Monday, September 06, 2010

The current state of logging frameworks

Logging is a commodity, which is argot to mean that I simply don't care as long as it gets out of the way. Recently I had to revisit my own logging decisions based on a single tip: Google Closure Compiler (the library we are using to minify our javascript files) was ignoring my logging setup.

It's like opening the closet you buried ten years ago when log4j was the one and only available choice.

First, the basics

Logging reliably since 1999

One thing is the logging framework (the thingie that writes stuff out):

  • log4j: Started by Ceki Gülcü back in the day, it was the framework used by the Java community until Sun decided to include a different logging implementation in the JRE. After Ceki split from the project, it has been mostly dead in the water.
  • Java Util Logging (JUL): First included with JRE 1.4.2, it is a simple (some would say naïve) implementation that includes just a handful of possible logging setups. You can extend it, but you should code it yourself.
  • Logback: It seems that Ceki Gülcü didn't stop after leaving the Apache Software Foundation. Logback is an implementation of the SLF4J API that can replace log4j or JUL.

A separate concept is the logging isolation library, because when you are implementing a library you cannot require (or expect) a specific logging framework:

  • Commons Logging: Supported by the Apache Software Foundation, this framework has been sadly well-known because of their memory leaks. At the time of this writing their last release dates back to 2007.
  • SLF4J: A logging API layer by Ceki Gülcü that includes adapters for Log4J, JUL, and almost anything you can think of. The adapter plugin mechanism is quite well-thought, just drop your jar in the classpath and you are ready to go.

The Google case


Google makes a point of not including external dependencies with their libraries, and when they do the package name is jarjar'ed to avoid version conflicts. This would not be needed in an ideal world, but as long as framework designers keep making backwards-incompatible changes I think it's a sane decision. Anyone that has used ASM or even Google Collections / Guava knows what I'm talking about (fun trivia fact: the earliest occurrence of this practice I can recall was inside Weblogic 8.1 in 2003, but my memory is not what I would call a trustable source :)

With this strategy logging becomes a problem because at some point you cannot keep renaming classes. This makes the JDK logging the only remaining alternative - for Google, that is.

The choice


A bit of history here: when log4j was out, it could write anywhere: JMS, Database, Email, the OS event system, Grandma's Cheesecacke Recipes Book... But that was soooo 1999. Today we have aspects, servlet filters and all kind of interceptor mechanisms that make these options obsolete. Of the handful times I have configured a log that goes anywhere other than the standard output, I have always ended up replacing it with a more elaborate solution.

"Zillions of appenders" is thus not exactly an important feature in my book. Performance, stability and support are. A quick search using Google pretty much confirms that the competition is over. SLF4J is being widely used by other frameworks, includes an awesome support (with releases every couple of months, the latest at the time of this writing was July 2010), no major known glitches, and fits perfectly with all the logging frameworks mentioned before.

About the logging framework... well, I will keep playing with JUL and Logback, but log4j is left behind as far as I am concerned.

UPDATE: You can also see a longer explanation by the SpringSource team here.

No comments:

Post a Comment

Something on your mind?

Note: Only a member of this blog may post a comment.