Census for Python. Census provides a framework to measure a server's resource usage and collect performance stats. This repository contains Python related utilities and supporting software needed by Census.
Install the opencensus-trace package using pip or pipenv:
pip install opencensus pipenv install opencensus
Initialize a tracer for your application:
from opencensus.trace import tracer as tracer_module tracer = tracer_module.Tracer()
If you want to use the unreleased packages (like stats and tags), you need to build the package from source using the below commands: (The stats and tags packages are expected to be released in 0.1.6)
git clone https://github.com/census-instrumentation/opencensus-python.git cd opencensus-python python setup.py bdist_wheel pip install dist/*
You can collect traces using the Tracer
context manager:
from opencensus.trace import tracer as tracer_module
# Initialize a tracer, by default using the `PrintExporter`
tracer = tracer_module.Tracer()
# Example for creating nested spans
with tracer.span(name='span1') as span1:
do_something_to_trace()
with span1.span(name='span1_child1') as span1_child1:
do_something_to_trace()
with span1.span(name='span1_child2') as span1_child2:
do_something_to_trace()
with tracer.span(name='span2') as span2:
do_something_to_trace()
Census will collect everything within the with
statement as a single span.
Alternatively, you can explicitly start and end a span:
from opencensus.trace import tracer as tracer_module
# Initialize a tracer, by default using the `PrintExporter`
tracer = tracer_module.Tracer()
tracer.start_span(name='span1')
do_something_to_trace()
tracer.end_span()
You can specify different samplers when initializing a tracer, default
is using AlwaysOnSampler
, the other options are AlwaysOffSampler
and ProbabilitySampler
from opencensus.trace.samplers import probability
from opencensus.trace import tracer as tracer_module
# Sampling the requests at the rate equals 0.5
sampler = probability.ProbabilitySampler(rate=0.5)
tracer = tracer_module.Tracer(sampler=sampler)
You can choose different exporters to send the traces to. By default, the traces are printed to stdout in JSON format. Other options include writing to a file, sending to Python logging, or reporting to Stackdriver.
This example shows how to configure Census to save the traces to a file:
from opencensus.trace.exporters import file_exporter
from opencensus.trace.tracers import context_tracer
exporter = file_exporter.FileExporter(file_name='traces')
tracer = context_tracer.ContextTracer(exporter=exporter)
This example shows how to report the traces to Stackdriver Trace:
from opencensus.trace.exporters import stackdriver_exporter
from opencensus.trace import tracer as tracer_module
exporter = stackdriver_exporter.StackdriverExporter(
project_id='your_cloud_project')
tracer = tracer_module.Tracer(exporter=exporter)
StackdriverExporter requires the google-cloud-trace package. Install google-cloud-trace using pip or pipenv:
pip install google-cloud-trace pipenv install google-cloud-trace
By default, traces are exported synchronously, which introduces latency during your code's execution. To avoid blocking code execution, you can initialize your exporter to use a background thread.
This example shows how to configure Census to use a background thread:
from opencensus.trace.exporters import stackdriver_exporter
from opencensus.trace import tracer as tracer_module
from opencensus.trace.exporters.transports.background_thread \
import BackgroundThreadTransport
exporter = stackdriver_exporter.StackdriverExporter(
project_id='your_cloud_project', transport=BackgroundThreadTransport)
tracer = tracer_module.Tracer(exporter=exporter)
You can specify the propagator type for serializing and deserializing the
SpanContext
and its headers. There are currently three built in propagators:
GoogleCloudFormatPropagator
, TextFormatPropagator
and TraceContextPropagator
.
This example shows how to use the GoogleCloudFormatPropagator
:
from opencensus.trace.propagation import google_cloud_format
propagator = google_cloud_format.GoogleCloudFormatPropagator()
# Deserialize
span_context = propagator.from_header(header)
# Serialize
header = propagator.to_header(span_context)
This example shows how to use the TraceContextPropagator
:
import requests
from opencensus.trace import config_integration
from opencensus.trace.propagation.trace_context_http_header_format import TraceContextPropagator
from opencensus.trace.tracer import Tracer
config_integration.trace_integrations(['httplib'])
tracer = Tracer(propagator = TraceContextPropagator())
with tracer.span(name = 'parent'):
with tracer.span(name = 'child'):
response = requests.get('http://localhost:5000')
You can specify which paths you do not want to trace by configuring the blacklist paths.
