Don't trace if it's a http request made by the exporter (#289)

Disable httplib tracing for http calls made by the exporter, which can cause an
infinite loop.

Fixes #135

Author: geobeau

README.rst

@@ -418,8 +418,29 @@ method, and status will be collected.
 
 You can enable Requests integration by specifying ``'requests'`` to ``trace_integrations``.
 
+It's possible to configure a list of URL you don't want traced. By default the request to exporter
+won't be traced. It's configurable by giving an array of hostname/port to the attribute
+``blacklist_hostnames`` in OpenCensus context's attributes:
+
+.. code:: python
+
+    execution_context.set_opencensus_attr('blacklist_hostnames',['hostname:port'])
+
+Only the hostname must be specified if only the hostname is specified in the URL request.
+
 .. _Requests package: https://pypi.python.org/pypi/requests
 
+Httplib
+~~~~~~~~
+
+Census can trace HTTP requests made with the httplib library.
+
+You can enable Requests integration by specifying ``'httplib'`` to ``trace_integrations``.
+
+It's possible to configure a list of URL you don't want traced. See requests integration
+for more information. The only difference is that you need to specify hostname and port
+every time.
+
 Google Cloud Client Libraries
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

opencensus/trace/ext/django/config.py

@@ -37,6 +37,7 @@
     'ZIPKIN_EXPORTER_PORT': 9411,
     'ZIPKIN_EXPORTER_PROTOCOL': 'http',
     'OCAGENT_TRACE_EXPORTER_ENDPOINT': None,
+    'BLACKLIST_HOSTNAMES': None,
     'TRANSPORT': 'opencensus.trace.exporters.transports.sync.SyncTransport',
 }
 

opencensus/trace/ext/django/middleware.py

@@ -45,6 +45,7 @@
 ZIPKIN_EXPORTER_PORT = 'ZIPKIN_EXPORTER_PORT'
 ZIPKIN_EXPORTER_PROTOCOL = 'ZIPKIN_EXPORTER_PROTOCOL'
 OCAGENT_TRACE_EXPORTER_ENDPOINT = 'OCAGENT_TRACE_EXPORTER_ENDPOINT'
+BLACKLIST_HOSTNAMES = 'BLACKLIST_HOSTNAMES'
 
 log = logging.getLogger(__name__)
 
@@ -161,6 +162,9 @@ def __init__(self, get_response=None):
         else:
             self.exporter = self._exporter(transport=transport)
 
+        self.blacklist_hostnames = settings.params.get(
+            BLACKLIST_HOSTNAMES, None)
+
         # Initialize the propagator
         self.propagator = self._propagator()
 
@@ -179,6 +183,10 @@ def process_request(self, request):
             REQUEST_THREAD_LOCAL_KEY,
             request)
 
+        execution_context.set_opencensus_attr(
+            'blacklist_hostnames',
+            self.blacklist_hostnames)
+
         try:
             # Start tracing this request
             span_context = self.propagator.from_headers(

opencensus/trace/ext/flask/flask_middleware.py

@@ -45,6 +45,7 @@
 ZIPKIN_EXPORTER_PORT = 'ZIPKIN_EXPORTER_PORT'
 ZIPKIN_EXPORTER_PROTOCOL = 'ZIPKIN_EXPORTER_PROTOCOL'
 OCAGENT_TRACE_EXPORTER_ENDPOINT = 'OCAGENT_TRACE_EXPORTER_ENDPOINT'
+BLACKLIST_HOSTNAMES = 'BLACKLIST_HOSTNAMES'
 
 log = logging.getLogger(__name__)
 
@@ -165,6 +166,7 @@ def init_app(self, app):
         else:
             self.exporter = self.exporter(transport=transport)
 
+        self.blacklist_hostnames = params.get(BLACKLIST_HOSTNAMES, None)
         # Initialize the propagator
         if inspect.isclass(self.propagator):
             self.propagator = self.propagator()
@@ -203,6 +205,9 @@ def _before_request(self):
             tracer.add_attribute_to_current_span(
                 HTTP_METHOD, flask.request.method)
             tracer.add_attribute_to_current_span(HTTP_URL, flask.request.url)
+            execution_context.set_opencensus_attr(
+                'blacklist_hostnames',
+                self.blacklist_hostnames)
         except Exception:  # pragma: NO COVER
             log.error('Failed to trace request', exc_info=True)
 

opencensus/trace/ext/httplib/trace.py

@@ -18,6 +18,7 @@
 from opencensus.trace import attributes_helper
 from opencensus.trace import execution_context
 from opencensus.trace import span as span_module
+from opencensus.trace.ext import utils
 
