fix(lancedb): capture span attributes for positionally-passed arguments

packages/opentelemetry-instrumentation-lancedb/opentelemetry/instrumentation/lancedb/utils.py

@@ -1,3 +1,4 @@
+import inspect
 import logging
 import traceback
 from opentelemetry.instrumentation.lancedb.config import Config
@@ -26,3 +27,28 @@ def wrapper(*args, **kwargs):
                 Config.exception_logger(e)
 
     return wrapper
+
+
+def get_call_arguments(wrapped, args, kwargs):
+    """
+    Returns the arguments of a wrapped call as a single name -> value dict.
+
+    SDK methods are commonly called positionally (for example
+    ``table.add(data)``), so reading ``kwargs`` alone would
+    silently drop those arguments. Extra keyword arguments collected by a
+    ``**kwargs`` parameter are flattened into the result. Falls back to
+    ``kwargs`` if the signature cannot be bound.
+    """
+    try:
+        signature = inspect.signature(wrapped)
+        bound = signature.bind_partial(*args, **kwargs)
+        call_args = {}
+        for name, value in bound.arguments.items():
+            kind = signature.parameters[name].kind
+            if kind is inspect.Parameter.VAR_KEYWORD:
+                call_args.update(value)
+            elif kind is not inspect.Parameter.VAR_POSITIONAL:
+                call_args[name] = value
+        return call_args
+    except Exception:
+        return dict(kwargs)

packages/opentelemetry-instrumentation-lancedb/opentelemetry/instrumentation/lancedb/wrapper.py

@@ -1,4 +1,4 @@
-from opentelemetry.instrumentation.lancedb.utils import dont_throw
+from opentelemetry.instrumentation.lancedb.utils import dont_throw, get_call_arguments
 from opentelemetry.semconv.trace import SpanAttributes
 from opentelemetry.semconv.attributes.error_attributes import ERROR_TYPE
 from opentelemetry.trace.status import Status, StatusCode
@@ -42,12 +42,13 @@ def _wrap(tracer, to_wrap, wrapped, instance, args, kwargs):
         span.set_attribute(SpanAttributes.DB_SYSTEM, "lancedb")
         span.set_attribute(SpanAttributes.DB_OPERATION, to_wrap.get("method"))
 
+        call_args = get_call_arguments(wrapped, args, kwargs)
         if to_wrap.get("method") == "add":
-            _set_add_attributes(span, kwargs)
+            _set_add_attributes(span, call_args)
         elif to_wrap.get("method") == "search":
-            _set_search_attributes(span, kwargs)
+            _set_search_attributes(span, call_args)
         elif to_wrap.get("method") == "delete":
-            _set_delete_attributes(span, kwargs)
+            _set_delete_attributes(span, call_args)
 
         try:
             return_value = wrapped(*args, **kwargs)

packages/opentelemetry-instrumentation-lancedb/tests/test_positional_args.py

@@ -0,0 +1,49 @@
+"""Regression tests: span attributes must be captured for positional arguments.
+
+LanceDB documents ``table.add(data)``, ``table.search(query)`` and
+``table.delete(where)`` as positional calls, so the instrumentation must not
+rely on ``kwargs`` alone. These tests use a local, temporary LanceDB.
+"""
+
+import lancedb
+import pytest
+from opentelemetry.semconv_ai import SpanAttributes
+
+ROWS = [
+    {"vector": [1.3, 1.4], "item": "fizz", "price": 100.0},
+    {"vector": [9.5, 56.2], "item": "buzz", "price": 200.0},
+]
+
+
+@pytest.fixture
+def table(tmp_path):
+    db = lancedb.connect(str(tmp_path))
+    return db.create_table("positional_table", data=ROWS)
+
+
+def _span(exporter, name):
+    return next(s for s in exporter.get_finished_spans() if s.name == name)
+
+
+def test_add_positional_args(exporter, table):
+    exporter.clear()
+    table.add(ROWS)
+
+    span = _span(exporter, "lancedb.add")
+    assert span.attributes.get(SpanAttributes.MILVUS_INSERT_DATA_COUNT) == 2
+
+
+def test_search_positional_args(exporter, table):
+    exporter.clear()
+    table.search([100, 100]).limit(2).to_pandas()
+
+    span = _span(exporter, "lancedb.search")
+    assert span.attributes.get(SpanAttributes.MILVUS_SEARCH_FILTER) == "[100, 100]"
+
+
+def test_delete_positional_args(exporter, table):
+    exporter.clear()
+    table.delete("item = 'fizz'")
+
+    span = _span(exporter, "lancedb.delete")
+    assert span.attributes.get(SpanAttributes.CHROMADB_DELETE_WHERE) == "item = 'fizz'"