Bristol

Insanely configurable logging for Node.js

Why another logger?

NPM has no shortage of loggers. Bristol was created to address a few common shortcomings in the current pool of options:

The message format should be decoupled from the message target (file, console, etc). You never know what system will be ingesting log files.
Structured logging should be the norm, and simple log functions should make that easy! No more shortcuts to util.format; if you're injecting values into a string, chances are they should be logged under their own keys.
No restrictions on data type. If you log an Error object, your logger should know what to do with that. Same goes for a string, an object, a date, as well as application-specific data types.
Lightweight! Why load modules that your configuration never uses? Bristol lazy-loads everything it can.
No callbacks, return values, or anything that would disincentivize a developer from throwing in a quick log message.

Those points and more drove the development of a brand new breed of logging library. Introducing Bristol.

Installation

In your project folder, type:

npm install bristol --save

Usage

Quick start

var log = require('bristol');
log.addTarget('console');

log.info("We're up and running!", {port: 3000});

Outputs (pretty-printed for README only):

{
    "message": "We're up and running!",
    "date": "2014-04-09 00:45:37",
    "severity": "info",
    "file": "/path/to/my/file.js",
    "line": "4",
    "port": 3000
}

Bristol can now be require()'d in any other file, and the settings will follow it.

Customizing the output

    log.addTarget('console')
        .withFormatter('human');

    log.debug("Hello, world", {event: 'bootstrap:welcome'});

Outputs:

[2014-04-09 00:53:59] DEBUG: Hello, world (/path/to/my/file.js:5)
    event: bootstrap:welcome

Available formatters

human: Multi-line messages suitable for humans during development
json: One line per message, valid json
syslog: Syslog-compliant format, single-line
commonInfoModel: Common Info Model format, friendly to many log aggregators. More human-readable than JSON with similar benefits. One line per message.

Available targets

console

Outputs directly to stdout using console.log() to ensure writes are blocking and synchronous. This is excellent for debugging, but is not recommended in production as this is not performant. There are no options for this target.

file

Streams output to a file using a logrotate-friendly WriteStream. Required options:

file string: The full path to the file to be created or opened

Severity levels

By default, Bristol provides error, warn, info, debug, and trace severity levels. These can be called as function names from the log object, or passed to the log function as the first parameter.

To change these levels to something else:

// List in order from MOST to LEAST severe:
log.setSeverities(['panic', 'omg', 'uhoh', 'crap', 'ok']);

// Functions matching the provided severities now exist
log.panic("WHAT ARE WE GOING TO DO?!");
log.log('omg', new Error("Something failed."));

// The old ones do not.
log.debug("This will throw an exception now");
log.log('trace', "So will this.");

Transforming your data types

Bristol allows you to send near-infinite arguments to the logging functions, and attempts to intelligently turn those into useful log messages. If your application has its own data structures, it can be useful to inform Bristol of them so that only pertinent values are logged if one of those is passed to a logging function.

log.addTransform(function(elem) {
    // This function should return data that should be logged instead of
    // the raw 'elem', or NULL if 'elem' isn't a type that we care about.
    if (elem.userName && elem.userId) {
        // elem is our enormous proprietary user object!
        return {
            username: elem.userName,
            connected: elem.lastSuccessfulConnection,
            ping: elem.getLastPingTime()
        };
    }
    return null;
});

// Now our "incomingConn" user object will just add its 3 most important
// properties to the resulting log message, rather than all of them.
log.info("New connection", incomingConn, {connections: server.getUserCount()});

With every log call, any transform functions are first called with an object parameter containing the file path of the calling script (.file) and the line number where the call originated (.line).

Setting global log values

Do you need certain pieces of information logged with every message?

log.setGlobal('hostname', require('os').hostname());
log.setGlobal('msg_uuid', function() {
    return uuid.v4();
});

log.info("This message contains both of those key/value pairs!");

Functions provided to setGlobal will be executed for every log message. Note that they're only called once per message even if multiple targets have been added; therefore, the uuid in the example above will be consistent across all configured targets.

Of course, you can also delete globals with:

log.deleteGlobal('msg_uuid');

Restricting targets to certain severities

// high_priority.log will contain only errors and warns
log.addTarget('file', {file: '/tmp/high_priority.log'})
    .withLowestSeverity('warn');

// debug.log will only store debug messages
log.addTarget('file', {file: '/tmp/debug.log'})
    .withLowestSeverity('debug')
    .withHighestSeverity('debug');

// The console will only output trace messages
log.addTarget('console')
    .withHighestSeverity('trace');

Restricting targets to certain types of messages

Sometimes, even within a severity level, it can be useful to filter out some kinds of log messages. Bristol leverages its key/value logging to allow you to blacklist or whitelist messages on a target, based on the values of certain keys.

// Don't log messages with event=disconnect, OR messages for certain channels
log.addTarget('console')
    .withFormatter('human')
    .excluding({
        event: 'disconnect',
        channel: ['#help', /^#anime.*$/]
    });

// Only log the trace messages with event starting with "test", from userId
// 1, 2, or 3, and with destination set to a test server.
log.addTarget('file')
    .withFormatter('commonInfoModel')
    .withHighestSeverity('trace')
    .onlyIncluding({
        event: /^test:/,
        userId: [1, 2, 3],
        destination: function(val) { return isTestServer(val); }

Restrictions on keys can be static types like strings or numbers, RegExp objects to check for a match, functions to test the value each time the target is hit, or arrays of any of the above to allow more than one match. Exclusions and inclusions can also be combined in one target to summon Captain Planet.

More than one logger

Have a use case that requires more than one 'log' object, so you can maintain different sets of targets? No problem:

var log2 = new log.Bristol();

The default log object will remain the default, but now log2 is a completely independent instance with zero configuration.

Extensions

Instead of passing in a target or formatter as a string, you can pass your own functions! Both of these modules are simply functions that take an options object as the first argument, and context-specific arguments after that. Check out some of the built-in targets and formatters for examples. They're super easy!

Testing

To test with full coverage report and enforcement of coverage percentage minimums:

npm test

For simple iterative testing, run just the mocha tests with:

npm run mocha

License

Bristol is distributed under the MIT license.

Credits

Bristol was created by Tom Shawver in 2014. Development of this library has been sponsored by TechnologyAdvice and Leadnomics.

TomFrost / Bristol

readme