feat(ai-monitoring): Sort conversation lists by supported aliases (#123735)

AI conversation lists can sort by one supported alias behind
`organizations:gen-ai-conversations-querying-enhancements`. Ordering
applies before pagination and defaults to `-max(timestamp)`, with
conversation ID breaking ties. Custom expressions and multiple sort
entries are rejected while enabled. Start and end timestamp sort aliases
are no longer accepted.

For now, sorting uses only spans matching the search, while returned
summaries include all conversation spans within the selected scope.
Whole-conversation filtering and frontend ordering will follow
separately.

Author: vgrozdanic

src/sentry/ai_monitoring/constants.py

@@ -2,7 +2,7 @@
 
 _AI_CONVERSATIONS_ALIAS_DEFINITIONS = {
     "conversationId": ("gen_ai.conversation.id", "gen_ai.conversation.id"),
-    "timestamp": ("max(timestamp)", "max(timestamp)"),
+    "age": ("max(timestamp)", "max(timestamp)"),
     "duration": (
         "sum_if(`has:gen_ai.operation.type`,span.duration)",
         "duration",
@@ -37,14 +37,10 @@
         "failure_count_if(gen_ai.operation.type,equals,tool)",
         "tool_errors",
     ),
-    "startTimestamp": ("min(precise.start_ts)", "start_timestamp"),
-    "endTimestamp": ("max(precise.finish_ts)", "end_timestamp"),
 }
 
 # Accept response field names, selected aliases, and aliases generated by Explore.
-AI_CONVERSATIONS_ALIASES = {
-    field: alias for field, (_, alias) in _AI_CONVERSATIONS_ALIAS_DEFINITIONS.items()
-}
-for _expression, _alias in _AI_CONVERSATIONS_ALIAS_DEFINITIONS.values():
-    AI_CONVERSATIONS_ALIASES[_alias] = _alias
-    AI_CONVERSATIONS_ALIASES[get_function_alias(_expression)] = _alias
+AI_CONVERSATIONS_FIELDS: dict[str, tuple[str, str]] = {}
+for _field, (_expression, _alias) in _AI_CONVERSATIONS_ALIAS_DEFINITIONS.items():
+    for _accepted_name in (_field, _alias, get_function_alias(_expression)):
+        AI_CONVERSATIONS_FIELDS[_accepted_name] = (_expression, _alias)

src/sentry/ai_monitoring/endpoints/organization_ai_conversations.py

@@ -1,14 +1,16 @@
 import logging
 import re
 from collections import defaultdict
-from collections.abc import Mapping
+from collections.abc import Mapping, Sequence
 from typing import Any, TypedDict
 
 import sentry_sdk
 from drf_spectacular.utils import OpenApiParameter, extend_schema
 from rest_framework.request import Request
 from rest_framework.response import Response
 
+from sentry import features
+from sentry.ai_monitoring.constants import AI_CONVERSATIONS_FIELDS
 from sentry.ai_monitoring.conversation_titles import fetch_conversation_titles
 from sentry.ai_monitoring.serializers import OrganizationAIConversationsSerializer
 from sentry.ai_monitoring.utils import get_first_input_message, get_last_output, timestamp_to_float
@@ -236,7 +238,12 @@ def get(
         except NoProjects:
             return Response(status=404)
 
-        serializer = OrganizationAIConversationsSerializer(data=request.GET)
+        sorting_enabled = features.has(
+            "organizations:gen-ai-conversations-querying-enhancements", organization
+        )
+        serializer = OrganizationAIConversationsSerializer(
+            data=request.GET, context={"sorting_enabled": sorting_enabled}
+        )
         if not serializer.is_valid():
             return Response(as_validation_errors(serializer), status=400)
 
@@ -249,6 +256,7 @@ def data_fn(offset: int, limit: int) -> list[AIConversationResponse]:
                 limit=limit,
                 user_query=validated_data.get("query", ""),
                 sampling_mode=validated_data.get("samplingMode", "NORMAL"),
+                sorts=validated_data["sort"] if sorting_enabled else None,
             )
 
         with handle_query_errors():
@@ -278,12 +286,13 @@ def _get_conversations(
         limit: int,
         user_query: str,
         sampling_mode: SAMPLING_MODES = "NORMAL",
+        sorts: Sequence[str] | None = None,
     ) -> list[AIConversationResponse]:
         base_filter = "has:gen_ai.conversation.id has:gen_ai.operation.type"
         query_string = _build_conversation_query(base_filter, user_query)
 
         conversation_ids_results = self._fetch_conversation_ids(
-            snuba_params, query_string, offset, limit, sampling_mode
+            snuba_params, query_string, offset, limit, sampling_mode, sorts
         )
         conversation_ids = _extract_conversation_ids(conversation_ids_results)
 
