pyramid_metrics 0.3.1

Performance metrics for Pyramid using StatsD

Performance metrics for Pyramid using StatsD. The project aims at providing ways to instrument a Pyramid application in the least intrusive way.

• PyPI: https://pypi.python.org/pypi/pyramid_metrics

• Github: https://github.com/ludia/pyramid_metrics

• 

Installation

Install using setuptools, e.g. (within a virtualenv):

$ pip install pyramid_metrics

Setup

Once pyramid_metrics is installed, you must use the config.include mechanism to include it into your Pyramid project’s configuration. In your Pyramid project’s __init__.py:

config = Configurator(.....)
config.include('pyramid_metrics')

Alternately you can use the pyramid.includes configuration value in your .ini file:

[app:myapp]
pyramid.includes = pyramid_metrics

Usage

Pyramid_metrics configuration (values are defaults):

[app:myapp]
metrics.host = localhost
metrics.port = 8125

metrics.prefix = application.stage

metrics.route_performance = true

Route performance

If enabled, the route performance feature will time the request processing. By using the StatsD Timer type metric, pre-aggregation will provide information on latency, rate and total number. The information is sent two times: per route and globally.

The key name is composed of the route name, the HTTP method and the outcome (as HTTP status code or ‘exc’ for exception).

• Global key request.<HTTP_METHOD>.<STATUS_CODE_OR_EXC>

• Per route key route.<ROUTE_NAME>.request.<HTTP_METHOD>.<STATUS_CODE_OR_EXC>

API

Counter

StatsD type: https://github.com/etsy/statsd/blob/master/docs/metric_types.md#counting

# Increment a counter named cache.hit by 1
request.metrics.incr('cache.hit')

# Increment by N
request.metrics.incr(('cache.hit.read.total', count=len(cacheresult)))

# Stat names can be composed from list or tuple
request.metrics.incr(('cache', cache_action))

Gauge

StatsD type: https://github.com/etsy/statsd/blob/master/docs/metric_types.md#gauges

# Set the number of SQL connections to 8
request.metrics.gauge('sql.connections', 8)

# Increase the value of the metrics by some amount
request.metrics.gauge('network.egress', 34118, delta=True)

Timer

StatsD type: https://github.com/etsy/statsd/blob/master/docs/metric_types.md#timing

# Simple timing
time_in_ms = requests.get('http://example.net').elapsed.microseconds/1000
request.metrics.timing('net.example.responsetime', time_in_ms)

# Using the time marker mechanism
request.metrics.marker_start('something_slow')
httpclient.get('http://example.net')
request.metrics.marker_stop('something_slow')

# Measure different outcome
request.metrics.marker_start('something_slow')
try:
    httpclient.get('http://example.net').raise_for_status()
except:
    # Send measure to key 'something_slow.error'
    request.metrics.marker_stop('something_slow', suffix='error')
else:
    # Send measure to key 'something_slow.ok'
    request.metrics.marker_stop('something_slow', suffix='ok')

# Using the context manager
with request.metrics.timer(['longprocess', processname]):
   run_longprocess(processname)
   # Send measure to 'longprocess.foobar' or 'longprocess.foobar.exc'

Currently implemented

• Collection utility as a request method

• Ability to send metrics per Pyramid route

• Simple time marker mechanism

• Simple counter

• Context manager for Timing metric type

TODO

• Full StatsD metric types

• Extensions for automatic metrology (SQLAlchemy, MongoDB, Requests…)

• Whitelist/blacklist of metrics

• Time allocation per subsystem (using the time marker mechanism)

Considerations

• The general error policy is: always failsafe. Pyramid_metrics should NEVER break your application.

• The DNS resolution is done during configuration to avoid recurring latencies.

Development

Run tests

The tests are run by nose and all dependencies are in requirements-test.txt.

$ pip install -r requirements-test
...

$ nosetests
...

Run tests with tox

$ pip install tox
...

$ tox          # Run on python 2.7 and python 3.4
...

$ tox -e py34  # Run on python 3.4 only

Contributors

• Pior Bastida (@pior)

• Philippe Gauthier (@deuxpi)

• Hadrien David (@hadrien)

• Jay R. Wren (@jrwren)