mutornadomon

µtornadomon is a library designed to be used with Tornado web applications. It adds an endpoint (/mutornadomon) to HTTP servers which outputs application statistics for use with standard metric collectors.

Usage

The monitor is initialized using mutornadomon.config.initialize_mutornadomon.

Exposing an HTTP endpoint

If you only pass a tornado web application, it will include request/response statistics, and expose an HTTP endpoint for polling by external processes:

from mutornadomon.config import initialize_mutornadomon
import signal

[...]

application = tornado.web.Application(...)
monitor = initialize_mutornadomon(application)

def shut_down(*args):
    monitor.stop()
    some_other_application_stop_function()
    tornado.ioloop.IOLoop.current().stop()

for sig in (signal.SIGQUIT, signal.SIGINT, signal.SIGTERM):
    signal.signal(sig, shut_down)

This will add a /mutornadomon endpoint to the web application.

Here is an example request to that endpoint:

$ curl http://localhost:8080/mutornadomon
{"process": {"uptime": 38.98995113372803, "num_fds": 8, "meminfo": {"rss_bytes": 14020608, "vsz_bytes": 2530562048}, "cpu": {"num_threads": 1, "system_time": 0.049356776, "user_time": 0.182635456}}, "max_gauges": {"ioloop_pending_callbacks": 0, "ioloop_handlers": 2, "ioloop_excess_callback_latency": 0.0006290912628173773}, "min_gauges": {"ioloop_pending_callbacks": 0, "ioloop_handlers": 2, "ioloop_excess_callback_latency": -0.004179096221923834}, "gauges": {"ioloop_pending_callbacks": 0, "ioloop_handlers": 2, "ioloop_excess_callback_latency": 0.0006290912628173773}, "counters": {"callbacks": 388, "requests": 6, "localhost_requests": 6, "private_requests": 6}}

If you want to add your own metrics, you can do so by calling the .kv() or .count() methods on the monitor object at any time.

The HTTP endpoint is restricted to only respond to request from loopback.

Providing a publishing callback

Alternatively, instead of polling the HTTP interface, you can pass in a publisher callback:

import pprint

def publisher(metrics):
    pprint.pprint(metrics)

monitor = initialize_mutornadomon(application, publisher=publisher)

By default, this will call the publisher callback every 10 seconds. To override this pass the publish_interval parameter (in miliseconds).

Monitoring non-web applications

If you don't pass an application object, other stats can still be collected:

import pprint

def publisher(metrics):
    pprint.pprint(metrics)

monitor = initialize_mutornadomon(publisher=publisher)

This only works with the publisher callback interface.