perf: Change query-exporting to use generators instead of expanding fully into memory (onyx-dot-app#4729)

raunakab · greptile-apps[bot] · web-flow · commit 5bf4dab47615 · 2025-05-19T20:09:45.000Z
* Change query-exporting to use generators instead of expanding fully into memory

* Fix pagination logic

Co-authored-by: greptile-apps[bot] &lt;165735046+greptile-apps[bot]@users.noreply.github.com&gt;

* Add type annotation

* Add early break if list of chat_sessions is empty

---------

Co-authored-by: greptile-apps[bot] &lt;165735046+greptile-apps[bot]@users.noreply.github.com&gt;
diff --git a/backend/ee/onyx/background/celery/apps/heavy.py b/backend/ee/onyx/background/celery/apps/heavy.py
@@ -9,7 +9,6 @@
 from ee.onyx.background.task_name_builders import query_history_task_name
 from ee.onyx.server.query_history.api import fetch_and_process_chat_session_history
 from ee.onyx.server.query_history.api import ONYX_ANONYMIZED_EMAIL
-from ee.onyx.server.query_history.models import ChatSessionSnapshot
 from ee.onyx.server.query_history.models import QuestionAnswerPairSnapshot
 from onyx.background.celery.apps.heavy import celery_app
 from onyx.background.task_utils import construct_query_history_report_name
@@ -45,6 +44,13 @@ def export_query_history_task(self: Task, *, start: datetime, end: datetime) ->
     task_id = self.request.id
     start_time = datetime.now(tz=timezone.utc)
 
+    stream = io.StringIO()
+    writer = csv.DictWriter(
+        stream,
+        fieldnames=list(QuestionAnswerPairSnapshot.model_fields.keys()),
+    )
+    writer.writeheader()
+
     with get_session_with_current_tenant() as db_session:
         try:
             register_task(
@@ -55,15 +61,23 @@ def export_query_history_task(self: Task, *, start: datetime, end: datetime) ->
                 start_time=start_time,
             )
 
-            complete_chat_session_history: list[ChatSessionSnapshot] = (
-                fetch_and_process_chat_session_history(
-                    db_session=db_session,
-                    start=start,
-                    end=end,
-                    feedback_type=None,
-                    limit=None,
-                )
+            snapshot_generator = fetch_and_process_chat_session_history(
+                db_session=db_session,
+                start=start,
+                end=end,
             )
+
+            for snapshot in snapshot_generator:
+                if ONYX_QUERY_HISTORY_TYPE == QueryHistoryType.ANONYMIZED:
+                    snapshot.user_email = ONYX_ANONYMIZED_EMAIL
+
+                writer.writerows(
+                    qa_pair.to_json()
+                    for qa_pair in QuestionAnswerPairSnapshot.from_chat_session_snapshot(
+                        snapshot
+                    )
+                )
+
         except Exception:
             logger.exception(f"Failed to export query history with {task_id=}")
             mark_task_as_finished_with_id(
@@ -73,31 +87,6 @@ def export_query_history_task(self: Task, *, start: datetime, end: datetime) ->
             )
             raise
 
-    if ONYX_QUERY_HISTORY_TYPE == QueryHistoryType.ANONYMIZED:
-        complete_chat_session_history = [
-            ChatSessionSnapshot(
-                **chat_session_snapshot.model_dump(), user_email=ONYX_ANONYMIZED_EMAIL
-            )
-            for chat_session_snapshot in complete_chat_session_history
-        ]
-
-    qa_pairs: list[QuestionAnswerPairSnapshot] = [
-        qa_pair
-        for chat_session_snapshot in complete_chat_session_history
-        for qa_pair in QuestionAnswerPairSnapshot.from_chat_session_snapshot(
-            chat_session_snapshot
-        )
-    ]
-
-    stream = io.StringIO()
-    writer = csv.DictWriter(
-        stream,
-        fieldnames=list(QuestionAnswerPairSnapshot.model_fields.keys()),
-    )
-    writer.writeheader()
-    for row in qa_pairs:
-        writer.writerow(row.to_json())
-
     report_name = construct_query_history_report_name(task_id)
     with get_session_with_current_tenant() as db_session:
         try:
diff --git a/backend/ee/onyx/server/query_history/api.py b/backend/ee/onyx/server/query_history/api.py
@@ -1,3 +1,4 @@
+from collections.abc import Generator
 from datetime import datetime
 from datetime import timezone
 from http import HTTPStatus
@@ -10,7 +11,6 @@
 from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
 
-from ee.onyx.db.query_history import fetch_chat_sessions_eagerly_by_time
 from ee.onyx.db.query_history import get_all_query_history_export_tasks
 from ee.onyx.db.query_history import get_page_of_chat_sessions
 from ee.onyx.db.query_history import get_total_filtered_chat_sessions_count
@@ -45,6 +45,7 @@
 from onyx.server.documents.models import PaginatedReturn
 from onyx.server.query_and_chat.models import ChatSessionDetails
 from onyx.server.query_and_chat.models import ChatSessionsResponse
+from onyx.utils.threadpool_concurrency import parallel_yield
 
 router = APIRouter()
 
@@ -61,41 +62,55 @@ def ensure_query_history_is_enabled(
         )
 
 
+def yield_snapshot_from_chat_session(
+    chat_session: ChatSession,
+    db_session: Session,
+) -> Generator[ChatSessionSnapshot | None]:
+    yield snapshot_from_chat_session(chat_session=chat_session, db_session=db_session)
+
+
 def fetch_and_process_chat_session_history(
     db_session: Session,
     start: datetime,
     end: datetime,
-    feedback_type: QAFeedbackType | None,
     limit: int | None = 500,
-) -> list[ChatSessionSnapshot]:
-    # observed to be slow a scale of 8192 sessions and 4 messages per session
+) -> Generator[ChatSessionSnapshot]:
+    PAGE_SIZE = 100
+
+    page = 0
+    while True:
+        paged_chat_sessions = get_page_of_chat_sessions(
+            start_time=start,
+            end_time=end,
+            db_session=db_session,
+            page_num=page,
+            page_size=PAGE_SIZE,
+        )
 
-    # 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
-    )
+        if not paged_chat_sessions:
+            break
+
+        paged_snapshots = parallel_yield(
+            [
+                yield_snapshot_from_chat_session(
+                    db_session=db_session,
+                    chat_session=chat_session,
+                )
+                for chat_session in paged_chat_sessions
+            ]
+        )
 
-    # 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
-    ]
-
-    valid_snapshots = [
-        snapshot for snapshot in chat_session_snapshots if snapshot is not None
-    ]
-
-    if feedback_type:
-        valid_snapshots = [
-            snapshot
-            for snapshot in valid_snapshots
-            if any(
-                message.feedback_type == feedback_type for message in snapshot.messages
-            )
-        ]
+        for snapshot in paged_snapshots:
+            if snapshot:
+                yield snapshot
+
+        # If we've fetched *less* than a `PAGE_SIZE` worth
+        # of data, we have reached the end of the
+        # pagination sequence; break.
+        if len(paged_chat_sessions) < PAGE_SIZE:
+            break
 
-    return valid_snapshots
+        page += 1
 
 
 def snapshot_from_chat_session(
