chore(runtime metrics): fix telemetry reporting [backport 3.1]

.github/CODEOWNERS

@@ -48,7 +48,6 @@ tests/internal                      @DataDog/apm-core-python
 tests/lib-injection                 @DataDog/apm-core-python
 tests/opentelemetry                 @DataDog/apm-core-python
 tests/opentracer                    @DataDog/apm-core-python
-tests/runtime                       @DataDog/apm-core-python
 tests/tracer                        @DataDog/apm-core-python
 
 # Test Visibility and related
@@ -166,7 +165,7 @@ ddtrace/internal/remoteconfig       @DataDog/remote-config @DataDog/apm-core-pyt
 tests/internal/remoteconfig         @DataDog/remote-config @DataDog/apm-core-python
 
 # API SDK
-ddtrace/trace/                                    @DataDog/apm-sdk-api-python
+ddtrace/trace/                                     @DataDog/apm-sdk-api-python
 ddtrace/_trace/                                    @DataDog/apm-sdk-api-python
 ddtrace/opentelemetry/                             @DataDog/apm-sdk-api-python
 ddtrace/internal/opentelemetry                     @DataDog/apm-sdk-api-python
@@ -179,9 +178,12 @@ ddtrace/sampler.py                                 @DataDog/apm-sdk-api-python
 ddtrace/sampling_rule.py                           @DataDog/apm-sdk-api-python
 ddtrace/internal/sampling.py                       @DataDog/apm-sdk-api-python
 ddtrace/internal/tracemethods.py                   @DataDog/apm-sdk-api-python
-ddttace/settings/_otel_remapper.py                 @DataDog/apm-sdk-api-python
+ddtrace/runtime/                                   @DataDog/apm-sdk-api-python
+ddtrace/internal/runtime/                          @DataDog/apm-sdk-api-python
+ddtrace/settings/_otel_remapper.py                 @DataDog/apm-sdk-api-python
 tests/integration/test_priority_sampling.py        @DataDog/apm-sdk-api-python
 tests/integration/test_propagation.py              @DataDog/apm-sdk-api-python
+tests/runtime/                                     @DataDog/apm-sdk-api-python
 tests/test_sampling.py                             @DataDog/apm-sdk-api-python
 tests/test_tracemethods.py                         @DataDog/apm-sdk-api-python
 tests/opentelemetry/                               @DataDog/apm-sdk-api-python

ddtrace/internal/runtime/runtime_metrics.py

@@ -7,7 +7,6 @@
 import ddtrace
 from ddtrace.internal import atexit
 from ddtrace.internal import forksafe
-from ddtrace.internal import telemetry
 from ddtrace.vendor.dogstatsd import DogStatsd
 
 from .. import periodic
@@ -20,8 +19,6 @@
 from .tag_collectors import TracerTagCollector
 
 
-TELEMETRY_RUNTIMEMETRICS_ENABLED = "DD_RUNTIME_METRICS_ENABLED"
-
 log = get_logger(__name__)
 
 
@@ -75,13 +72,13 @@ class RuntimeWorker(periodic.PeriodicService):
     _instance = None  # type: ClassVar[Optional[RuntimeWorker]]
     _lock = forksafe.Lock()
 
-    def __init__(self, interval=_get_interval_or_default(), tracer=ddtrace.tracer, dogstatsd_url=None) -> None:
+    def __init__(self, interval=_get_interval_or_default(), tracer=None, dogstatsd_url=None) -> None:
         super().__init__(interval=interval)
         self.dogstatsd_url: Optional[str] = dogstatsd_url
         self._dogstatsd_client: DogStatsd = get_dogstatsd_client(
             self.dogstatsd_url or ddtrace.internal.agent.get_stats_url()
         )
-        self.tracer: Optional[ddtrace.trace.Tracer] = tracer
+        self.tracer: ddtrace.trace.Tracer = tracer or ddtrace.tracer
         self._runtime_metrics: RuntimeMetrics = RuntimeMetrics()
         self._platform_tags: List[str] = self._format_tags(PlatformTags())
 
@@ -107,9 +104,6 @@ def disable(cls):
             cls._instance = None
             cls.enabled = False
 
-        # Report status to telemetry
-        telemetry.telemetry_writer.add_configuration(TELEMETRY_RUNTIMEMETRICS_ENABLED, False, origin="unknown")
-
     @classmethod
     def _restart(cls):
         cls.disable()
@@ -132,9 +126,6 @@ def enable(cls, flush_interval=None, tracer=None, dogstatsd_url=None):
             cls._instance = runtime_worker
             cls.enabled = True
 
