OPENTRACING PYTHON UTILS

Early stage WIP + Experimental

Convenient utilities for adding OpenTracing support in your python projects.

Features

opentracing-utils should provide and aims at the following:

• No external dependencies, only opentracing-python.

• No threadlocals. Either pass spans explicitly or fallback to callstack frames inspection!

• Context agnostic, so no external context implementation dependency (no Tornado, Flask, Django etc ...).

• Try to be less verbose - just add the @trace decorator.

• Could be more verbose when needed, without complexity - just accept **kwargs and get the span passed to your traced functions via @trace(pass_span=True).

• Support asyncio/async-await coroutines. (drop support for py2.7)

• Support gevent.

• Ability to add OpenTracing support to external libs/frameworks/clients:

  • Django (via OpenTracingHttpMiddleware)
  • Flask (via trace_flask())
  • Requests (via trace_requests())
  • SQLAlchemy (via trace_sqlalchemy())

Install

Using pip (not released yet to PyPi)

pip install -U opentracing-utils

or by cloning the repo

python setup.py install

Usage

init_opentracing_tracer

The first step needed in OpenTracing instrumentation is to initialize a tracer. Each vendor defines how the tracer can be initialized. Currently the following tracers are supported:

• BasicTracer
• Instana
• Jaeger
• LightStep

BasicTracer

This is the basic noop tracer. It could be initialized with a recorder (e.g. Memory Recorder), which can be useful in debugging and playing around with OpenTracing concepts.

import opentracing
from opentracing_utils import OPENTRACING_BASIC, init_opentracing_tracer

# Initialize upon application start
init_opentracing_tracer(OPENTRACING_BASIC)

# It is possible to pass custom recorder
# init_opentracing_tracer(OPENTRACING_BASIC, recorder=custom_recorder)

# Now use the opentracing.tracer
root_span = opentracing.tracer.start_span(operation_name='root_span')

Instana

Config Vars

The following config variables can be used in initialization if set as env variables

OPENTRACING_INSTANA_SERVICE

The service name.

import opentracing
from opentracing_utils import OPENTRACING_INSTANA, init_opentracing_tracer

# Initialize upon application start
init_opentracing_tracer(OPENTRACING_INSTANA)

# It is possible to pass args
# init_opentracing_tracer(OPENTRACING_INSTANA, service='python-server')

# Now use the opentracing.tracer
root_span = opentracing.tracer.start_span(operation_name='root_span')

Dependency

Add instana to the dependencies.txt of your project.

Jaeger

Config Vars

The following config variables can be used in initialization if set as env variables

OPENTRACING_JAEGER_SERVICE_NAME

The service name.

Note

Jaeger configuration should be passed by the instrumentated code. Default is {}.

import opentracing
from opentracing_utils import OPENTRACING_JAEGER, init_opentracing_tracer

# Initialize upon application start
init_opentracing_tracer(OPENTRACING_JAEGER)

# It is possible to pass args
# init_opentracing_tracer(OPENTRACING_JAEGER, service_name='python-server', config=custom_config_with_sampling)

# Now use the opentracing.tracer
root_span = opentracing.tracer.start_span(operation_name='root_span')

Dependency

Add jaeger_client to the dependencies.txt of your project.

LightStep

Config Vars

The following config variables can be used in initialization if set as env variables

OPENTRACING_LIGHTSTEP_COMPONENT_NAME

The component name.

OPENTRACING_LIGHTSTEP_ACCESS_TOKEN

The LightStep collector access token.

OPENTRACING_LIGHTSTEP_COLLECTOR_HOST

The LightStep collector host. Default: collector.lightstep.com.

OPENTRACING_LIGHTSTEP_COLLECTOR_PORT

The LightStep collector port (int). Default: 443.

OPENTRACING_LIGHTSTEP_VERBOSITY

The verbosity of the tracer (int). Default: 0.

import opentracing
from opentracing_utils import OPENTRACING_LIGHTSTEP, init_opentracing_tracer

# Initialize upon application start
init_opentracing_tracer(OPENTRACING_LIGHTSTEP)

# It is possible to pass args
# init_opentracing_tracer(OPENTRACING_LIGHTSTEP, component_name='python-server', access_token='123', collector_host='production-collector.com')

# Now use the opentracing.tracer
root_span = opentracing.tracer.start_span(operation_name='root_span')

Dependency

Add lightstep to the dependencies.txt of your project.