 PYTHON2 = sys.version_info.major == 2
 
@@ -61,6 +62,12 @@ def wrap_httplib_request(request_func):
 
     def call(self, method, url, body, headers, *args, **kwargs):
         _tracer = execution_context.get_opencensus_tracer()
+        blacklist_hostnames = execution_context.get_opencensus_attr(
+            'blacklist_hostnames')
+        dest_url = '{}:{}'.format(self._dns_host, self.port)
+        if utils.disable_tracing_hostname(dest_url, blacklist_hostnames):
+            return request_func(self, method, url, body,
+                                headers, *args, **kwargs)
         _span = _tracer.start_span()
         _span.span_kind = span_module.SpanKind.CLIENT
         _span.name = '[httplib]{}'.format(request_func.__name__)
@@ -100,7 +107,7 @@ def call(self, *args, **kwargs):
         span = _tracer.current_span()
 
         # No corresponding request span is found, request not traced.
-        if span.span_id != current_span_id:
+        if not span or span.span_id != current_span_id:
             return response_func(self, *args, **kwargs)
 
         result = response_func(self, *args, **kwargs)

opencensus/trace/ext/requests/trace.py

@@ -15,10 +15,15 @@
 import logging
 import requests
 import wrapt
+try:
+    from urllib.parse import urlparse
+except ImportError:
+    from urlparse import urlparse
 
 from opencensus.trace import attributes_helper
 from opencensus.trace import execution_context
 from opencensus.trace import span as span_module
+from opencensus.trace.ext import utils
 
 log = logging.getLogger(__name__)
 
@@ -56,6 +61,16 @@ def trace_integration(tracer=None):
 def wrap_requests(requests_func):
     """Wrap the requests function to trace it."""
     def call(url, *args, **kwargs):
+        blacklist_hostnames = execution_context.get_opencensus_attr(
+            'blacklist_hostnames')
+        parsed_url = urlparse(url)
+        if parsed_url.port is None:
+            dest_url = parsed_url.hostname
+        else:
+            dest_url = '{}:{}'.format(parsed_url.hostname, parsed_url.port)
+        if utils.disable_tracing_hostname(dest_url, blacklist_hostnames):
+            return requests_func(url, *args, **kwargs)
+
         _tracer = execution_context.get_opencensus_tracer()
         _span = _tracer.start_span()
         _span.name = '[requests]{}'.format(requests_func.__name__)
@@ -80,6 +95,17 @@ def wrap_session_request(wrapped, instance, args, kwargs):
     """Wrap the session function to trace it."""
     method = kwargs.get('method') or args[0]
     url = kwargs.get('url') or args[1]
+
+    blacklist_hostnames = execution_context.get_opencensus_attr(
+        'blacklist_hostnames')
+    parsed_url = urlparse(url)
+    if parsed_url.port is None:
+        dest_url = parsed_url.hostname
+    else:
+        dest_url = '{}:{}'.format(parsed_url.hostname, parsed_url.port)
+    if utils.disable_tracing_hostname(dest_url, blacklist_hostnames):
+        return wrapped(*args, **kwargs)
+
     _tracer = execution_context.get_opencensus_tracer()
     _span = _tracer.start_span()
 

opencensus/trace/ext/utils.py

@@ -14,6 +14,8 @@
 
 import re
 
+from opencensus.trace import execution_context
+
 # By default the blacklist urls are not tracing, currently just include the
 # health check url. The paths are literal string matched instead of regular
 # expressions. Do not include the '/' at the beginning of the path.
@@ -63,3 +65,31 @@ def disable_tracing_url(url, blacklist_paths=None):
             return True
 
