.gitlab/benchmarks/bp-runner.microbenchmarks.fail-on-breach.yml

-Original file line number
+Diff line change
@@ Expand Up / @@ -1152,3 +1152,13 @@ experiments: @@
                 thresholds:
                   - execution_time < 0.37 ms
                   - max_rss_usage < 38.75 MB
+              # forktime
+              - name: forktime-baseline
+                thresholds:
+                  - execution_time < 2.30 ms
+                  - max_rss_usage < 33 MB
+              - name: forktime-configured
+                thresholds:
+                  - execution_time < 10 ms
+                  - max_rss_usage < 60 MB

.gitlab/benchmarks/microbenchmarks.yml

-Original file line number
+Diff line change
@@ Expand Up / @@ -169,7 +169,7 @@ microbenchmarks: @@
             SCENARIOS:
             - "span tracer core_api set_http_meta telemetry_add_metric otel_span otel_sdk_span recursive_computation sampling_rule_matches"
             - "http_propagation_extract http_propagation_inject rate_limiter appsec_iast_aspects appsec_iast_aspects_ospath appsec_iast_aspects_re_module appsec_iast_aspects_split appsec_iast_propagation"
-            - "packages_package_for_root_module_mapping packages_update_imported_dependencies"
+            - "packages_package_for_root_module_mapping packages_update_imported_dependencies fork_time"
           - CPUS_PER_RUN: "2"
             SCENARIOS:
             - "django_simple flask_simple flask_sqli errortracking_django_simple errortracking_flask_sqli startup"
@@ Expand Down @@

benchmarks/fork_time/config.yaml

-Original file line number
+Diff line change
@@ -0,0 +1,4 @@
+    baseline:
+      configure: false
+    configured:
+      configure: true

benchmarks/fork_time/requirements_scenario.txt

Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		flask>=2.0.0

benchmarks/fork_time/scenario.py

-Original file line number
+Diff line change
@@ -0,0 +1,81 @@
+    """
+    Benchmark fork time overhead with ddtrace.
+    Measures the time to fork a process when ddtrace is imported and configured
+    in the parent process (simulating gunicorn preload mode with a real Flask app).
+    """
+    import multiprocessing
+    import os
+    import time
+    import bm
+    def _child_process(conn, parent_fork_time):
+        """Child process
+        Measures fork overhead by recording when the child starts executing
+        and comparing to when the parent initiated the fork.
+        """
+        fork_overhead = time.perf_counter() - parent_fork_time
+        conn.send(fork_overhead)
+        conn.close()
+    class ForkTime(bm.Scenario):
+        """Measure fork overhead with ddtrace."""
+        configure: bool
+        cprofile_loops: int = 0  # Fork benchmarks don't work well with cprofile
+        def __post_init__(self):
+            self._setup_done = False
+        def _one_time_setup(self):
+            """Run ddtrace and Flask setup exactly once per process.
+            _pyperf is called once per warmup/value by pyperf. Moving setup here
+            prevents accumulating Flask app instances and ddtrace instrumentation
+            state across calls, which would artificially inflate fork times.
+            """
+            if self._setup_done:
+                return
+            if self.configure:
+                os.environ["DD_TRACE_ENABLED"] = "true"
+                os.environ["DD_SERVICE"] = "fork-benchmark"
+                os.environ["DD_ENV"] = "benchmark"
+                import ddtrace.auto  # noqa: F401
+            self._setup_flask()
+            self._setup_done = True
+        def _setup_flask(self):
+            from flask import Flask
+            try:
+                app = Flask(__name__)  # noqa: F841
+                @app.route("/")
+                def hello():
+                    return "Hello"
+            except ImportError:
+                pass
+        def _pyperf(self, loops: int) -> float:
+            self._one_time_setup()
+            total = 0.0
+            for _ in range(loops):
+                parent_conn, child_conn = multiprocessing.Pipe()
+                fork_start = time.perf_counter()
+                p = multiprocessing.Process(target=_child_process, args=(child_conn, fork_start))
+                p.start()
+                fork_overhead = parent_conn.recv()
+                p.join()
+                parent_conn.close()
+                total += fork_overhead
+            return total

chore(CI): add benchmark for fork_time #16346

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open

juanjux wants to merge 14 commits into main from juanjux/fork-benchmarks

+97 −1

-Original file line number
+Diff line change
@@ Expand Up / @@ -1152,3 +1152,13 @@ experiments: @@
                 thresholds:
                   - execution_time < 0.37 ms
                   - max_rss_usage < 38.75 MB
+              # forktime
+              - name: forktime-baseline
+                thresholds:
+                  - execution_time < 2.30 ms
+                  - max_rss_usage < 33 MB
+              - name: forktime-configured
+                thresholds:
+                  - execution_time < 10 ms
+                  - max_rss_usage < 60 MB

