Permalink
Fetching contributors…
Cannot retrieve contributors at this time
119 lines (100 sloc) 6.07 KB

Big Picture

Elektra solves a non-trivial issue: how to abstract configuration in a way that software can be integrated and reconfiguration can be automated. Elektra solves this problem in a holistic way. Read why Elektra for an explanation of why such a solution is necessary.

Virtual File System Analogy

If you know virtual file systems, you already know a very similar solution to a very similar problem (otherwise first read about what a virtual file system is here).

Before file systems (or for devices without operating system) software simply wrote at discs (think of dd of=/dev/sda) to persistently store data. This obviously does not work with multiple applications. To allow multiple applications to access data in /dev/sda, a file system structures them in a way that every application knows where its bits are. In analogy configuration files are application-specific initialization that cannot be shared with other applications. So as first steps we need to have a "file system" that describes the content of a configuration file in a uniform way. In Elektra plugins represent file systems: they know how the data in configuration files should be interpreted.

For file systems, the API is open, read, write, and then close. For configuration key-value access is more suitable because values are so small that a single read/write always suffices. Thus Elektra has a key-value API with kdbOpen, kdbGet, kdbSet and kdbClose. Not every application is written in C, thus many bindings where written to access file systems. For example, in C++ you have fstream, and in Java FileReader. Also Elektra provides different bindings: In C++ you have a class KDB and can use kdb.get or kdb.set. Furthermore, every language has native support for the language's iterators which make Elektra easier to use.

Furthermore, command-line tools like cat and ls provide an additional interface to the content of file systems for users and administrators. In Elektra also command-line tools for the analog purpose exist: With kdb cp, kdb mv, and kdb ls some command-line tools operating on file systems are mimicked.

There is not one file system satisfying every need, e.g., proc, tmpfs, nfs and ext4 have quite different use cases and are needed at the same time in parallel. In analogy, different configuration file format (parsers) have different advantages and disadvantages. For example, /etc/passwd or /etc/hosts are structured and very compact, while many other configuration files are semi-structured.

To have multiple file systems present at the same time a virtual file system is able to mount (2) concrete file systems and thus give applications a way to uniformly access them. Similarly, Elektra also implements a mount functionality in its core. In Elektra a contract specifies the obligations between plugins.

Sometimes, it is even useful to have multiple file systems at the same mountpoint, so called stacked file systems. They allow you to combine features of different file systems. For example, eCryptfs allows you to encrypt directories or files. In Elektra, stacking plugins is a core feature and heavily used to avoid feature-bloated plugins. For example, the crypto plugin allows you to encrypt individual keys or the iconv plugin to change the character encoding.

In file systems metadata describes information about files, e.g. when they were last accessed and who they are owned by (ls -l). In the same way Elektra has metadata that describe individual key-value pairs. In Elektra metadata is defined globally but implemented in many plugins.

Implementations of file systems is not an easy task. The idea of FUSE (Filesystem in Userspace) is to make file system development easier by having the conveniences of userspace together with a helper library libfuse. In particular this allowed developers to use any programming language and easier abstractions. Elektra also tries hard to make plugin development simple. For example, special interpreter plugins enable developers to also write plugins in different languages. Furthermore, other libraries also assist in creating plugins.

Of course not every feature of virtual file systems or Elektra has an analogy in the other system. If they would solve the same problem, one of them would be useless. Main differences are:

  • API (get/set vs. read/write)
  • commit semantics: one kdbSet can change many configuration files atomically. This is important if you want, e.g., a new host in /etc/hosts and use this host in some other configuration files.
  • namespaces there are many places where the same configuration is stored. All of these configuration files have the same semantics and they override each other (think of command-line arguments, /etc, $HOME/.config, ...)
  • Elektra interacts closely with the program execution environment such as command-line parsing. The namespace proc is specifically reserved for this purpose.
  • in Elektra it is possible to create holes (files without directories above them) which are needed because of these override semantics: we want to be able to override a single value without duplicating the whole skeleton.
  • validation: in Elektra you can describe how valid configuration should look like and reject invalid configuration.
  • and much more...

Further pointers