@trace decorator

from opentracing_utils import trace, extract_span_from_kwargs

# decorate all your functions that require tracing

# Normal traced function
@trace()
def trace_me():
    pass


# Traced function with access to created span in ``kwargs``
@trace(operation_name='user.operation', pass_span=True)
def user_operation(user, op, **kwargs):
    current_span = extract_span_from_kwargs(**kwargs)

    current_span.set_tag('user.id', user.id)

    # Then do stuff ...

    # trace_me will have ``current_span`` as its parent.
    trace_me()

# Traced function using ``follows_from`` instead of ``child_of`` reference.
@trace(use_follows_from=True)
def trace_me_later():
    pass


# Start a fresh trace - any parent spans will be ignored
@trace(operation_name='epoch', ignore_parent_span=True)
def start_fresh():

    user = {'id': 1}

    # trace decorator will handle trace heirarchy
    user_operation(user, 'create')

    # trace_me will have ``epoch`` span as its parent.
    trace_me()

Skip Spans

In certain cases you might need to skip certain spans while using the @trace decorator.

def skip_this_span(arg1, arg2, **kwargs):
    if arg1 == 'special':
        # span should be skipped
        return True

    return False


@trace(skip_span=skip_this_span)
def traced(arg1, arg2):
    pass


top_span = opentracing.tracer.start_span(operation_name='top_trace')
with top_span:
    # this call will be traced and have a span!
    traced('open', 'tracing')

    # this call won't be traced and no span to be added!
    traced('special', 'tracing')

Broken traces

If you plan to break nested traces, then it is recommended to pass the span to traced functions

top_span = opentracing.tracer.start_span(operation_name='top_trace')
with top_span:

    # This one gets ``top_span`` as parent span
    call_traced()

    # Here, we break the trace, since we create a new span with no parents
    broken_span = opentracing.tracer.start_span(operation_name='broken_trace')
    with broken_span:
        # This one gets ``broken_span`` as parent span (not consistent in 2.7 and 3.5)
        call_traced()

        # pass span as safer/guaranteed trace here
        call_traced(span=broken_span)

    # ISSUE: Due to stack call inspection, next call will get ``broken_span`` instead of ``top_span``, which is wrong!!
    call_traced()

    # To get the ``top_span`` as parent span, then pass it to the traced call
    call_traced(span=top_span)

Multiple traces

If you plan to use multiple traces then it is better to always pass the span as it is safer/guaranteed.

first_span = opentracing.tracer.start_span(operation_name='first_trace')
with first_span:

    # This one gets ``first_span`` as parent span
    call_traced()

second_span = opentracing.tracer.start_span(operation_name='second_trace')
with second_span:

    # ISSUE: This one **could** get ``first_span`` as parent span (not consistent among Python versions)
    call_traced()

    # It is better to pass ``second_span`` explicitly
    call_traced(span=second_span)

Generators (yield)

Using generators could get tricky and leads to invalid parent span inspection. It is recommended to pass the span explicitly.

@trace(pass_span=True)
def gen(**kwargs):
    s = extract_span_from_kwargs(**kwargs)  # noqa

    # Extract and pass span to ``f2()`` otherwise it could get ``f1()`` as parent span instead of ``gen()``
    f2(span=s)

    for i in range(10):
        yield i

@trace()
def f2():
    pass

@trace()
def f1():
    list(gen())

first_span = opentracing.tracer.start_span(operation_name='first_trace')
with first_span:
    f1()

External libraries and clients

Django

For tracing Django applications. You can use the following:

• OpenTracingHttpMiddleware: for tracing incoming HTTP requests

# In settings.py or equivalent Django config
from opentracing_utils import init_opentracing_tracer
init_opentracing_tracer(YOUR_TRACER)  # make sure opentracing.tracer is initialized properly.