-Original file line number
+Diff line change
@@ Expand Up / @@ -169,7 +169,7 @@ microbenchmarks: @@
             SCENARIOS:
             - "span tracer core_api set_http_meta telemetry_add_metric otel_span otel_sdk_span recursive_computation sampling_rule_matches"
             - "http_propagation_extract http_propagation_inject rate_limiter appsec_iast_aspects appsec_iast_aspects_ospath appsec_iast_aspects_re_module appsec_iast_aspects_split appsec_iast_propagation"
-            - "packages_package_for_root_module_mapping packages_update_imported_dependencies"
+            - "packages_package_for_root_module_mapping packages_update_imported_dependencies fork_time"
           - CPUS_PER_RUN: "2"
             SCENARIOS:
             - "django_simple flask_simple flask_sqli errortracking_django_simple errortracking_flask_sqli startup"
@@ Expand Down @@

-Original file line number
+Diff line change
@@ -0,0 +1,4 @@
+    baseline:
+      configure: false
+    configured:
+      configure: true

Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		flask>=2.0.0

-Original file line number
+Diff line change
@@ -0,0 +1,81 @@
+    """
+    Benchmark fork time overhead with ddtrace.
+    Measures the time to fork a process when ddtrace is imported and configured
+    in the parent process (simulating gunicorn preload mode with a real Flask app).
+    """
+    import multiprocessing
+    import os
+    import time
+    import bm
+    def _child_process(conn, parent_fork_time):
+        """Child process
+        Measures fork overhead by recording when the child starts executing
+        and comparing to when the parent initiated the fork.
+        """
+        fork_overhead = time.perf_counter() - parent_fork_time
+        conn.send(fork_overhead)
+        conn.close()
+    class ForkTime(bm.Scenario):
+        """Measure fork overhead with ddtrace."""
+        configure: bool
+        cprofile_loops: int = 0  # Fork benchmarks don't work well with cprofile
+        def __post_init__(self):
+            self._setup_done = False
+        def _one_time_setup(self):
+            """Run ddtrace and Flask setup exactly once per process.
+            _pyperf is called once per warmup/value by pyperf. Moving setup here
+            prevents accumulating Flask app instances and ddtrace instrumentation
+            state across calls, which would artificially inflate fork times.
+            """
+            if self._setup_done:
+                return
+            if self.configure:
+                os.environ["DD_TRACE_ENABLED"] = "true"
+                os.environ["DD_SERVICE"] = "fork-benchmark"
+                os.environ["DD_ENV"] = "benchmark"
+                import ddtrace.auto  # noqa: F401
+            self._setup_flask()
+            self._setup_done = True
+        def _setup_flask(self):
+            from flask import Flask
+            try:
+                app = Flask(__name__)  # noqa: F841
+                @app.route("/")
+                def hello():
+                    return "Hello"
+            except ImportError:
+                pass
+        def _pyperf(self, loops: int) -> float:
+            self._one_time_setup()
+            total = 0.0
+            for _ in range(loops):
+                parent_conn, child_conn = multiprocessing.Pipe()
+                fork_start = time.perf_counter()
+                p = multiprocessing.Process(target=_child_process, args=(child_conn, fork_start))
+                p.start()
+                fork_overhead = parent_conn.recv()
+                p.join()
+                parent_conn.close()
+                total += fork_overhead
+            return total

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

chore(CI): add benchmark for fork_time #16346

Diff view

Diff view

There are no files selected for viewing

Uh oh!

chore(CI): add benchmark for fork_time #16346

Are you sure you want to change the base?

chore(CI): add benchmark for fork_time #16346

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!