Update src/agentcore/orchestrator.py

src/agentcore/orchestrator.py

@@ -108,6 +108,33 @@ _LONG_DIGITS_RE = re.compile(r"\d{6,}")
 _TAG_RE = re.compile(r"<[^>]+>")
 
 
+# Strong references to in-flight analytics POSTs.
+#
+# asyncio keeps only WEAK references to tasks: a task created with
+# create_task() and then dropped can be garbage-collected before it finishes,
+# silently cancelling the request. That is invisible when the network is fast
+# and drops writes unpredictably when the handshake is slow. Holding a
+# reference until the task completes is the documented fix.
+_ANALYTICS_TASKS = set()
+
+
+def _spawn_background(coro) -> bool:
+    """Run coro in the background, keeping a strong reference. True if scheduled."""
+    try:
+        loop = asyncio.get_running_loop()
+    except RuntimeError:
+        try:
+            loop = asyncio.get_event_loop()
+        except Exception:
+            return False
+    if not loop.is_running():
+        return False
+    task = loop.create_task(coro)
+    _ANALYTICS_TASKS.add(task)
+    task.add_done_callback(_ANALYTICS_TASKS.discard)
+    return True
+
+
 def _redact_for_analytics(text: str, max_len: int = 200) -> str:
     """Privacy pass before anything reaches the analytics store: e-mail
     addresses and long digit runs (IDs, phone numbers) are masked and the
@@ -147,7 +174,6 @@ def _log_agent_query(question: str, response: AgentResponse, start_time: float)
         settings = get_settings()
         worker_url = getattr(settings, "cloudflare_worker_url", "") or ""
         if worker_url:
-            import asyncio as _al
             async def _post_to_worker():
                 try:
                     from src.services.http_client import make_client as _mk
@@ -182,14 +208,12 @@ def _log_agent_query(question: str, response: AgentResponse, start_time: float)
                             )
                         elif _resp.status_code >= 400:
                             logger.warning(f"D1 logging failed: HTTP {_resp.status_code}")
+                        else:
+                            logger.debug("D1 logged OK")
                 except Exception as _exc:
                     logger.warning(f"D1 logging unreachable: {_exc}")
-            try:
-                loop = _al.get_event_loop()
-                if loop.is_running():
-                    loop.create_task(_post_to_worker())
-            except Exception:
-                pass
+            if not _spawn_background(_post_to_worker()):
+                logger.warning("D1 logging skipped: no running event loop")
     except Exception:
         pass