Skip to content

Latest commit

 

History

History
93 lines (77 loc) · 3.91 KB

README.md

File metadata and controls

93 lines (77 loc) · 3.91 KB

What?

Liblogging is an easy to use library for logging. It offers an enhanced replacement for the syslog() call, but retains its ease of use.

If you dig deeper, liblogging actually has three components, which address different needs.

stdlog

This is the core component. Think of it as a the next version of the syslog(3) API. We retain the easy semantics, but make it more sophisticated "behind the scenes" with better support for multiple threads and flexibility for different log destinations. It also is signal-safe and thus can be used from within signal handlers.

Right now, it actually does more or less what syslog() did. In the next couple of weeks, this will change. It supports different log destionations:

  • syslog()
  • systemd journal native API
  • unix domain socket
  • files

The key point here is that we provide a separation of concerns: the application developer will do the logging calls, but the sysadmin will be able to configure where log messages actually are send to. We will use a driver layer to do so.

With the current version, the logging driver is set via an environment variable or directly specified from within the application code. The latter provides applications the ability to request specific logging drivers. At the smame time, a single application can log to multiple channels which in turn use multiple logging providers -- all at the same time.

This follows much the successful log4j paradigm. However, we have a focus on simplicity. No app developer likes logging, and so we really want to make it as easy, simple and unintrusive as syslog() was.

An interesting side-effect of that approach is that an application developer can write code that natively logs to the journal on platforms that support it but uses different methods on platforms that do not.

journalemu

This component (not yet committed) emulates the most important logging calls of systemd journal. This permits applications to be written to this spec, even though the journal is not available on all platforms. Of course, we recommend writing directly to libstdlog, as this solves the problem more cleanly.

rfc3195

This is liblogging's original component. Back in 2002, we thought that logging would be taken over by rfc3195 and thus we begun working on a library -liblogging- that makes this easy. While the lib is there, rfc3195 has turned out to be a big failure. This component is no longer enhanced, but it is still available for those apps that need it.

Motivation for this library

The syslog(3) API is the de-facto standard for application logging on Linux and Unix. While very simplistic, this is exactly the main feature the app developer wants. We want to keep this but improve it to

  • provide separation of concerns as described above under the stdlog component
  • support reentrancy
  • provide multiple concurrent log channels to different destinations
  • provide an interface for structured logging - but only if there is a real demand for it (to be seen based on feedback)

Liblogging's goal is also to be a very slim library without any notable memory or performance footprint (and also simple and small code).

Note to packagers

We recommend to create three different packages for this library, one for each component. As we do not anticipate new applications to use the rfc3195 component, we do not suggest building a package for it. On systems where systemd journal is present, there is hardly a point in packaging the journalemu component. So in esscence, the stdlog component is the only one we suggest to be packaged on many platforms.

Depending on distro policies, package names like liblogging1-stdlog are suggested. This prevents confusion with liblogging v0, which only supported rfc3195.

History

Liblogging is around since 2002. See HISTORY file for some background information on how it evolved and why it is structured like it currently is.