Skip to content

Latest commit

 

History

History
276 lines (191 loc) · 7.32 KB

README.rst

File metadata and controls

276 lines (191 loc) · 7.32 KB

graypy

PyPI Status Build Status Documentation Status Coverage Status

Description

Python logging handlers that send log messages in the Graylog Extended Log Format (GELF).

graypy supports sending GELF logs to both Graylog2 and Graylog3 servers.

Installing

Using pip

Install the basic graypy python logging handlers:

pip install graypy

Install with requirements for GELFRabbitHandler:

pip install graypy[amqp]

Using easy_install

Install the basic graypy python logging handlers:

easy_install graypy

Install with requirements for GELFRabbitHandler:

easy_install graypy[amqp]

Usage

graypy sends GELF logs to a Graylog server via subclasses of the python logging.Handler class.

Below is the list of ready to run GELF logging handlers defined by graypy:

  • GELFUDPHandler - UDP log forwarding
  • GELFTCPHandler - TCP log forwarding
  • GELFTLSHandler - TCP log forwarding with TLS support
  • GELFHTTPHandler - HTTP log forwarding
  • GELFRabbitHandler - RabbitMQ log forwarding

UDP Logging

UDP Log forwarding to a locally hosted Graylog server can be easily done with the GELFUDPHandler:

import logging
import graypy

my_logger = logging.getLogger('test_logger')
my_logger.setLevel(logging.DEBUG)

handler = graypy.GELFUDPHandler('localhost', 12201)
my_logger.addHandler(handler)

my_logger.debug('Hello Graylog.')

UDP GELF Chunkers

GELF UDP Chunking is supported by the GELFUDPHandler and is defined by the gelf_chunker argument within its constructor. By default the GELFWarningChunker is used, thus, GELF messages that chunk overflow (i.e. consisting of more than 128 chunks) will issue a GELFChunkOverflowWarning and will be dropped.

Other gelf_chunker options are also available:

  • BaseGELFChunker silently drops GELF messages that chunk overflow
  • GELFTruncatingChunker issues a GELFChunkOverflowWarning and simplifies and truncates GELF messages that chunk overflow in a attempt to send some content to Graylog. If this process fails to prevent another chunk overflow a GELFTruncationFailureWarning is issued.

RabbitMQ Logging

Alternately, use GELFRabbitHandler to send messages to RabbitMQ and configure your Graylog server to consume messages via AMQP. This prevents log messages from being lost due to dropped UDP packets (GELFUDPHandler sends messages to Graylog using UDP). You will need to configure RabbitMQ with a gelf_log queue and bind it to the logging.gelf exchange so messages are properly routed to a queue that can be consumed by Graylog (the queue and exchange names may be customized to your liking).

import logging
import graypy

my_logger = logging.getLogger('test_logger')
my_logger.setLevel(logging.DEBUG)

handler = graypy.GELFRabbitHandler('amqp://guest:guest@localhost/', exchange='logging.gelf')
my_logger.addHandler(handler)

my_logger.debug('Hello Graylog.')

Django Logging

It's easy to integrate graypy with Django's logging settings. Just add a new handler in your settings.py:

LOGGING = {
    'version': 1,
    # other dictConfig keys here...
    'handlers': {
        'graypy': {
            'level': 'WARNING',
            'class': 'graypy.GELFUDPHandler',
            'host': 'localhost',
            'port': 12201,
        },
    },
    'loggers': {
        'django.request': {
            'handlers': ['graypy'],
            'level': 'ERROR',
            'propagate': True,
        },
    },
}

Traceback Logging

By default log captured exception tracebacks are added to the GELF log as full_message fields:

import logging
import graypy

my_logger = logging.getLogger('test_logger')
my_logger.setLevel(logging.DEBUG)

handler = graypy.GELFUDPHandler('localhost', 12201)
my_logger.addHandler(handler)

try:
    puff_the_magic_dragon()
except NameError:
    my_logger.debug('No dragons here.', exc_info=1)

Default Logging Fields

By default a number of debugging logging fields are automatically added to the GELF log if available:

  • function
  • pid
  • process_name
  • thread_name

You can disable automatically adding these debugging logging fields by specifying debugging_fields=False in the handler's constructor:

handler = graypy.GELFUDPHandler('localhost', 12201, debugging_fields=False)

Adding Custom Logging Fields

graypy also supports including custom fields in the GELF logs sent to Graylog. This can be done by using Python's LoggerAdapter and Filter classes.

Using LoggerAdapter

LoggerAdapter makes it easy to add static information to your GELF log messages:

import logging
import graypy

my_logger = logging.getLogger('test_logger')
my_logger.setLevel(logging.DEBUG)

handler = graypy.GELFUDPHandler('localhost', 12201)
my_logger.addHandler(handler)

my_adapter = logging.LoggerAdapter(logging.getLogger('test_logger'),
                                   {'username': 'John'})

my_adapter.debug('Hello Graylog from John.')

Using Filter

Filter gives more flexibility and allows for dynamic information to be added to your GELF logs:

import logging
import graypy

class UsernameFilter(logging.Filter):
    def __init__(self):
        # In an actual use case would dynamically get this
        # (e.g. from memcache)
        self.username = 'John'

    def filter(self, record):
        record.username = self.username
        return True

my_logger = logging.getLogger('test_logger')
my_logger.setLevel(logging.DEBUG)

handler = graypy.GELFUDPHandler('localhost', 12201)
my_logger.addHandler(handler)

my_logger.addFilter(UsernameFilter())

my_logger.debug('Hello Graylog from John.')

Contributors

  • Sever Banesiu
  • Daniel Miller
  • Tushar Makkar
  • Nathan Klapstein