For some tracer implementation (like Jaeger), the initialization of the tracer should be deferred until the wsgi server forks. We cannot directly create a jaeger tracer instance and pass it to FlaskTracer constructor upon startup.

This PR addresses this issue. The underlying tracer initialization can be deferred if we pass a callable as argument to the constructor.

What is the status of getting master released? I need the broader opentracing version pinning in order to incorporate this library into my project, and master seems to work in my project, but I would much prefer to install a released package.

If the flask app has a before_request method that shortcuts request execution, the request will not be present in the _current_spans dict but the tracer will still try to finish the span. Adding this default prevents an exception on those calls.

Hi,

opentracing 1.3 can run on python 3, 1.2 and below he can't.

In the setup, 'opentracing>=1.1,<1.2' can be upgrade or there is some issue?

Regards,

Hello @carlosalberto and @yurishkuro,

The main purpose of this PR is to simplify the deployment flow. Please note that user and encrypted password must be set properly.

Useful links:

• docs.travis-ci.com/user/encryption-keys
• docs.travis-ci.com/user/deployment/pypi
• docs.travis-ci.com/user/deployment#examples-of-conditional-deployment
• docs.travis-ci.com/user/conditional-builds-stages-jobs
• docs.travis-ci.com/user/conditions-v1

Best regards!

We have a recursive route we're using for testing, when using the python aws xray tracing here: https://github.com/aws/aws-xray-sdk-python

It includes all depth in the traces. However with the python-flask opentracing module, it's only showing one level of recursion in the waterfall graph. Is there anyway to get more detail?

This is my config:

    config = jaeger_client.Config(config={'sampler': {'type': 'const', 'param': 1}, 
                            'logging': True,
                            'local_agent':
                            {'reporting_host': JAEGER_HOST}},
                    service_name="jaeger_opentracing_example")
    jaeger_tracer = config.initialize_tracer()
    # This will trace all requests; routes can also be chosen individually instead
    # Traced attributes lists additional request attributes we want to capture
    tracing = FlaskTracing(jaeger_tracer, True, app, traced_atrributes=['remote_addr'])

Here is the flask route:

@app.route("/recurse/<number>", methods=['GET'])
def recurse(number):
    '''Recursively call itself with sleep'''
    num = int(number)
    if num> 0:
        val = num - 1
        sleep((randint(200,600)/1000.0))
        r = requests.get("http://turquoise.default/recurse/%d" % val)
        string = "%sAt depth %d!<br>" % (r.text, num)
        return string, r.status_code
    else:
        return "Depth 0!", 200

For instance calling /recurse/7 results in the following waterfalls:

Following the standard example but calling Flask.run explicitly, like:

       def initialize_tracer():
            from jaeger_client import Config
            config = Config(
                config={
                    'sampler': {'type': 'const', 'param': 1}
                },
                service_name='hello-world')
            return config.initialize_tracer() # also sets opentracing.tracer                                      

        def rest_prediction_server():
            app = seldon_microservice.get_rest_microservice(
                user_object, debug=DEBUG)


            from flask_opentracing import FlaskTracer

            flask_tracer = FlaskTracer(initialize_tracer, True, app)

            app.run(host='0.0.0.0', port=port)

I get on REST call:

Traceback (most recent call last):
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/app.py", line 2292, in wsgi_app
    response = self.full_dispatch_request()
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/app.py", line 1815, in full_dispatch_request
    rv = self.handle_user_exception(e)
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask_cors/extension.py", line 161, in wrapped_function
    return cors_after_request(app.make_response(f(*args, **kwargs)))
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/app.py", line 1718, in handle_user_exception
    reraise(exc_type, exc_value, tb)
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/_compat.py", line 35, in reraise
    raise value
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/app.py", line 1811, in full_dispatch_request
    rv = self.preprocess_request()
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask/app.py", line 2087, in preprocess_request
    rv = func()
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask_opentracing/tracing.py", line 39, in start_trace
    self._before_request_fn(traced_attributes)
  File "/home/clive/anaconda3/lib/python3.6/site-packages/flask_opentracing/tracing.py", line 121, in _before_request_fn
    child_of=span_ctx)
  File "/home/clive/anaconda3/lib/python3.6/site-packages/opentracing/tracer.py", line 126, in start_active_span
    return self._noop_scope
AttributeError: 'Tracer' object has no attribute '_noop_scope'

Any ideas?

I've added examples that use Jaeger instead of Lightstep, introduce serialization to the wire, and are bare minimal for ease of understanding.

Before running both files you need to set JAEGER_HOST and WEBSERVER_HOST environment variables.

@bhs @yurishkuro - please let me know if the approach seems good for you. I'll be writing README after your thumbs up on general direction.