This example shows how to configure the blacklist to ignore the _ah/health endpoint for a Flask application:
from opencensus.trace.ext.flask.flask_middleware import FlaskMiddleware
app = flask.Flask(__name__)
blacklist_paths = ['_ah/health']
middleware = FlaskMiddleware(app, blacklist_paths=blacklist_paths)
For Django, you can configure the blacklist in the OPENCENSUS_PARAMS
in settings.py
:
OPENCENSUS_PARAMS: {
...
'BLACKLIST_PATHS': ['_ah/health',],
}
Note
By default, the health check path for the App Engine flexible environment is not traced, but you can turn it on by excluding it from the blacklist setting.
Census supports integration with popular web frameworks including Django, Flask, Pyramid, and Webapp2. When the application receives a HTTP request, the tracer will automatically generate a span context using the trace information extracted from the request headers and propagated to the child spans.
In your application, use the middleware to wrap your app and the requests will be automatically traced.
from opencensus.trace.ext.flask.flask_middleware import FlaskMiddleware
app = flask.Flask(__name__)
# You can also specify the sampler, exporter, propagator in the middleware,
# default is using `AlwaysOnSampler` as sampler, `PrintExporter` as exporter,
# `GoogleCloudFormatPropagator` as propagator.
middleware = FlaskMiddleware(app)
For tracing Django requests, you will need to add the following line to
the MIDDLEWARE_CLASSES
section in the Django settings.py
file.
MIDDLEWARE_CLASSES = [
...
'opencensus.trace.ext.django.middleware.OpencensusMiddleware',
]
And add this line to the INSTALLED_APPS
section:
INSTALLED_APPS = [
...
'opencensus.trace.ext.django',
]
You can configure the sampler, exporter, propagator using the OPENCENSUS_TRACE
setting in
settings.py
:
OPENCENSUS_TRACE = {
'SAMPLER': 'opencensus.trace.samplers.probability.ProbabilitySampler',
'EXPORTER': 'opencensus.trace.exporters.print_exporter.PrintExporter',
'PROPAGATOR': 'opencensus.trace.propagation.google_cloud_format.'
'GoogleCloudFormatPropagator',
}
You can configure the sampling rate and other parameters using the OPENCENSUS_TRACE_PARAMS
setting in settings.py
:
OPENCENSUS_TRACE_PARAMS = {
'BLACKLIST_PATHS': ['/_ah/health'],
'GCP_EXPORTER_PROJECT': None,
'SAMPLING_RATE': 0.5,
'SERVICE_NAME': 'my_service',
'ZIPKIN_EXPORTER_HOST_NAME': 'localhost',
'ZIPKIN_EXPORTER_PORT': 9411,
'ZIPKIN_EXPORTER_PROTOCOL': 'http',
}
In your application, add the pyramid tween and your requests will be traced.
def main(global_config, **settings):
config = Configurator(settings=settings)
config.add_tween('opencensus.trace.ext.pyramid'
'.pyramid_middleware.OpenCensusTweenFactory')
To configure the sampler, exporter, and propagator, pass the instances into the pyramid settings
from opencensus.trace.exporters import print_exporter
from opencensus.trace.propagation import google_cloud_format
from opencensus.trace.samplers import probability
settings = {}
settings['OPENCENSUS_TRACE'] = {
'EXPORTER': print_exporter.PrintExporter(),
'SAMPLER': probability.ProbabilitySampler(rate=0.5),
'PROPAGATOR': google_cloud_format.GoogleCloudFormatPropagator(),
}
config = Configurator(settings=settings)
OpenCensus provides the implementation of interceptors for both the client side and server side to instrument the gRPC requests and responses. The client interceptors are used to create a decorated channel that intercepts client gRPC calls and server interceptors act as decorators over handlers.
gRPC interceptor is a new feature in the grpcio1.8.0 release, please upgrade your grpcio to the latest version to use this feature.