-        # Report status to telemetry
-        telemetry.telemetry_writer.add_configuration(TELEMETRY_RUNTIMEMETRICS_ENABLED, True, origin="unknown")
-
     def flush(self):
         # type: () -> None
         # Ensure runtime metrics have up-to-date tags (ex: service, env, version)

ddtrace/runtime/__init__.py

@@ -1,6 +1,10 @@
 from typing import Optional  # noqa:F401
 
 import ddtrace.internal.runtime.runtime_metrics
+from ddtrace.internal.telemetry import telemetry_writer
+
+
+TELEMETRY_RUNTIMEMETRICS_ENABLED = "DD_RUNTIME_METRICS_ENABLED"
 
 
 class _RuntimeMetricsStatus(type):
@@ -38,7 +42,7 @@ def enable(tracer=None, dogstatsd_url=None, flush_interval=None):
         :param dogstatsd_url: The DogStatsD URL.
         :param flush_interval: The flush interval.
         """
-
+        telemetry_writer.add_configuration(TELEMETRY_RUNTIMEMETRICS_ENABLED, True, origin="code")
         ddtrace.internal.runtime.runtime_metrics.RuntimeWorker.enable(
             tracer=tracer, dogstatsd_url=dogstatsd_url, flush_interval=flush_interval
         )
@@ -52,6 +56,7 @@ def disable():
         Once disabled, runtime metrics can be re-enabled by calling ``enable``
         again.
         """
+        telemetry_writer.add_configuration(TELEMETRY_RUNTIMEMETRICS_ENABLED, False, origin="code")
         ddtrace.internal.runtime.runtime_metrics.RuntimeWorker.disable()
 
 

tests/runtime/test_runtime_metrics_api.py

@@ -51,51 +51,22 @@ def test_manually_start_runtime_metrics_telemetry(test_agent_session, run_python
     """
     code = """
 from ddtrace.internal.telemetry import telemetry_writer
-telemetry_writer.start()
 
 from ddtrace.runtime import RuntimeMetrics
 
 assert not RuntimeMetrics._enabled
 RuntimeMetrics.enable()
 assert RuntimeMetrics._enabled
-
-telemetry_writer.stop()
-telemetry_writer.join(3)
+telemetry_writer.periodic(force_flush=True)
     """
 
-    def find_telemetry_event(events, request_type):
-        e = [e for e in events if e["request_type"] == request_type]
-        assert len(e) == 1
-        return e[0]
-
     _, stderr, status, _ = run_python_code_in_subprocess(code)
     assert status == 0, stderr
 
-    events = test_agent_session.get_events(subprocess=True)
-    # app-started, app-closing, app-client-configuration-change, app-dependencies-loaded
-    assert len(events) == 4
-
-    # Note: the initial app-started event is going to say that it is not enabled, because
-    #       we only look at the env variable DD_RUNTIME_METRICS_ENABLED to set the initial
-    #       value.
-    #
-    #       This test helps validate that manually enabling runtime metrics via code will
-    #       will report the correct config status.
-    app_started_event = find_telemetry_event(events, "app-started")
-    config_changed_event = find_telemetry_event(events, "app-client-configuration-change")
-
-    runtimemetrics_enabled = [
-        c for c in app_started_event["payload"]["configuration"] if c["name"] == "runtimemetrics_enabled"
-    ]
+    runtimemetrics_enabled = test_agent_session.get_configurations("DD_RUNTIME_METRICS_ENABLED")
     assert len(runtimemetrics_enabled) == 1
-    assert runtimemetrics_enabled[0]["value"] is False
-    assert runtimemetrics_enabled[0]["origin"] == "unknown"
-
-    config = config_changed_event["payload"]["configuration"]
-    assert len(config) == 1
-    assert config[0]["name"] == "runtimemetrics_enabled"
-    assert config[0]["value"] is True
-    assert config[0]["origin"] == "unknown"
+    assert runtimemetrics_enabled[0]["value"]
+    assert runtimemetrics_enabled[0]["origin"] == "code"
 
 
 def test_manually_stop_runtime_metrics_telemetry(test_agent_session, ddtrace_run_python_code_in_subprocess):
@@ -105,53 +76,24 @@ def test_manually_stop_runtime_metrics_telemetry(test_agent_session, ddtrace_run
     """
     code = """
 from ddtrace.internal.telemetry import telemetry_writer
-telemetry_writer.start()
 
 from ddtrace.runtime import RuntimeMetrics
 
 assert RuntimeMetrics._enabled
 RuntimeMetrics.disable()
 assert not RuntimeMetrics._enabled
-
-telemetry_writer.stop()
-telemetry_writer.join(3)
+telemetry_writer.periodic(force_flush=True)
     """
 
