chore: update log adapter to support session-scoped api method logging

Updates `log_adapter` to optionally store API method names in a `Session` instance instead of the global list. This improves label accuracy when running tests in parallel.

- Updates `Session` to initialize `_api_methods` list and lock.
- Updates `log_adapter.add_api_method` and `get_and_reset_api_methods` to handle session-scoped logging.
- Updates `log_adapter.method_logger` and `property_logger` to identify the session from arguments.
- Propagates `session` through `start_query_with_client` and its callers to ensure labels are correctly associated with the session.

Author: google-labs-jules[bot]

bigframes/core/log_adapter.py

@@ -174,7 +174,8 @@ def wrapper(*args, **kwargs):
             full_method_name = f"{base_name.lower()}-{api_method_name}"
             # Track directly called methods
             if len(_call_stack) == 0:
-                add_api_method(full_method_name)
+                session = _find_session(*args, **kwargs)
+                add_api_method(full_method_name, session=session)
 
             _call_stack.append(full_method_name)
 
@@ -220,7 +221,8 @@ def wrapped(*args, **kwargs):
             full_property_name = f"{class_name.lower()}-{property_name.lower()}"
 
             if len(_call_stack) == 0:
-                add_api_method(full_property_name)
+                session = _find_session(*args, **kwargs)
+                add_api_method(full_property_name, session=session)
 
             _call_stack.append(full_property_name)
             try:
@@ -250,25 +252,40 @@ def wrapper(func):
     return wrapper
 
 
-def add_api_method(api_method_name):
+def add_api_method(api_method_name, session=None):
     global _lock
     global _api_methods
-    with _lock:
-        # Push the method to the front of the _api_methods list
-        _api_methods.insert(0, api_method_name.replace("<", "").replace(">", ""))
-        # Keep the list length within the maximum limit (adjust MAX_LABELS_COUNT as needed)
-        _api_methods = _api_methods[:MAX_LABELS_COUNT]
+
+    if session is not None:
+        with session._api_methods_lock:
+            session._api_methods.insert(
+                0, api_method_name.replace("<", "").replace(">", "")
+            )
+            session._api_methods = session._api_methods[:MAX_LABELS_COUNT]
+    else:
+        with _lock:
+            # Push the method to the front of the _api_methods list
+            _api_methods.insert(0, api_method_name.replace("<", "").replace(">", ""))
+            # Keep the list length within the maximum limit (adjust MAX_LABELS_COUNT as needed)
+            _api_methods = _api_methods[:MAX_LABELS_COUNT]
 
 
-def get_and_reset_api_methods(dry_run: bool = False):
+def get_and_reset_api_methods(dry_run: bool = False, session=None):
     global _lock
+    methods = []
+    if session is not None:
+        with session._api_methods_lock:
+            methods.extend(session._api_methods)
+            if not dry_run:
+                session._api_methods.clear()
+
     with _lock:
-        previous_api_methods = list(_api_methods)
+        methods.extend(_api_methods)
 
         # dry_run might not make a job resource, so only reset the log on real queries.
         if not dry_run:
             _api_methods.clear()
-    return previous_api_methods
+    return methods
 
 
 def _get_bq_client(*args, **kwargs):
@@ -283,3 +300,19 @@ def _get_bq_client(*args, **kwargs):
             return kwargv._block.session.bqclient
 
     return None
+
+
+def _find_session(*args, **kwargs):
+    # This function cannot import Session at the top level because Session
+    # imports log_adapter.
+    # We can't import bigframes.session in type checking block either.
+    from bigframes.session import Session
+
+    if args and isinstance(args[0], Session):
+        return args[0]
+
+    session = kwargs.get("session")
+    if session is not None and isinstance(session, Session):
+        return session
+
+    return None

bigframes/session/__init__.py

@@ -208,6 +208,9 @@ def __init__(
         self._session_id: str = "session" + secrets.token_hex(3)
         # store table ids and delete them when the session is closed
 
+        self._api_methods: list[str] = []
+        self._api_methods_lock = threading.Lock()
+
         self._objects: list[
             weakref.ReferenceType[
                 Union[
@@ -2160,6 +2163,7 @@ def _start_query_ml_ddl(
             query_with_job=True,
             job_retry=third_party_gcb_retry.DEFAULT_ML_JOB_RETRY,
             publisher=self._publisher,
+            session=self,
         )
         return iterator, query_job
 
@@ -2188,6 +2192,7 @@ def _create_object_table(self, path: str, connection: str) -> str:
             timeout=None,
             query_with_job=True,
             publisher=self._publisher,
+            session=self,
         )
 
         return table

bigframes/session/_io/bigquery/__init__.py

@@ -228,12 +228,14 @@ def format_option(key: str, value: Union[bool, str]) -> str:
     return f"{key}={repr(value)}"
 
 
-def add_and_trim_labels(job_config):
+def add_and_trim_labels(job_config, session=None):
     """
     Add additional labels to the job configuration and trim the total number of labels
     to ensure they do not exceed MAX_LABELS_COUNT labels per job.
     """
-    api_methods = log_adapter.get_and_reset_api_methods(dry_run=job_config.dry_run)
+    api_methods = log_adapter.get_and_reset_api_methods(
+        dry_run=job_config.dry_run, session=session
+    )
     job_config.labels = create_job_configs_labels(
         job_configs_labels=job_config.labels,
         api_methods=api_methods,
@@ -270,6 +272,7 @@ def start_query_with_client(
     metrics: Optional[bigframes.session.metrics.ExecutionMetrics],
     query_with_job: Literal[True],
     publisher: bigframes.core.events.Publisher,
+    session=None,
 ) -> Tuple[google.cloud.bigquery.table.RowIterator, bigquery.QueryJob]:
     ...
 
