Bugfix/query history notes (#4204)

* early work in progress

* rename utility script

* move actual data seeding to a shareable function

* add test

* make the test pass with the fix

* fix comment

* slight improvements and notes to query history and seeding

* update test

---------

Co-authored-by: Richard Kuo (Danswer) <rkuo@onyx.app>

backend/ee/onyx/db/usage_export.py

@@ -27,6 +27,8 @@ def get_empty_chat_messages_entries__paginated(
     first element is the most recent timestamp out of the sessions iterated
     - this timestamp can be used to paginate forward in time
     second element is a list of messages belonging to all the sessions iterated
+
+    Only messages of type USER are returned
     """
     chat_sessions = fetch_chat_sessions_eagerly_by_time(
         start=period[0],

backend/ee/onyx/server/query_history/api.py

@@ -48,10 +48,15 @@ def fetch_and_process_chat_session_history(
     feedback_type: QAFeedbackType | None,
     limit: int | None = 500,
 ) -> list[ChatSessionSnapshot]:
+    # observed to be slow a scale of 8192 sessions and 4 messages per session
+
+    # this is a little slow (5 seconds)
     chat_sessions = fetch_chat_sessions_eagerly_by_time(
         start=start, end=end, db_session=db_session, limit=limit
     )
 
+    # this is VERY slow (80 seconds) due to create_chat_chain being called
+    # for each session. Needs optimizing.
     chat_session_snapshots = [
         snapshot_from_chat_session(chat_session=chat_session, db_session=db_session)
         for chat_session in chat_sessions
@@ -246,6 +251,8 @@ def get_query_history_as_csv(
             detail="Query history has been disabled by the administrator.",
         )
 
+    # this call is very expensive and is timing out via endpoint
+    # TODO: optimize call and/or generate via background task
     complete_chat_session_history = fetch_and_process_chat_session_history(
         db_session=db_session,
         start=start or datetime.fromtimestamp(0, tz=timezone.utc),

backend/onyx/db/seeding/chat_history_seeding.py

@@ -1,6 +1,7 @@
 import random
 from datetime import datetime
 from datetime import timedelta
+from logging import getLogger
 
 from onyx.configs.constants import MessageType
 from onyx.db.chat import create_chat_session
@@ -9,6 +10,8 @@ from onyx.db.chat import get_or_create_root_message
 from onyx.db.engine import get_session_with_current_tenant
 from onyx.db.models import ChatSession
 
+logger = getLogger(__name__)
+
 
 def seed_chat_history(num_sessions: int, num_messages: int, days: int) -> None:
     """Utility function to seed chat history for testing.
@@ -19,12 +22,18 @@ def seed_chat_history(num_sessions: int, num_messages: int, days: int) -> None:
     the times.
     """
     with get_session_with_current_tenant() as db_session:
+        logger.info(f"Seeding {num_sessions} sessions.")
         for y in range(0, num_sessions):
             create_chat_session(db_session, f"pytest_session_{y}", None, None)
 
         # randomize all session times
+        logger.info(f"Seeding {num_messages} messages per session.")
         rows = db_session.query(ChatSession).all()
-        for row in rows:
+        for x in range(0, len(rows)):
+            if x % 1024 == 0:
+                logger.info(f"Seeded messages for {x} sessions so far.")
+
+            row = rows[x]
             row.time_created = datetime.utcnow() - timedelta(
                 days=random.randint(0, days)
             )
@@ -34,20 +43,37 @@ def seed_chat_history(num_sessions: int, num_messages: int, days: int) -> None:
 
             root_message = get_or_create_root_message(row.id, db_session)
 
+            current_message_type = MessageType.USER
+            parent_message = root_message
             for x in range(0, num_messages):
+                if current_message_type == MessageType.USER:
+                    msg = f"pytest_message_user_{x}"
+                else:
+                    msg = f"pytest_message_assistant_{x}"
+
                 chat_message = create_new_chat_message(
                     row.id,
-                    root_message,
+                    parent_message,
-                    f"pytest_message_{x}",
+                    msg,
                     None,
                     0,
-                    MessageType.USER,
+                    current_message_type,
                     db_session,
                 )
 
                 chat_message.time_sent = row.time_created + timedelta(
                     minutes=random.randint(0, 10)
                 )
-            db_session.commit()
+
+                db_session.commit()
+
+                current_message_type = (
+                    MessageType.ASSISTANT
+                    if current_message_type == MessageType.USER
+                    else MessageType.USER
+                )
+                parent_message = chat_message
 
         db_session.commit()
+
+        logger.info(f"Seeded messages for {len(rows)} sessions. Finished.")

backend/tests/integration/tests/query_history/test_usage_reports.py

@@ -10,7 +10,9 @@ from onyx.db.seeding.chat_history_seeding import seed_chat_history
 def test_usage_reports(reset: None) -> None:
     EXPECTED_SESSIONS = 2048
     MESSAGES_PER_SESSION = 4
-    EXPECTED_MESSAGES = EXPECTED_SESSIONS * MESSAGES_PER_SESSION
+
+    # divide by 2 because only messages of type USER are returned
+    EXPECTED_MESSAGES = EXPECTED_SESSIONS * MESSAGES_PER_SESSION / 2
 
     seed_chat_history(EXPECTED_SESSIONS, MESSAGES_PER_SESSION, 90)