The following script never starts the webserver and just hangs forever:

import opentracing
from flask import Flask
from flask_opentracing import FlaskTracer
from jaeger_client import Config

app = Flask(__name__)

config = Config(
config={ # usually read from some yaml config
                          'sampler': {
                          'type': 'const',
                          'param': 1,
                  },
                          'logging': True,
                  },
service_name='flask_app_1',
)


opentracing_tracer = config.initialize_tracer()
tracer = FlaskTracer(opentracing_tracer)

@app.route('/some_url')
@tracer.trace()
def some_view_func():
    return "some_view"

With debug logging turned on, I see this:

2017-11-06 15:10:16,177 Initializing Jaeger Tracer with UDP reporter

Same script with minimal changes started by python scriptname.py works as a charm.

The python packaging doesn't declare itself to prefer utf-8, but the README contains it, so the long_description=open('README.rst').read() is failing with: UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 3800: ordinal not in range(128)

There's only this single character, though, so just replace it with ASCII and solve the issue the easy way.

I have an flask app using flasktracing, all the requests are shown in my jaeger backend UI without WSGI server. But when I run with gunicorn, I saw my "no-root-span" messages in the UI. I am using threads only with only one sync worker, and I have tried lazy-init and normal init the global tracer . Any idea how to fix it?

How to reproduce the issue

Let's use whatever tracer (I used jaeger for this example) and use the trace decorator. Ex.:

from flask import Flask
from flask_opentracing import FlaskTracer

app = Flask(__name__)


def initialize_tracer():
    config = Config(
        config={
            'sampler': {'type': 'const', 'param': 1}
        },
        service_name='hello-world'
    )
    return config.initialize_tracer()


flask_tracer = FlaskTracer(initialize_tracer)


@app.route('/')
@flask_tracer.trace()
def hello():
    return 'Hello'

In the resulting span, the HTTP_STATUS_CODE tag should not be present.

Possible cause

Looking at the code, it seems the response is not being passed to the _after_request_fn method when using the trace decorator.

https://github.com/opentracing-contrib/python-flask/blob/master/flask_opentracing/tracing.py#L83

Possible fix

I attached a PR with one possible solution.

In the opentracing API the scope manager is stored in the property _scope_manager and can be accessed as scope_manager. Currently, the property _scope_manager is not set in the FlaskTracing class. This leads to issues with opentracing instrumentation that needs access to it, e.g. the boto3 part of the https://github.com/uber-common/opentracing-python-instrumentation. (When I have more time, I will try to add a minimal example where failures actually occur.)

I could imagine two fixes to this a) adding

@property 
def _scope_manager(self):
    return self.tracer._scopemanager

or b) explicitly doing:

self._scope_manager = self.tracer.scope_manager

in the initializer. Happy to write the PR for this, if you provide me your preferences :slightly_smiling_face:

Traced_attributes is given and scanned in _after_request_fn too. The motivation is that this way the attributes of the response object could be presented in the trace also. Our goal is to have more control over the traced attributes because our task requires also the input and the output data in one point.

The wavefront tracer specifically requires span names be non-None:

ValueError
ValueError: Span name cannot be blank

ValueError
ValueError: Span name cannot be blank

Traceback (most recent call last)
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2463, in __call__
return self.wsgi_app(environ, start_response)
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2449, in wsgi_app
response = self.handle_exception(e)
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 1866, in handle_exception
reraise(exc_type, exc_value, tb)
File "/usr/local/lib/python3.7/site-packages/flask/_compat.py", line 39, in reraise
raise value
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2446, in wsgi_app
response = self.full_dispatch_request()
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 1952, in full_dispatch_request
return self.finalize_request(rv)
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 1969, in finalize_request
response = self.process_response(response)
File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2266, in process_response
response = handler(response)
File "/usr/local/lib/python3.7/site-packages/flask_opentracing/tracing.py", line 43, in end_trace
self._after_request_fn(response)
File "/usr/local/lib/python3.7/site-packages/flask_opentracing/tracing.py", line 160, in _after_request_fn
scope.close()
File "/usr/local/lib/python3.7/site-packages/opentracing/scope_managers/__init__.py", line 77, in close
self.span.finish()
File "/usr/local/lib/python3.7/site-packages/wavefront_opentracing_sdk/span.py", line 165, in finish
self._do_finish(time.time() - self.start_time)
File "/usr/local/lib/python3.7/site-packages/wavefront_opentracing_sdk/span.py", line 191, in _do_finish
self.tracer.report_span(self)
File "/usr/local/lib/python3.7/site-packages/wavefront_opentracing_sdk/tracer.py", line 274, in report_span
self._reporter.report(span)
File "/usr/local/lib/python3.7/site-packages/wavefront_opentracing_sdk/reporting/composite.py", line 32, in report
rep.report(wavefront_span)
File "/usr/local/lib/python3.7/site-packages/wavefront_opentracing_sdk/reporting/console.py", line 35, in report
default_source='unknown')
File "/usr/local/lib/python3.7/site-packages/wavefront_sdk/common/utils.py", line 236, in tracing_span_to_line_data
raise ValueError('Span name cannot be blank')
ValueError: Span name cannot be blank

Is there a mechanism to provide a default name for that case?