@@ -286,6 +289,7 @@ def start_query_with_client(
     metrics: Optional[bigframes.session.metrics.ExecutionMetrics],
     query_with_job: Literal[False],
     publisher: bigframes.core.events.Publisher,
+    session=None,
 ) -> Tuple[google.cloud.bigquery.table.RowIterator, Optional[bigquery.QueryJob]]:
     ...
 
@@ -303,6 +307,7 @@ def start_query_with_client(
     query_with_job: Literal[True],
     job_retry: google.api_core.retry.Retry,
     publisher: bigframes.core.events.Publisher,
+    session=None,
 ) -> Tuple[google.cloud.bigquery.table.RowIterator, bigquery.QueryJob]:
     ...
 
@@ -320,6 +325,7 @@ def start_query_with_client(
     query_with_job: Literal[False],
     job_retry: google.api_core.retry.Retry,
     publisher: bigframes.core.events.Publisher,
+    session=None,
 ) -> Tuple[google.cloud.bigquery.table.RowIterator, Optional[bigquery.QueryJob]]:
     ...
 
@@ -340,14 +346,15 @@ def start_query_with_client(
     # version 3.36.0 or later.
     job_retry: google.api_core.retry.Retry = third_party_gcb_retry.DEFAULT_JOB_RETRY,
     publisher: bigframes.core.events.Publisher,
+    session=None,
 ) -> Tuple[google.cloud.bigquery.table.RowIterator, Optional[bigquery.QueryJob]]:
     """
     Starts query job and waits for results.
     """
     # Note: Ensure no additional labels are added to job_config after this
     # point, as `add_and_trim_labels` ensures the label count does not
     # exceed MAX_LABELS_COUNT.
-    add_and_trim_labels(job_config)
+    add_and_trim_labels(job_config, session=session)
 
     try:
         if not query_with_job:

bigframes/session/bq_caching_executor.py

@@ -323,6 +323,7 @@ def _export_gbq(
         iterator, job = self._run_execute_query(
             sql=sql,
             job_config=job_config,
+            session=array_value.session,
         )
 
         has_timedelta_col = any(
@@ -389,6 +390,7 @@ def _run_execute_query(
         sql: str,
         job_config: Optional[bq_job.QueryJobConfig] = None,
         query_with_job: bool = True,
+        session=None,
     ) -> Tuple[bq_table.RowIterator, Optional[bigquery.QueryJob]]:
         """
         Starts BigQuery query job and waits for results.
@@ -415,6 +417,7 @@ def _run_execute_query(
                     timeout=None,
                     query_with_job=True,
                     publisher=self._publisher,
+                    session=session,
                 )
             else:
                 return bq_io.start_query_with_client(
@@ -427,6 +430,7 @@ def _run_execute_query(
                     timeout=None,
                     query_with_job=False,
                     publisher=self._publisher,
+                    session=session,
                 )
 
         except google.api_core.exceptions.BadRequest as e:
@@ -661,6 +665,7 @@ def _execute_plan_gbq(
             sql=compiled.sql,
             job_config=job_config,
             query_with_job=(destination_table is not None),
+            session=plan.session,
         )
 
         # we could actually cache even when caching is not explicitly requested, but being conservative for now

bigframes/session/direct_gbq_execution.py

@@ -60,6 +60,7 @@ def execute(
 
         iterator, query_job = self._run_execute_query(
             sql=compiled.sql,
+            session=plan.session,
         )
 
         # just immediately downlaod everything for simplicity
@@ -75,6 +76,7 @@ def _run_execute_query(
         self,
         sql: str,
         job_config: Optional[bq_job.QueryJobConfig] = None,
+        session=None,
     ) -> Tuple[bq_table.RowIterator, Optional[bigquery.QueryJob]]:
         """
         Starts BigQuery query job and waits for results.
@@ -89,4 +91,5 @@ def _run_execute_query(
             metrics=None,
             query_with_job=False,
             publisher=self._publisher,
+            session=session,
         )

bigframes/session/loader.py

@@ -1284,6 +1284,7 @@ def _start_query_with_job_optional(
             metrics=None,
             query_with_job=False,
             publisher=self._publisher,
+            session=self._session,
         )
         return rows
 
@@ -1310,6 +1311,7 @@ def _start_query_with_job(
             metrics=None,
             query_with_job=True,
             publisher=self._publisher,
+            session=self._session,
         )
         return query_job