refactor(profiling): migrate to product interface

We migrate the profiler product to the product interface for automatic
lifecycle and telemetry management.

ddtrace/bootstrap/preload.py

@@ -6,7 +6,6 @@
 import os  # noqa:I001
 
 from ddtrace import config  # noqa:F401
-from ddtrace.settings.profiling import config as profiling_config  # noqa:F401
 from ddtrace.internal.logger import get_logger  # noqa:F401
 from ddtrace.internal.module import ModuleWatchdog  # noqa:F401
 from ddtrace.internal.products import manager  # noqa:F401
@@ -59,14 +58,6 @@ def register_post_preload(func: t.Callable) -> None:
     except Exception:
         log.error("failed to enable crashtracking", exc_info=True)
 
-
-if profiling_config.enabled:
-    log.debug("profiler enabled via environment variable")
-    try:
-        import ddtrace.profiling.auto  # noqa: F401
-    except Exception:
-        log.error("failed to enable profiling", exc_info=True)
-
 if config._runtime_metrics_enabled:
     RuntimeWorker.enable()
 

ddtrace/profiling/product.py

@@ -0,0 +1,34 @@
+from ddtrace.settings.profiling import config
+
+
+def post_preload():
+    pass
+
+
+def start():
+    if config.enabled:
+        import ddtrace.profiling.auto  # noqa: F401
+
+
+def restart(join=False):
+    if config.enabled:
+        from ddtrace.profiling import bootstrap
+
+        try:
+            bootstrap.profiler._restart_on_fork()
+        except AttributeError:
+            pass
+
+
+def stop(join=False):
+    if config.enabled:
+        from ddtrace.profiling import bootstrap
+
+        try:
+            bootstrap.profiler.stop()
+        except AttributeError:
+            pass
+
+
+def at_exit(join=False):
+    stop(join=join)

ddtrace/profiling/profiler.py

@@ -10,14 +10,9 @@
 
 import ddtrace
 from ddtrace import config
-from ddtrace.internal import atexit
-from ddtrace.internal import forksafe
 from ddtrace.internal import service
-from ddtrace.internal import uwsgi
 from ddtrace.internal.datadog.profiling import ddup
 from ddtrace.internal.module import ModuleWatchdog
-from ddtrace.internal.telemetry import telemetry_writer
-from ddtrace.internal.telemetry.constants import TELEMETRY_APM_PRODUCT
 from ddtrace.profiling import collector
 from ddtrace.profiling import exporter  # noqa:F401
 from ddtrace.profiling import recorder
@@ -46,39 +41,18 @@ class Profiler(object):
     def __init__(self, *args, **kwargs):
         self._profiler = _ProfilerInstance(*args, **kwargs)
 
-    def start(self, stop_on_exit=True, profile_children=True):
-        """Start the profiler.
-
-        :param stop_on_exit: Whether to stop the profiler and flush the profile on exit.
-        :param profile_children: Whether to start a profiler in child processes.
-        """
-
-        if profile_children:
-            try:
-                uwsgi.check_uwsgi(self._restart_on_fork, atexit=self.stop if stop_on_exit else None)
-            except uwsgi.uWSGIMasterProcess:
-                # Do nothing, the start() method will be called in each worker subprocess
-                return
+    def start(self):
+        """Start the profiler."""
 
         self._profiler.start()
 
-        if stop_on_exit:
-            atexit.register(self.stop)
-
-        if profile_children:
-            forksafe.register(self._restart_on_fork)
-
-        telemetry_writer.product_activated(TELEMETRY_APM_PRODUCT.PROFILER, True)
-
     def stop(self, flush=True):
         """Stop the profiler.
 
         :param flush: Flush last profile.
         """
-        atexit.unregister(self.stop)
         try:
             self._profiler.stop(flush)
-            telemetry_writer.product_activated(TELEMETRY_APM_PRODUCT.PROFILER, False)
         except service.ServiceStatusError:
             # Not a best practice, but for backward API compatibility that allowed to call `stop` multiple times.
             pass

pyproject.toml

@@ -63,6 +63,7 @@ ddtrace = "ddtrace.contrib.internal.pytest.plugin"
 "code-origin-for-spans" = "ddtrace.debugging._products.code_origin.span"
 "dynamic-instrumentation" = "ddtrace.debugging._products.dynamic_instrumentation"
 "exception-replay" = "ddtrace.debugging._products.exception_replay"
+"profiling" = "ddtrace.profiling.product"
 "remote-configuration" = "ddtrace.internal.remoteconfig.product"
 "symbol-database" = "ddtrace.internal.symbol_db.product"
 "appsec" = "ddtrace.internal.appsec.product"

tests/profiling/gevent_fork.py

@@ -9,7 +9,7 @@
 from ddtrace.profiling import profiler  # noqa:E402,F401
 
 
-p = profiler.Profiler().start(profile_children=True)
+p = profiler.Profiler().start()
 
 pid = os.fork()
 if pid == 0:

tests/profiling/recorder_fork.py

@@ -6,7 +6,7 @@
 
 
 p = Profiler()
-p.start(profile_children=False, stop_on_exit=False)
+p.start()
 
 
 e = stack_event.StackSampleEvent()

tests/profiling/simple_program_fork.py

@@ -2,8 +2,8 @@
 import sys
 import threading
 
+import ddtrace.auto
 from ddtrace.internal import service
-import ddtrace.profiling.auto
 import ddtrace.profiling.bootstrap
 from ddtrace.profiling.collector import stack_event
 from ddtrace.profiling.collector import threading as cthreading

tests/profiling/test_main.py

@@ -38,10 +38,7 @@ def test_call_script_gevent(monkeypatch):
 
