Support transport options in exporter

[ocervell]

This PR adds support to pass options to the transport initialized in our exporters.

This will allow, for instance, to set the grace_period and max_batch_size for the BackgroundThread transport when initializing the exporter:

from opencensus.trace.exporters.stackdriver_exporter import StackdriverExporter
from opencensus.trace.exporters.transports.background_thread import BackgroundThreadTransport

exporter = StackdriverExporter(
    transport=BackgroundThreadTransport, 
    transport_config={'grace_period': 1, 'max_batch_size': 30})

[c24t]

I wonder why we're initializing the transport here instead of passing the exporter constructor a preconfigured transport. This may be a simpler solution to your problem.

[felippe-mendonca]

@c24t, I suggested this in #249 (pass an object instead), but add a configuration parameter could be interesting for backwards compatibility.

[ocervell]

I agree with @felippe-mendonca, this is mostly to keep backwards compatibility while giving more options. Maybe we can think of a more global reformat later on.

[c24t]

@ocervell it looks like this branch still has some lint and coverage failures.

To check for new lint errors quickly you can do:

flake8 $(git diff --name-only upstream/master)

And to see the detailed coverage report:

% python -m pytest --cov=opencensus --cov-config=.coveragerc
% python -m coverage html
% open htmlcov/index.html

[ocervell]

Need to add a test for the init_transport function to get full coverage.

[felippe-mendonca]

Need to add a test for the init_transport function to get full coverage.

@ocervell , I'd suggest only the modification below on test_base_exporter.py file to test the init_transport function, since it's already tested indirectly on all exporters' tests.

class TestBaseExporter(unittest.TestCase):
    def test_export_abstract(self):
        from opencensus.trace.exporters import base

        exporter = base.Exporter()
        trace = {}

        with self.assertRaises(NotImplementedError):
            exporter.export(trace)

    def test_emit_abstract(self):
        from opencensus.trace.exporters import base

        exporter = base.Exporter()
        trace = {}

        with self.assertRaises(NotImplementedError):
            exporter.emit(trace)

    def test_init_transport(self):
        from opencensus.trace.exporters import base

        with self.assertRaises(TypeError):
            transport = base.init_transport(
                exporter=base.Exporter(),
                transport=MockTransport,
                config={
                    "par2": 0,
                })


class MockTransport(object):
    def __init__(self, exporter=None, par1=0):
        self.export_called = False
        self.exporter = exporter
        self.par1 = par1

    def export(self, trace):
        self.export_called = True

[davewalkexpel]

Hi, this is cool but would it be possible to make _WAIT_PERIOD configurable via the new transport_config dict? Or would that be out of scope and should I start a new PR?

My use case is I want to send less write requests to Stackdriver so that we don't exceed our write quota. I can control the max batch size of the requests now, but the app is still going to send a write request every second even if there's only one span to emit (traces are getting created continuously). If I could up the wait period to something like five seconds I would really be in business. Thanks!

[c24t]

would it be possible to make _WAIT_PERIOD configurable via the new transport_config dict?

It'd mean adding an arg to the AsyncTransport constructor, but there's at least as good an argument for making this configurable as the grace period (which already has an arg).

[felippe-mendonca]

@davewalkexpel, Seems good to be possible to also configure _WAIT_TIME in your use case. I had a similar problem but I'd solved only changing the batch size. Besides, adjust _WAIT_TIME could also be interesting in cases where you have multiples components sending traces, which mean multiple requests to the collector. In that case, increase interval and batch size might reduce collector's pressure. However, I think that this feature fits to another PR, and maybe change the name from 'wait_time' to 'interval'. Or, even more, open an discussion about the AsyncTransport export strategy.

[reyang]

This has been addressed by the configuration refactor work in #394.
The following mechanism has been added in #621:
https://github.com/census-instrumentation/opencensus-python/tree/master/opencensus/common/configuration

[reyang]

There is plan to refactor exporter, worker, queue and transport, refer to the conversations in #642.