How it works...

Even though the Debian/Ubuntu and Red Hat file layouts are different, they both follow the Linux Filesystem Hierarchy Standard (FHS), so neither layout is wrong.

The Red Hat layout is simpler and easier to understand. The Debian/Ubuntu layout is more complex, but it has different and more adventurous goals. The Debian/Ubuntu layout is similar to the Optimal Flexible Architecture (OFA) of other database systems. As pointed out earlier, the goals are to provide a file layout that will allow you to have multiple PostgreSQL database servers on one system, and to allow many versions of the software to exist in the filesystem at once.

Again, the layout for the Windows and OS X installers is different. Multiple database clusters are possible, but they are also more complex than on Debian/Ubuntu.

I recommend that you follow the Debian/Ubuntu layout on whichever platform you are using. It doesn't really have a name, so I call it the PostgreSQL Flexible Architecture (PFA). Clearly, if you are using Debian or Ubuntu, then the Debian/Ubuntu layout is already being used. If you do this on other platforms, you'll need to lay things out yourself, but it does pay off in the long run. To implement PFA, you can set the following environment variables to name parts of the file layout:

export PGROOT=/var/lib/pgsql/ 
export PGRELEASE=10
export PGSERVERNAME=mamba 
export PGDATA=$PGROOT/$PGRELEASE/$PGSERVERNAME

In this example, PGDATA is /var/lib/pgsql/10/mamba.

Finally, you must run initdb to actually initialize the data directory, as noted earlier, and custom administration scripts should be prepared to automate actions such as starting or stopping the database server when the system undergoes similar procedures.

Note that server applications such as initdb can only work with one major PostgreSQL version. On distributions that allow several major versions, such as Debian or Ubuntu, these applications are placed in dedicated directories, which are not put in the default command path. This means that if you just type initdb, the system will not find the executable, and you will get an error message.

This may look like a bug, but in fact it is the desired behavior. Instead of directly accessing initdb, you are supposed to use the pg_createcluster utility from postgresql-common, which will select the right initdb depending on the major version you specify.

If you plan to run more than one database server on the same host, you must set the preceding variables differently for each server, as they mandate the name of the data directory. For instance, you can set them in the script that you use to start or stop the database server, which would be enough because PGDATA is mostly used only by the database server process.