-    def find_telemetry_event(events, request_type):
-        e = [e for e in events if e["request_type"] == request_type]
-        assert len(e) == 1
-        return e[0]
-
     env = os.environ.copy()
     env["DD_RUNTIME_METRICS_ENABLED"] = "true"
     _, stderr, status, _ = ddtrace_run_python_code_in_subprocess(code, env=env)
     assert status == 0, stderr
 
-    events = test_agent_session.get_events(subprocess=True)
-    # app-started, app-closing, app-client-configuration-change, app-integrations-change, app-dependencies-loaded
-    assert len(events) == 5
-
-    # Note: the initial app-started event is going to say that it is enabled, because we
-    #       set DD_RUNTIME_METRICS_ENABLED=true and are using ddtrace-run which will enable
-    #       runtime metrics for us.
-    #
-    #       This test helps validate that manually disabling runtime metrics via code will
-    #       will report the correct config status.
-    app_started_event = find_telemetry_event(events, "app-started")
-    config_changed_event = find_telemetry_event(events, "app-client-configuration-change")
-
-    runtimemetrics_enabled = [
-        c for c in app_started_event["payload"]["configuration"] if c["name"] == "runtimemetrics_enabled"
-    ]
+    runtimemetrics_enabled = test_agent_session.get_configurations("DD_RUNTIME_METRICS_ENABLED")
     assert len(runtimemetrics_enabled) == 1
-    assert runtimemetrics_enabled[0]["value"] is True
-    assert runtimemetrics_enabled[0]["origin"] == "unknown"
-
-    config = config_changed_event["payload"]["configuration"]
-    assert len(config) == 1
-    assert config[0]["name"] == "runtimemetrics_enabled"
-    assert config[0]["value"] is False
-    assert config[0]["origin"] == "unknown"
+    assert runtimemetrics_enabled[0]["value"] is False
+    assert runtimemetrics_enabled[0]["origin"] == "code"
 
 
 def test_start_runtime_metrics_via_env_var(monkeypatch, ddtrace_run_python_code_in_subprocess):

tests/telemetry/test_writer.py

@@ -112,7 +112,7 @@ def test_app_started_event(telemetry_writer, test_agent_session, mock_time):
                     {"name": "DD_PROFILING_MAX_FRAMES", "origin": "unknown", "value": 64},
                     {"name": "DD_REMOTE_CONFIGURATION_ENABLED", "origin": "unknown", "value": False},
                     {"name": "DD_REMOTE_CONFIG_POLL_INTERVAL_SECONDS", "origin": "unknown", "value": 5.0},
-                    {"name": "DD_RUNTIME_METRICS_ENABLED", "origin": "unknown", "value": False},
+                    {"name": "DD_RUNTIME_METRICS_ENABLED", "origin": "unknown", "value": True},
                     {"name": "DD_SERVICE_MAPPING", "origin": "unknown", "value": ""},
                     {"name": "DD_SPAN_SAMPLING_RULES", "origin": "unknown", "value": None},
                     {"name": "DD_SPAN_SAMPLING_RULES_FILE", "origin": "unknown", "value": None},
@@ -434,7 +434,7 @@ def test_app_started_event_configuration_override(test_agent_session, run_python
         {"name": "DD_PROFILING__FORCE_LEGACY_EXPORTER", "origin": "default", "value": False},
         {"name": "DD_REMOTE_CONFIGURATION_ENABLED", "origin": "env_var", "value": True},
         {"name": "DD_REMOTE_CONFIG_POLL_INTERVAL_SECONDS", "origin": "env_var", "value": 1.0},
-        {"name": "DD_RUNTIME_METRICS_ENABLED", "origin": "unknown", "value": False},
+        {"name": "DD_RUNTIME_METRICS_ENABLED", "origin": "env_var", "value": True},
         {"name": "DD_SERVICE", "origin": "default", "value": DEFAULT_DDTRACE_SUBPROCESS_TEST_SERVICE_NAME},
         {"name": "DD_SERVICE_MAPPING", "origin": "env_var", "value": "default_dd_service:remapped_dd_service"},
         {"name": "DD_SITE", "origin": "env_var", "value": "datadoghq.com"},