MIDDLEWARE = (
'opentracing_utils.OpenTracingHttpMiddleware',  # goes first in the list
# ... more middlewares here
)

# Further options

# Add default tags to all incoming HTTP requests spans
OPENTRACING_UTILS_DEFAULT_TAGS = {'my-default-tag': 'tag-value'}

# Add error tag on 4XX responses (default is ``True``)
OPENTRACING_UTILS_ERROR_4XX = False

# Override span operation_name (default is ``view_func.__name__``)
OPENTRACING_UTILS_OPERATION_NAME_CALLABLE = 'my_app.utils.span_operation_name'

# Exclude certain requests from OpenTracing
OPENTRACING_UTILS_SKIP_SPAN_CALLABLE = 'my_app.utils.skip_span'

Here are the callables examples for overriding span operation names and skipping spans:

# my_app/utils.py
def span_operation_name(request, view_func, view_args, view_kwargs):
    return 'edge_{}'.format(view_func.__name__)

def skip_span(request, view_func, view_args, view_kwargs):
    if view_func.__name__.startswith('no_trace_'):
        return True
    return False

In order to follow traces in your views, you can use extract_span_from_django_request utility function.

# my_app/views.py

from opentracing_utils import trace, extract_span_from_django_request

@trace(span_extractor=extract_span_from_django_request, operation_name='custom_view')
def my_traced_view(request):
    ...

Flask

For tracing Flask applications. This utility function adds a middleware that handles all incoming requests to the Flask application.

from opentracing_utils import trace_flask, extract_span_from_flask_request
from flask import Flask

app = Flask(__name__)

trace_flask(app)

# You can add default_tags or optionally treat 4xx responses as not an error (i.e no error tag in span)
# trace_flask(app, default_tags={'always-there': True}, error_on_4xx=False)

# Extract current span from request context
def internal_function():
    current_span = extract_span_from_flask_request()

    current_span.set_tag('internal', True)

# You can skip requests spans.
def skip_health_checks(request):
    return request.path == '/health'

# trace_flask(skip_span=skip_health_checks)

Requests

For tracing requests client library for all outgoing requests.

# trace_requests should be called as early as possible, before importing requests
from opentracing_utils import trace_requests
trace_requests()  # noqa

# In case you want to include default span tags to be sent with every outgoing request.
# trace_requests(default_tags={'account_id': '123'}, set_error_tag=False)

# In case you want to keep the URL query args (masked by default in order to avoid leaking auth tokens etc...)
# trace_requests(mask_url_query=False)

# You can also mask URL path parameters (e.g. http://hostname/1 will be http://hostname/??/)
# trace_requests(mask_url_path=True)

# The library patches the requests library send functionality. This causes
# all requests to propagate the span id's in the headers. Sometimes this is
# undesireable so it's also possible to avoid tracing specific URL's or
# endpoints. trace_requests accepts a list of regex patterns and matches the
# request.url against these patterns, ignoring traces if any pattern matches.
# trace_requests(ignore_url_patterns=[r".*hostname/endpoint"])

import requests

def main():

    span = opentracing.tracer.start_span(operation_name='main')
    with span:
        # Following call will be traced as a ``child span`` and propagated via HTTP headers.
        requests.get('https://example.org')

SQLAlchemy

For tracing SQLAlchemy client library for all SQL queries.

# trace_sqlalchemy can be used to trace all SQL queries.
# By default, span operation_name will be deduced from the query statement (e.g. select, update, delete).
from opentracing_utils import trace_sqlalchemy
trace_sqlalchemy()

# You can customize the span operation_name via supplying a callable
def get_sqlalchemy_span_op_name(conn, cursor, statement, parameters, context, executemany):
    # inspect statement and parameters etc...
    return 'custom_operation_name'
# trace_sqlalchemy(operation_name=get_sqlalchemy_span_op_name)

# By default, trace_sqlalchemy will not set error tags for SQL errors/exceptions. You can change that via ``set_error_tag`` param.
# trace_sqlalchemy(set_error_tag=True)

# you can skip spans for certain SQL queries.
def skip_inserts(conn, cursor, statement, parameters, context, executemany):
    return statement.lower().startswith('insert')

# trace_sqlalchemy(skip_span=skip_inserts)

# you can enrich the span with by supplying an ``enrich_span`` callable.
def enrich_sql_span_parameters(span, conn, cursor, statement, parameters, context, executemany):
    span.set_tag('parameters', parameters)

# trace_sqlalchemy(enrich_span=enrich_sql_span_parameters)

License

The MIT License (MIT)

Copyright (c) 2017 Zalando SE, https://tech.zalando.com

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.