 
 def test_call_script_pprof_output(tmp_path, monkeypatch):
-    """This checks if the pprof output and atexit register work correctly.
-
-    The script does not run for one minute, so if the `stop_on_exit` flag is broken, this test will fail.
-    """
+    """This checks if the pprof output and atexit register work correctly."""
     filename = str(tmp_path / "pprof")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     monkeypatch.setenv("DD_PROFILING_CAPTURE_PCT", "1")
@@ -61,13 +58,14 @@ def test_call_script_pprof_output(tmp_path, monkeypatch):
 @pytest.mark.skipif(sys.platform == "win32", reason="fork only available on Unix")
 def test_fork(tmp_path, monkeypatch):
     filename = str(tmp_path / "pprof")
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
     monkeypatch.setenv("DD_PROFILING_API_TIMEOUT", "0.1")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     monkeypatch.setenv("DD_PROFILING_CAPTURE_PCT", "100")
     stdout, stderr, exitcode, pid = call_program(
         "python", os.path.join(os.path.dirname(__file__), "simple_program_fork.py")
     )
-    assert exitcode == 0
+    assert exitcode == 0, stderr
     child_pid = stdout.decode().strip()
     utils.check_pprof_file(filename + "." + str(pid) + ".1")
     utils.check_pprof_file(filename + "." + str(child_pid) + ".1")

tests/profiling/test_uwsgi.py

@@ -42,8 +42,11 @@ def uwsgi(monkeypatch):
 
 def test_uwsgi_threads_disabled(uwsgi):
     proc = uwsgi()
-    stdout, _ = proc.communicate()
-    assert proc.wait() != 0
+    try:
+        stdout, _ = proc.communicate(timeout=1)
+    except TimeoutExpired:
+        proc.terminate()
+        stdout, _ = proc.communicate()
     assert THREADS_MSG in stdout
 
 
@@ -59,6 +62,7 @@ def test_uwsgi_threads_number_set(uwsgi):
 
 def test_uwsgi_threads_enabled(uwsgi, tmp_path, monkeypatch):
     filename = str(tmp_path / "uwsgi.pprof")
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     proc = uwsgi("--enable-threads")
     worker_pids = _get_worker_pids(proc.stdout, 1)
@@ -71,8 +75,14 @@ def test_uwsgi_threads_enabled(uwsgi, tmp_path, monkeypatch):
 
 
 def test_uwsgi_threads_processes_no_master(uwsgi, monkeypatch):
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
+    monkeypatch.setenv("STOP_AFTER_LOAD", "1")
     proc = uwsgi("--enable-threads", "--processes", "2")
-    stdout, _ = proc.communicate()
+    try:
+        stdout, _ = proc.communicate(timeout=1)
+    except TimeoutExpired:
+        proc.terminate()
+        stdout, _ = proc.communicate()
     assert (
         b"ddtrace.internal.uwsgi.uWSGIConfigError: master option must be enabled when multiple processes are used"
         in stdout
@@ -101,6 +111,7 @@ def _get_worker_pids(stdout, num_worker, num_app_started=1):
 
 def test_uwsgi_threads_processes_master(uwsgi, tmp_path, monkeypatch):
     filename = str(tmp_path / "uwsgi.pprof")
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     proc = uwsgi("--enable-threads", "--master", "--py-call-uwsgi-fork-hooks", "--processes", "2")
     worker_pids = _get_worker_pids(proc.stdout, 2)
@@ -114,6 +125,7 @@ def test_uwsgi_threads_processes_master(uwsgi, tmp_path, monkeypatch):
 
 def test_uwsgi_threads_processes_master_lazy_apps(uwsgi, tmp_path, monkeypatch):
     filename = str(tmp_path / "uwsgi.pprof")
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     proc = uwsgi("--enable-threads", "--master", "--processes", "2", "--lazy-apps")
     worker_pids = _get_worker_pids(proc.stdout, 2, 2)
@@ -127,6 +139,7 @@ def test_uwsgi_threads_processes_master_lazy_apps(uwsgi, tmp_path, monkeypatch):
 
 def test_uwsgi_threads_processes_no_master_lazy_apps(uwsgi, tmp_path, monkeypatch):
     filename = str(tmp_path / "uwsgi.pprof")
+    monkeypatch.setenv("DD_PROFILING_ENABLED", "1")
     monkeypatch.setenv("DD_PROFILING_OUTPUT_PPROF", filename)
     proc = uwsgi("--enable-threads", "--processes", "2", "--lazy-apps")
     worker_pids = _get_worker_pids(proc.stdout, 2, 2)

tests/profiling/uwsgi-app.py

@@ -1,5 +1,13 @@
-import ddtrace.profiling.auto  # noqa:F401
+import os
+
+import ddtrace.auto  # noqa:F401
 
 
 def application():
     pass
+
+
+if os.getenv("STOP_AFTER_LOAD"):
+    import sys
+
+    sys.exit(0)

tests/profiling_v2/simple_program_fork.py

@@ -2,8 +2,8 @@
 import sys
 import threading
 
+import ddtrace.auto
 from ddtrace.internal import service
-import ddtrace.profiling.auto
 import ddtrace.profiling.bootstrap
 import ddtrace.profiling.profiler
 

tests/profiling_v2/test_main.py

@@ -79,6 +79,7 @@ def test_fork(tmp_path):
 
     filename = str(tmp_path / "pprof")
     env = os.environ.copy()
+    env["DD_PROFILING_ENABLED"] = "1"
     env["DD_PROFILING_OUTPUT_PPROF"] = filename
     env["DD_PROFILING_CAPTURE_PCT"] = "100"
     stdout, stderr, exitcode, pid = call_program(