@@ -303,24 +312,45 @@ def _fetch_conversation_ids(
         offset: int,
         limit: int,
         sampling_mode: SAMPLING_MODES,
+        sorts: Sequence[str] | None = None,
     ) -> EAPResponse:
+        selected_columns = ["gen_ai.conversation.id", "max(precise.finish_ts)"]
+        orderby = ["-max(precise.finish_ts)"]
+        if sorts is not None:
+            # Keep groups with missing sort attributes: EAP filters for the presence
+            # of at least one selected aggregate attribute. Timestamp is always present.
+            selected_columns = ["gen_ai.conversation.id", "max(timestamp)"]
+            selected_aliases = set(selected_columns)
+            orderby = []
+            for sort in sorts:
+                expression, alias = AI_CONVERSATIONS_FIELDS[sort.removeprefix("-")]
+                orderby.append(("-" if sort.startswith("-") else "") + alias)
+                if alias not in selected_aliases:
+                    selected_columns.append(
+                        expression if expression == alias else f"{expression} as {alias}"
+                    )
+                    selected_aliases.add(alias)
+            if not any(column.removeprefix("-") == "gen_ai.conversation.id" for column in orderby):
+                orderby.append("gen_ai.conversation.id")
+
+        # TODO (vgrozdanic): Sort on whole conversations instead of only matching spans.
         return Spans.run_table_query(
             params=snuba_params,
             query_string=query_string,
-            selected_columns=["gen_ai.conversation.id", "max(precise.finish_ts)"],
-            orderby=["-max(precise.finish_ts)"],
+            selected_columns=selected_columns,
+            orderby=orderby,
             offset=offset,
             limit=limit,
             referrer=Referrer.API_AI_CONVERSATIONS.value,
-            config=SearchResolverConfig(auto_fields=True),
+            config=SearchResolverConfig(auto_fields=True, disable_aggregate_extrapolation=True),
             sampling_mode=sampling_mode,
         )
 
     @trace
     def _get_conversations_data(
         self, snuba_params: SnubaParams, conversation_ids: list[str]
     ) -> list[AIConversationResponse]:
-        config = SearchResolverConfig(auto_fields=True)
+        config = SearchResolverConfig(auto_fields=True, disable_aggregate_extrapolation=True)
         resolver = Spans.get_resolver(snuba_params, config)
 
         # Build queries

src/sentry/ai_monitoring/serializers.py

@@ -2,7 +2,7 @@
 
 from rest_framework import serializers
 
-from sentry.ai_monitoring.constants import AI_CONVERSATIONS_ALIASES
+from sentry.ai_monitoring.constants import AI_CONVERSATIONS_FIELDS
 from sentry.search.events.types import SAMPLING_MODES
 
 
@@ -13,9 +13,7 @@ class AIConversationsQuery(TypedDict):
 
 
 class OrganizationAIConversationsSerializer(serializers.Serializer[AIConversationsQuery]):
-    sort = serializers.ListField(
-        child=serializers.CharField(), required=False, default=["-timestamp"]
-    )
+    sort = serializers.ListField(child=serializers.CharField(), required=False, default=["-age"])
     query = serializers.CharField(required=False, allow_blank=True)
     samplingMode = serializers.ChoiceField(
         choices=[
@@ -28,7 +26,9 @@ class OrganizationAIConversationsSerializer(serializers.Serializer[AIConversatio
     )
 
     def validate_sort(self, value: list[str]) -> list[str]:
+        if self.context.get("sorting_enabled") and len(value) != 1:
+            raise serializers.ValidationError("Provide exactly one sort option.")
         for sort in value:
-            if sort.removeprefix("-") not in AI_CONVERSATIONS_ALIASES:
+            if sort.removeprefix("-") not in AI_CONVERSATIONS_FIELDS:
                 raise serializers.ValidationError(f"Invalid sort option: {sort}")
         return value

src/sentry/features/temporary.py

@@ -471,6 +471,8 @@ def register_temporary_features(manager: FeatureManager) -> None:
     manager.add("organizations:trace-waterfall-version-message", OrganizationFeature, FeatureHandlerStrategy.FLAGPOLE, api_expose=True)
     # Enable Conversation focused views in AI Insights
     manager.add("organizations:gen-ai-conversations", OrganizationFeature, FeatureHandlerStrategy.FLAGPOLE, api_expose=True)
+    # Enable query enhancements for AI conversation lists
+    manager.add("organizations:gen-ai-conversations-querying-enhancements", OrganizationFeature, FeatureHandlerStrategy.FLAGPOLE, api_expose=True)
     # Enable AI conversation title generation from gen_ai spans
     manager.add("organizations:gen-ai-conversation-title-generation", OrganizationFeature, FeatureHandlerStrategy.FLAGPOLE, api_expose=False)
     # Enables Conduit demo endpoint and UI