Centos – SysV init run levels

When you switch runlevel, the only things executed are the scripts in /etc/rc.d/rc${NEW_LEVEL}.d/.

This means that you are right: Every rc*.d directory needs to be able to handle all of the process/service changes when switching from another runlevel. So every rc directory contains a full set of scripts for reaching that runlevel.

Say you're switching to runlevel 3. The /etc/rc.d/rc3.d/K* scripts will attempt to kill any processes that were running in whatever your previous runlevel was (could be any number), and /etc/rc.d/rc3.d/S* scripts will start any processes that need to be started (and weren't already started in the previous runlevel).

Clearly, managing all of these symlinks would be a real pain, so there are utilities to help manage this. On Debian and Ubuntu (at least, maybe others), you can use update-rc.d to selectively enable/disable scripts found in /etc/init.d, or to set them up to "default" or recommended settings for each script. This will create and update all the symlinks for you, to reflect any configuration changes you want to put in place. On CentOS, I understand you can use ntsysv or chkconfig to do the same thing.

Effectively, you never touch the files in /etc/rc*.d/ (or /etc/rc.d/rc*.d/) yourself; you always use the tool (eg. update-rc.d, ntsysv, chkconfig) to make any changes.

In Ubuntu, this is called Upstart. I suggest you read the Cookbook, it discusses how to create scripts to run at a specific (pseudo) runlevel.

What you're refering to, is the "System V" init system. The idea is to have things started concurrently. I suggest you take a look at the instance functionality of Upstart.