[requests] add a default service name, or use a user defined one or the parent if available

Author: Emanuele Palazzetti

ddtrace/contrib/requests/connection.py

@@ -2,6 +2,8 @@
 import logging
 import ddtrace
 
+from .constants import DEFAULT_SERVICE
+
 from ...ext import http
 from ...util import asbool
 from ...propagation.http import HTTPPropagator
@@ -16,7 +18,29 @@ def _wrap_session_init(func, instance, args, kwargs):
 
     # set tracer settings
     distributed_tracing = asbool(os.environ.get('DATADOG_REQUESTS_DISTRIBUTED_TRACING')) or False
+    service_name = os.environ.get('DATADOG_REQUESTS_SERVICE_NAME') or DEFAULT_SERVICE
     setattr(instance, 'distributed_tracing', distributed_tracing)
+    setattr(instance, 'service_name', service_name)
+
+
+def _extract_service_name(session, span):
+    """Extracts the right service name based on the following logic:
+    - `requests` is the default service name
+    - users can change it via `session.service_name = 'clients'`
+    - if the Span doesn't have a parent, use the set service name
+      or fallback to the default
+    - if the Span has a parent, use the set service name or the
+      parent service value if the set service name is the default
+
+    The priority can be represented as:
+    Updated service name > parent service name > default to `requests`.
+    """
+    service_name = getattr(session, 'service_name', DEFAULT_SERVICE)
+    if (service_name == DEFAULT_SERVICE and
+            span._parent is not None and
+            span._parent.service is not None):
+        service_name = span._parent.service
+    return service_name
 
 
 def _wrap_request(func, instance, args, kwargs):
@@ -35,6 +59,9 @@ def _wrap_request(func, instance, args, kwargs):
     headers = kwargs.get('headers', {})
 
     with tracer.trace("requests.request", span_type=http.TYPE) as span:
+        # update the span service name before doing any action
+        span.service = _extract_service_name(instance, span)
+
         if distributed_tracing:
             propagator = HTTPPropagator()
             propagator.inject(span.context, headers)

ddtrace/contrib/requests/constants.py

@@ -0,0 +1 @@
+DEFAULT_SERVICE = 'requests'

tests/contrib/requests/test_requests.py

@@ -139,3 +139,69 @@ def test_500(self):
         eq_(s.get_tag(http.METHOD), 'GET')
         eq_(s.get_tag(http.STATUS_CODE), '500')
         eq_(s.error, 1)
+
+    def test_default_service_name(self):
+        # ensure a default service name is set
+        out = self.session.get(URL_200)
+        eq_(out.status_code, 200)
+
+        spans = self.tracer.writer.pop()
+        eq_(len(spans), 1)
+        s = spans[0]
+
+        eq_(s.service, 'requests')
+
+    def test_user_set_service_name(self):
+        # ensure a service name set by the user has precedence
+        self.session.service_name = 'clients'
+        out = self.session.get(URL_200)
+        eq_(out.status_code, 200)
+
+        spans = self.tracer.writer.pop()
+        eq_(len(spans), 1)
+        s = spans[0]
+
+        eq_(s.service, 'clients')
+
+    def test_parent_service_name_precedence(self):
+        # ensure the parent service name has precedence if the value
+        # is not set by the user
+        with self.tracer.trace('parent.span', service='web'):
+            out = self.session.get(URL_200)
+            eq_(out.status_code, 200)
+
+        spans = self.tracer.writer.pop()
+        eq_(len(spans), 2)
+        s = spans[1]
+
+        eq_(s.name, 'requests.request')
+        eq_(s.service, 'web')
+
+    def test_parent_without_service_name(self):
+        # ensure the default value is used if the parent
+        # doesn't have a service
+        with self.tracer.trace('parent.span'):
+            out = self.session.get(URL_200)
+            eq_(out.status_code, 200)
+
+        spans = self.tracer.writer.pop()
+        eq_(len(spans), 2)
+        s = spans[1]
+
+        eq_(s.name, 'requests.request')
+        eq_(s.service, 'requests')
+
+    def test_user_service_name_precedence(self):
+        # ensure the user service name takes precedence over
+        # the parent Span
+        with self.tracer.trace('parent.span', service='web'):
+            self.session.service_name = 'clients'
+            out = self.session.get(URL_200)
+            eq_(out.status_code, 200)
+
+        spans = self.tracer.writer.pop()
+        eq_(len(spans), 2)
+        s = spans[1]
+
+        eq_(s.name, 'requests.request')
+        eq_(s.service, 'clients')