For sample usage, please refer to the hello world example in the examples directory.
More information about the gRPC interceptors please see the proposal.
Opencensus supports integration with various popular outbound services such as SQL packages, Requests and Google Cloud client libraries. To enable integration services to census: you will need to pass the list of services to census:
from opencensus.trace import config_integration
from opencensus.trace import tracer as tracer_module
import mysql.connector
# Trace both mysql-connection and psycopg2
integration = ['mysql', 'postgresql']
config_integration.trace_integrations(integration)
The integration with MySQL supports the mysql-connector library and is specified
to trace_integrations
using 'mysql'
.
The integration with PostgreSQL supports the psycopg2 library and is specified
to trace_integrations
using 'postgresql'
.
You can trace usage of the sqlalchemy package, regardless of the underlying
database, by specifying 'sqlalchemy'
to trace_integrations
.
Note
If you enable tracing of SQLAlchemy as well as the underlying database driver, you will get duplicate spans. Instead, just trace SQLAlchemy.
Census can trace HTTP requests made with the Requests package. The request URL, method, and status will be collected.
You can enable Requests integration by specifying 'requests'
to trace_integrations
.
Census can trace HTTP and gRPC requests made with the Cloud client libraries. The request URL, method, and status will be collected.
You can enable Google Cloud client libraries integration by specifying 'google_cloud_clientlibs'
to trace_integrations
.
The OpenCensus Stackdriver Stats Exporter allows users to export metrics to Stackdriver Monitoring. The API of this project is still evolving. The use of vendoring or a dependency management tool is recommended.
from opencensus.stats.exporters import stackdriver_exporter as stackdriver from opencensus.stats import stats as stats_module
- OpenCensus Python libraries require Python 2.7 or later.
- Google Cloud Platform account and project.
- Google Stackdriver Monitoring enabled on your project (Need help? Click here).
stats = stats_module.Stats() view_manager = stats.view_manager exporter = stackdriver.new_stats_exporter(stackdriver.Options(project_id="<id_value>")) view_manager.register_exporter(exporter) ...
In the examples folder, you can find all the necessary steps to get the exporter, register a view, put tags on the measure, and see the values against the Stackdriver monitoring tool once you have defined the project_id.
For further details for the Stackdriver implementation, see the file stackdriver_exporter.py.
Path & File | Short Description |
---|---|
examples/stats/exporter/stackdriver.py | End to end example |
opencensus/stats/exporters/stackdriver_exporter.py | Stats implementation for Stackdriver |
The OpenCensus Prometheus Stats Exporter allows users to export metrics to Prometheus monitoring solution. The API of this project is still evolving. The use of vendoring or a dependency management tool is recommended.
from opencensus.stats.exporters import prometheus_exporter as prometheus from opencensus.stats import stats as stats_module
- OpenCensus Python libraries require Python 2.7 or later.
- Prometheus up and running.
stats = stats_module.Stats() view_manager = stats.view_manager exporter = prometheus.new_stats_exporter(prometheus.Options(namespace="<namespace>")) view_manager.register_exporter(exporter) ...
In the examples folder, you can find all the necessary steps to get the exporter, register a view, put tags on the measure, and see the values against the Prometheus monitoring tool.
For further details for the Prometheus implementation, see the file prometheus_exporter.py.
Path & File | Short Description |
---|---|
examples/stats/exporter/prometheus.py | End to end example |
opencensus/stats/exporters/prometheus_exporter.py | Stats implementation for Prometheus |
Contributions to this library are always welcome and highly encouraged.
See CONTRIBUTING for more information on how to get started.
cd trace tox -e py34 source .tox/py34/bin/activate # Install nox with pip pip install nox-automation # See what's available in the nox suite nox -l # Run a single nox command nox -s "unit(py='2.7')" # Run all the nox commands nox # Integration test # We don't have script for integration test yet, but can test as below. python setup.py bdist_wheel cd dist pip install opencensus-0.0.1-py2.py3-none-any.whl # Then just run the tracers normally as you want to test.
Apache 2.0 - See LICENSE for more information.
This is not an official Google product.