automatic restart, socket activation, and resource limits per service.

Whichever init a system runs, it remains the process that must never crash: its exit brings down the kernel with it, since there is no parent left to notice and no supervisor above PID 1.