Skip to content

uber-common/larch

Repository files navigation

Larch

It's a logging framework. Currently it's main use is to reservoir sample logs and forward them to Logtron. In the long term, it may replace Logtron. It can be used like this:

var Logtron = require('logtron');

var Larch = require('larch');
var LogtronBackend = require('larch/logtron-backend');
var ReservoirBackend = require('larch/reservoir-backend');

var clients = ...
var logtronLogger = Logtron({...});

var logger = Larch({
    backends: [ReservoirBackend({
        backend: LogtronBackend(logtronLogger),
        statsd: clients.statsd,
        clusterStatsd: clients.clusterStatsd,
        size: 100,
        flushInterval: 50
    })]
});
logger.bootstrap();

logger.warn('warn string', {meta: 'object'});

Docs

Larch({ backends: Array })

Larch is a constructor that takes a list of backends.

Generally you want to use the ReservoirBackend

larch.bootstrap()

Remember to call bootstrap() on larch to start the reservoir.

ReservoirBackend(opts)

ReservoirBackend takes a set of options including:

  • backend; what to write to if we sample the log call
  • statsd; where to emit stats.
  • clusterStatsd; where to emit cluster-wide latency stats.
  • size; The amount of records that can be logged per interval
  • flushInterval; How often we should flush the reservoir

The reservoir will log (1000 / flushInterval) * size records per second.

This means by default it will log 2000 records per second.

Using willSample($level, $msg)

The top level Larch object, as well as each backend, has a method willSample. This method returns true if any backend after this backend in the object tree is interested in taking a log with level $level. This can be used to avoid allocating space for large meta objects, like so:

if (larch.willSample('warn', 'thing failed!')) {
    larch.swarn('thing failed!', {
        count: this.count,
        length: this.length,
        largeArrayOfThings: this.bigArray,
        hugeAmountOfUsefulDebuggingInfo: this.stuff
    });
}

This way, we can do less work when we have a lot of logs because we don't have to allocate the large meta object.

Regular log methods (.log, .error, etc) will first compute a sampling decision. Log methods prefixed with an s (.slog, .serror, etc) will use a previously computed sampling decision. Calling an s method without first calling .willSample($level, $msg) will throw.

Contributors

License

MIT.