fix(ray): error when ray.init() was not called [backport 3.17] (#15116)

Backport 0b92c07 from #15087 to 3.17.

Without an explicit `ray.init()` call at the beginning of the file, the
ray job would fail because the `ray_trace_ctx` was not injected at the
right moment.

This PR fixes that

Signed-off-by: Louis Tricot <[email protected]>
Co-authored-by: Louis Tricot <[email protected]>
Co-authored-by: Brett Langdon <[email protected]>

Author: 3 people

ddtrace/contrib/internal/ray/patch.py

@@ -152,10 +152,11 @@ def traced_submit_task(wrapped, instance, args, kwargs):
     # This is done under a lock as multiple task could be submit at the same time
     # and thus try to modify the signature as the same time
     with instance._inject_lock:
-        if instance._function_signature is None:
+        if not getattr(instance._function, "_dd_trace_wrapped", False):
             instance._function = _wrap_remote_function_execution(instance._function)
             instance._function.__signature__ = _inject_dd_trace_ctx_kwarg(instance._function)
             instance._function_signature = extract_signature(instance._function)
+            instance._function._dd_trace_wrapped = True
 
     with tracer.trace(
         "task.submit",
@@ -170,7 +171,11 @@ def traced_submit_task(wrapped, instance, args, kwargs):
             if config.ray.trace_args_kwargs:
                 set_tag_or_truncate(span, RAY_TASK_ARGS, kwargs.get("args", {}))
                 set_tag_or_truncate(span, RAY_TASK_KWARGS, kwargs.get("kwargs", {}))
-            _inject_context_in_kwargs(span.context, kwargs)
+
+            # Check if signature has the trace context parameter
+            has_trace_ctx = DD_RAY_TRACE_CTX in inspect.signature(instance._function).parameters
+            if has_trace_ctx:
+                _inject_context_in_kwargs(span.context, kwargs)
 
             resp = wrapped(*args, **kwargs)
 

releasenotes/notes/fix-task-submission-without-ray-init-ccac8820c43b3505.yaml

@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    ray: This fix resolves an issue where the tracer raised an error when submitting Ray tasks without explicitly calling ``ray.init()``.

tests/contrib/ray/test_ray.py

@@ -278,3 +278,22 @@ def increment(self, x, y=1):
             current_value = ray.get(counter_actor.increment.remote(1, y=2))
 
             assert current_value == 3, f"Unexpected result: {current_value}"
+
+
+class TestRayWithoutInit(TracerTestCase):
+    def tearDown(self):
+        if ray.is_initialized():
+            ray.shutdown()
+        super().tearDown()
+
+    @pytest.mark.snapshot(token="tests.contrib.ray.test_ray.test_task_without_init", ignores=RAY_SNAPSHOT_IGNORES)
+    def test_task_without_init(self):
+        """Test that tracing works when Ray auto-initializes without explicit ray.init()"""
+
+        @ray.remote
+        def add_one(x):
+            return x + 1
+
+        futures = [add_one.remote(i) for i in range(2)]
+        results = ray.get(futures)
+        assert results == [1, 2], f"Unexpected results: {results}"

tests/snapshots/tests.contrib.ray.test_ray.test_task_without_init.json

@@ -0,0 +1,69 @@
+[[
+  {
+    "name": "task.submit",
+    "service": "tests.contrib.ray",
+    "resource": "tests.contrib.ray.test_ray.add_one.remote",
+    "trace_id": 0,
+    "span_id": 1,
+    "parent_id": 0,
+    "type": "ray",
+    "error": 0,
+    "meta": {
+      "_dd.p.dm": "-0",
+      "_dd.p.tid": "69025c6200000000",
+      "component": "ray",
+      "language": "python",
+      "ray.hostname": "docker-desktop",
+      "ray.task.submit_status": "success",
+      "runtime-id": "467a8549e29149d3a0a561b9407dc4f6",
+      "span.kind": "producer"
+    },
+    "metrics": {
+      "_dd.ai_obs.enabled": 1,
+      "_dd.djm.enabled": 1,
+      "_dd.filter.kept": 1,
+      "_dd.measured": 1,
+      "_dd.top_level": 1,
+      "_dd.tracer_kr": 1.0,
+      "_sampling_priority_v1": 2,
+      "process_id": 605
+    },
+    "duration": 3830005877,
+    "start": 1761762402453522879
+  }],
+[
+  {
+    "name": "task.submit",
+    "service": "tests.contrib.ray",
+    "resource": "tests.contrib.ray.test_ray.add_one.remote",
+    "trace_id": 1,
+    "span_id": 1,
+    "parent_id": 0,
+    "type": "ray",
+    "error": 0,
+    "meta": {
+      "_dd.p.dm": "-0",
+      "_dd.p.tid": "69025c6600000000",
+      "component": "ray",
+      "language": "python",
+      "ray.hostname": "docker-desktop",
+      "ray.job_id": "01000000",
+      "ray.node_id": "9221982158053df5fa3e8b7cfe1f487b8aef5fb4b18f9fcfc3a76944",
+      "ray.task.submit_status": "success",
+      "ray.worker_id": "01000000ffffffffffffffffffffffffffffffffffffffffffffffff",
+      "runtime-id": "467a8549e29149d3a0a561b9407dc4f6",
+      "span.kind": "producer"
+    },
+    "metrics": {
+      "_dd.ai_obs.enabled": 1,
+      "_dd.djm.enabled": 1,
+      "_dd.filter.kept": 1,
+      "_dd.measured": 1,
+      "_dd.top_level": 1,
+      "_dd.tracer_kr": 1.0,
+      "_sampling_priority_v1": 2,
+      "process_id": 605
+    },
+    "duration": 389792,
+    "start": 1761762406283924423
+  }]]