feat(feedback): cache the feedback summary (#93461)

Author: vishnupsatish

src/sentry/api/endpoints/organization_feedback_summary.py

@@ -13,9 +13,11 @@
 from sentry.api.utils import get_date_range_from_stats_period
 from sentry.exceptions import InvalidParams
 from sentry.feedback.usecases.feedback_summaries import generate_summary
+from sentry.grouping.utils import hash_from_values
 from sentry.issues.grouptype import FeedbackGroup
 from sentry.models.group import Group, GroupStatus
 from sentry.models.organization import Organization
+from sentry.utils.cache import cache
 
 logger = logging.getLogger(__name__)
 
@@ -24,6 +26,9 @@
 # Token limit is 1,048,576 tokens, see https://ai.google.dev/gemini-api/docs/models#gemini-2.0-flash
 MAX_FEEDBACKS_TO_SUMMARIZE_CHARS = 1000000
 
+# One day since the cache key includes the start and end dates at hour granularity
+SUMMARY_CACHE_TIMEOUT = 86400
+
 
 @region_silo_endpoint
 class OrganizationFeedbackSummaryEndpoint(OrganizationEndpoint):
@@ -63,12 +68,30 @@ def get(self, request: Request, organization: Organization) -> Response:
         except InvalidParams:
             raise ParseError(detail="Invalid or missing date range")
 
+        projects = self.get_projects(request, organization)
+
+        # Sort first, then convert each element to a string
+        numeric_project_ids = sorted([project.id for project in projects])
+        project_ids = [str(project_id) for project_id in numeric_project_ids]
+        hashed_project_ids = hash_from_values(project_ids)
+
+        summary_cache_key = f"feedback_summary:{organization.id}:{start.strftime('%Y-%m-%d-%H')}:{end.strftime('%Y-%m-%d-%H')}:{hashed_project_ids}"
+        summary_cache = cache.get(summary_cache_key)
+        if summary_cache:
+            return Response(
+                {
+                    "summary": summary_cache["summary"],
+                    "success": True,
+                    "numFeedbacksUsed": summary_cache["numFeedbacksUsed"],
+                }
+            )
+
         filters = {
             "type": FeedbackGroup.type_id,
             "first_seen__gte": start,
             "first_seen__lte": end,
             "status": GroupStatus.UNRESOLVED,
-            "project__in": self.get_projects(request, organization),
+            "project__in": projects,
         }
 
         groups = Group.objects.filter(**filters).order_by("-first_seen")[
@@ -77,7 +100,13 @@ def get(self, request: Request, organization: Organization) -> Response:
 
         if groups.count() < MIN_FEEDBACKS_TO_SUMMARIZE:
             logger.error("Too few feedbacks to summarize")
-            return Response({"summary": None, "success": False, "numFeedbacksUsed": 0})
+            return Response(
+                {
+                    "summary": None,
+                    "success": False,
+                    "numFeedbacksUsed": 0,
+                }
+            )
 
         # Also cap the number of characters that we send to the LLM
         group_feedbacks = []
@@ -99,6 +128,16 @@ def get(self, request: Request, organization: Organization) -> Response:
             logger.exception("Error generating summary of user feedbacks")
             return Response({"detail": "Error generating summary"}, status=500)
 
+        cache.set(
+            summary_cache_key,
+            {"summary": summary, "numFeedbacksUsed": len(group_feedbacks)},
+            timeout=SUMMARY_CACHE_TIMEOUT,
+        )
+
         return Response(
-            {"summary": summary, "success": True, "numFeedbacksUsed": len(group_feedbacks)}
+            {
+                "summary": summary,
+                "success": True,
+                "numFeedbacksUsed": len(group_feedbacks),
+            }
         )

tests/sentry/api/endpoints/test_organization_feedback_summary.py

@@ -260,3 +260,55 @@ def test_get_feedback_summary_character_limit(self, mock_generate_summary):
         assert response.data["success"] is True
         assert response.data["summary"] == "Test summary of feedback"
         assert response.data["numFeedbacksUsed"] == 12
+
+    @django_db_all
+    @patch(
+        "sentry.api.endpoints.organization_feedback_summary.generate_summary",
+        return_value="Test summary of feedback",
+    )
+    @patch("sentry.api.endpoints.organization_feedback_summary.cache")
+    def test_get_feedback_summary_cache_hit(self, mock_cache, mock_generate_summary):
+        mock_cache.get.return_value = {
+            "summary": "Test cached summary of feedback",
+            "numFeedbacksUsed": 13,
+        }
+
+        for _ in range(15):
+            event = mock_feedback_event(self.project1.id)
+            create_feedback_issue(
+                event, self.project1.id, FeedbackCreationSource.NEW_FEEDBACK_ENVELOPE
+            )
+
+        with self.feature(self.features):
+            response = self.get_success_response(self.org.slug)
+
+        assert response.data["success"] is True
+        assert response.data["summary"] == "Test cached summary of feedback"
+        assert response.data["numFeedbacksUsed"] == 13
+
+        mock_cache.get.assert_called_once()
+        mock_cache.set.assert_not_called()
+
+    @django_db_all
+    @patch(
+        "sentry.api.endpoints.organization_feedback_summary.generate_summary",
+        return_value="Test summary of feedback",
+    )
+    @patch("sentry.api.endpoints.organization_feedback_summary.cache")
+    def test_get_feedback_summary_cache_miss(self, mock_cache, mock_generate_summary):
+        mock_cache.get.return_value = None
+
+        for _ in range(15):
+            event = mock_feedback_event(self.project1.id)
+            create_feedback_issue(
+                event, self.project1.id, FeedbackCreationSource.NEW_FEEDBACK_ENVELOPE
+            )
+
+        with self.feature(self.features):
+            response = self.get_success_response(self.org.slug)
+
+        assert response.data["success"] is True
+        assert response.data["summary"] == "Test summary of feedback"
+        assert response.data["numFeedbacksUsed"] == 15
+        mock_cache.get.assert_called_once()
+        mock_cache.set.assert_called_once()