     return False
+
+
+def disable_tracing_hostname(url, blacklist_hostnames=None):
+    """Disable tracing for the provided blacklist URLs, by default not tracing
+    the exporter url.
+
+    If the url path starts with the blacklisted path, return True.
+
+    :type blacklist_hostnames: list
+    :param blacklist_hostnames: URL that not tracing.
+
+    :rtype: bool
+    :returns: True if not tracing, False if tracing.
+    """
+    if blacklist_hostnames is None:
+        # Exporter host_name are not traced by default
+        _tracer = execution_context.get_opencensus_tracer()
+        try:
+            blacklist_hostnames = [
+                '{}:{}'.format(
+                    _tracer.exporter.host_name,
+                    _tracer.exporter.port
+                )
+            ]
+        except(AttributeError):
+            blacklist_hostnames = []
+
+    return url in blacklist_hostnames

tests/unit/trace/ext/django/test_config.py

@@ -86,6 +86,7 @@ def test__set_default_configs(self):
             'ZIPKIN_EXPORTER_PORT': 9411,
             'ZIPKIN_EXPORTER_PROTOCOL': 'http',
             'OCAGENT_TRACE_EXPORTER_ENDPOINT': None,
+            'BLACKLIST_HOSTNAMES': None,
             'TRANSPORT':
                 'opencensus.trace.exporters.transports.sync.SyncTransport',
         }

tests/unit/trace/ext/httplib/test_httplib_trace.py

@@ -99,7 +99,85 @@ def test_wrap_httplib_request(self):
         self.assertEqual(expected_attributes,
                          mock_tracer.span.attributes)
         self.assertEqual(expected_name, mock_tracer.span.name)
-        self.assertEqual(span_module.SpanKind.CLIENT, mock_tracer.span.span_kind)        
+        self.assertEqual(span_module.SpanKind.CLIENT, mock_tracer.span.span_kind)
+
+    def test_wrap_httplib_request_blacklist_ok(self):
+        mock_span = mock.Mock()
+        span_id = '1234'
+        mock_span.span_id = span_id
+        mock_tracer = MockTracer(mock_span)
+        mock_request_func = mock.Mock()
+        mock_request_func.__name__ = 'request'
+
+        patch_tracer = mock.patch(
+            'opencensus.trace.ext.requests.trace.execution_context.'
+            'get_opencensus_tracer',
+            return_value=mock_tracer)
+        patch_attr = mock.patch(
+            'opencensus.trace.ext.requests.trace.execution_context.'
+            'get_opencensus_attr',
+            return_value=None)
+
+        wrapped = trace.wrap_httplib_request(mock_request_func)
+
+        mock_self = mock.Mock()
+        method = 'GET'
+        url = 'http://localhost:8080'
+        body = None
+        headers = {}
+
+        with patch_tracer, patch_attr:
+            wrapped(mock_self, method, url, body, headers)
+
+        expected_attributes = {
+            'http.url': url,
+            'http.method': method}
+        expected_name = '[httplib]request'
+
+        mock_request_func.assert_called_with(
+            mock_self, method, url, body, {
+                'traceparent': '00-123-456-01',
+            }
+        )
+
+    def test_wrap_httplib_request_blacklist_nok(self):
+        mock_span = mock.Mock()
+        span_id = '1234'
+        mock_span.span_id = span_id
+        mock_tracer = MockTracer(mock_span)
+        mock_request_func = mock.Mock()
+        mock_request_func.__name__ = 'request'
+
+        patch_tracer = mock.patch(
+            'opencensus.trace.ext.requests.trace.execution_context.'
+            'get_opencensus_tracer',
+            return_value=mock_tracer)
+        patch_attr = mock.patch(
+            'opencensus.trace.ext.requests.trace.execution_context.'
+            'get_opencensus_attr',
+            return_value=['localhost:8080'])
+
+        wrapped = trace.wrap_httplib_request(mock_request_func)
+
+        mock_self = mock.Mock()
+        mock_self._dns_host = 'localhost'
+        mock_self.port = '8080'
+        method = 'GET'
+        url = 'http://{}:{}'.format(mock_self._dns_host, mock_self.port)
+        body = None
+        headers = {}
+
+        with patch_tracer, patch_attr:
+            wrapped(mock_self, method, url, body, headers)
+
+        expected_attributes = {
+            'http.url': url,
+            'http.method': method}
+        expected_name = '[httplib]request'
+
+        mock_request_func.assert_called_with(
+            mock_self, method, url, body, {}
+        )
 
     def test_wrap_httplib_response(self):
         mock_span = mock.Mock()