A Python structured logger for Fluentd

Many web/mobile applications generate huge amount of event logs (c,f. login, logout, purchase, follow, etc). To analyze these event logs could be really valuable for improving the service. However, the challenge is collecting these logs easily and reliably.

Fluentd solves that problem by having: easy installation, small footprint, plugins, reliable buffering, log forwarding, etc.

fluent-logger-python is a Python library, to record the events from Python application.

Requirements

• Python 2.6 or greater including 3.x

Installation

This library is distributed as 'fluent-logger' python package. Please execute the following command to install it.

$ pip install fluent-logger

Configuration

Fluentd daemon must be launched with a tcp source configuration:

<source>
type forward
port 24224
</source>

To quickly test your setup, add a matcher that logs to the stdout:

<match app.**>
type stdout
</match>

Usage

Event-Based Interface

First, you need to call logger.setup() to create global logger instance. This call needs to be called only once, at the beggining of the application for example.

By default, the logger assumes fluentd daemon is launched locally. You can also specify remote logger by passing the options.

fromfluentimportsender# for local fluentsender.setup('app')
# for remote fluentsender.setup('app', host='host', port=24224)

Then, please create the events like this. This will send the event to fluent, with tag 'app.follow' and the attributes 'from' and 'to'.

fromfluentimportevent# send event to fluentd, with 'app.follow' tagevent.Event('follow', {
'from': 'userA',
'to': 'userB'
})

Python logging.Handler interface

This client-library also has FluentHandler class for Python logging module.

importloggingfromfluentimporthandlercustom_format= {
'host': '%(hostname)s',
'where': '%(module)s.%(funcName)s',
'type': '%(levelname)s',
'stack_trace': '%(exc_text)s'
}
logging.basicConfig(level=logging.INFO)
l=logging.getLogger('fluent.test')
h=handler.FluentHandler('app.follow', host='host', port=24224)
formatter=handler.FluentRecordFormatter(custom_format)
h.setFormatter(formatter)
l.addHandler(h)
l.info({
'from': 'userA',
'to': 'userB'
})
l.info('{"from": "userC", "to": "userD"}')
l.info("This log entry will be logged with the additional key: 'message'.")

You can also customize formatter via logging.config.dictConfig

importlogging.configimportyamlwithopen('logging.yaml') asfd:
conf=yaml.load(fd)
logging.config.dictConfig(conf['logging'])

A sample configuration logging.yaml would be:

logging:
version: 1formatters:
brief:
format: '%(message)s'default:
format: '%(asctime)s %(levelname)-8s %(name)-15s %(message)s'datefmt: '%Y-%m-%d %H:%M:%S'fluent_fmt:
'()': fluent.handler.FluentRecordFormatterformat:
level: '%(levelname)s'hostname: '%(hostname)s'where: '%(module)s.%(funcName)s'handlers:
console:
class : logging.StreamHandlerlevel: DEBUGformatter: defaultstream: ext://sys.stdoutfluent:
class: fluent.handler.FluentHandlerhost: localhostport: 24224tag: test.loggingformatter: fluent_fmtlevel: DEBUGnull:
class: logging.NullHandlerloggers:
amqp:
handlers: [null]
propagate: Falseconf:
handlers: [null]
propagate: False'': # root loggerhandlers: [console, fluent]
level: DEBUGpropagate: False

Testing

Testing can be done using nose.

Contributors

Patches contributed by those people.

License

Apache License, Version 2.0