What is splitting

Splitlog reads changes from the watched log files line by line, as they appear. However, in order to provide some useful information here, it may group multiple lines into a single Message and infer some additional metadata from them. This process is called “splitting the log” and is what originally gave Splitlog its name.

In order to be able to separate the random set of lines into Messages, Splitlog needs to understand the boundaries of each message, and therefore have a good understanding of the file format. This is where the biggest limitations of Splitlog come from:

• It is not possible to parse most of logging files unambiguously. Let’s say we have a Logback logging pattern “%m%n”, where “%m” is a non-specific multi-line string and “%n” is a newline, used as a message separator. Where do messages end in this case? Which line ending is the end of the message, and which is just a continuation of “%m”?
• It is not possible to finalize a Message until we have seen the first line of the next one, or at least we are sure that the current line is the last.

This has multiple consequences:

1. There can be no generic splitter for Logback, JBoss Logging or what have you. The current JBossServerLogTailSplitter, for example, only supports the default JBoss log file layout.
2. Users have to write their own TailSplitter implementations for their particular logger patterns.

Message metadata

In cases where the splitter understands the log file, it can infer various metadata from them. There are various kinds of metadata that Splitlog is able to infer, such as:

• Message severity, such as ERROR, WARNING, INFO etc.
• The date when the line was recorded.
• Full stack trace of the exception, in case the Message carries one.

In case splitter can not infer any of these, it will try to provide some sensible defaults. If it couldn’t do even that, it will return null.

Available TailSplitter implementations

The most basic TailSplitter available is the SimpleTailSplitter and it is what you get when you obtain LogWatch through LogWatchBuilder without specifying any splitter implementation. This splitter:

• Treats every line as a new Message. Therefore even exception stack traces will not be merged into one Message.
• Infers no metadata.

The JBossServerLogTailSplitter is a bit more advanced and supports the default log file format of the JBoss Application Server. It will infer every possible metadata and properly parse exception stack traces. The following code snippet illustrates how to obtain it.

  LogWatch watch = LogWatchBuilder.getDefault()
      .watchingFile(...)
      .buildWith(new JBossServerLogTailSplitter());

The TailSplitter interface is fairly simple and users are encouraged to implement it to make Splitlog understand their particular log file formats. Pseudo-generic implementations for Logback and other logging systems would be welcome as pull requests.