Merge pull request #23202 from mubashir1osmani/fix/tag-usage-cost-panel-zero

fix: tag usage cost panel zero

litellm/proxy/management_endpoints/tag_management_endpoints.py

@@ -21,7 +21,6 @@ from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
 from litellm.proxy.management_endpoints.common_daily_activity import (
     SpendAnalyticsPaginatedResponse,
-    compute_tag_metadata_totals,
     get_daily_activity,
 )
 from litellm.proxy.management_helpers.utils import handle_budget_for_entity
@@ -554,5 +553,12 @@ async def get_tag_daily_activity(
         api_key=api_key,
         page=page,
         page_size=page_size,
-        metadata_metrics_func=compute_tag_metadata_totals,
+        # metadata_metrics_func=None because litellm_dailytagspend rows are
+        # pre-aggregated per (date, tag, model, …) and have no request_id.
+        # Deduplication across tags is therefore not possible at this level —
+        # a request tagged with N tags contributes its spend to N separate rows,
+        # so passing compute_tag_metadata_totals would double-count spend when
+        # multiple tags are present.  The panel is primarily used to inspect
+        # individual tags, making this trade-off acceptable.
+        metadata_metrics_func=None,
     )

tests/test_litellm/proxy/management_endpoints/test_common_daily_activity.py

@@ -10,7 +10,6 @@ sys.path.insert(
 
 from litellm.proxy.management_endpoints.common_daily_activity import (
     _is_user_agent_tag,
-    compute_tag_metadata_totals,
     get_api_key_metadata,
     get_daily_activity,
     get_daily_activity_aggregated,
@@ -77,57 +76,6 @@ def test_is_user_agent_tag():
     assert _is_user_agent_tag("user-agent-tag") is False  # no colon
 
 
-def test_compute_tag_metadata_totals():
-    """Test compute_tag_metadata_totals function."""
-    # Create mock records
-    class MockRecord:
-        def __init__(self, request_id, tag, spend, prompt_tokens=10, completion_tokens=5):
-            self.request_id = request_id
-            self.tag = tag
-            self.spend = spend
-            self.prompt_tokens = prompt_tokens
-            self.completion_tokens = completion_tokens
-            self.total_tokens = prompt_tokens + completion_tokens
-            self.cache_read_input_tokens = 0
-            self.cache_creation_input_tokens = 0
-            self.api_requests = 1
-            self.successful_requests = 1
-            self.failed_requests = 0
-
-    # Test deduplication by request_id (keeps max spend)
-    records = [
-        MockRecord("req-1", "production", spend=10.0),
-        MockRecord("req-1", "staging", spend=20.0),  # Higher spend, should be kept
-        MockRecord("req-2", "production", spend=15.0),
-    ]
-    result = compute_tag_metadata_totals(records)
-    assert result.spend == 35.0  # 20.0 + 15.0 (deduplicated req-1)
-    assert result.prompt_tokens == 20  # 10 + 10 (only deduplicated records)
-    assert result.completion_tokens == 10  # 5 + 5 (only deduplicated records)
-
-    # Test ignoring user-agent tags
-    records_with_ua = [
-        MockRecord("req-1", "production", spend=10.0),
-        MockRecord("req-1", "user-agent:chrome", spend=50.0),  # Should be ignored
-        MockRecord("req-2", "staging", spend=15.0),
-    ]
-    result = compute_tag_metadata_totals(records_with_ua)
-    assert result.spend == 25.0  # 10.0 + 15.0 (user-agent ignored)
-
-    # Test ignoring records without request_id
-    records_no_req_id = [
-        MockRecord("req-1", "production", spend=10.0),
-        MockRecord(None, "staging", spend=20.0),  # Should be ignored
-    ]
-    result = compute_tag_metadata_totals(records_no_req_id)
-    assert result.spend == 10.0
-
-    # Test empty records
-    result = compute_tag_metadata_totals([])
-    assert result.spend == 0.0
-    assert result.prompt_tokens == 0
-
-
 @pytest.mark.asyncio
 async def test_get_daily_activity_aggregated_with_endpoint_breakdown():
     """Test that endpoint breakdown is included in aggregated daily activity."""
@@ -405,6 +353,96 @@ async def test_get_api_key_metadata_regenerated_key_uses_most_recent_deleted_rec
     assert result["old-key-hash"]["team_id"] == "latest-team"
 
 
+@pytest.mark.asyncio
+async def test_tag_daily_activity_metadata_totals_not_zero():
+    """Test that tag daily activity returns correct metadata totals.
+
+    Regression test: the tag endpoint previously passed metadata_metrics_func=
+    compute_tag_metadata_totals, which skipped every row whose request_id is
+    NULL.  Rows in litellm_dailytagspend are pre-aggregated and always have
+    NULL request_id, so the totals panel showed $0.  The fix is to pass
+    metadata_metrics_func=None so the fallback aggregation path is used instead.
+    """
+    mock_prisma = MagicMock()
+    mock_prisma.db = MagicMock()
+
+    # Create mock tag spend records (request_id is NULL for aggregated rows)
+    mock_record_1 = MagicMock()
+    mock_record_1.request_id = None  # NULL in aggregated daily rows
+    mock_record_1.tag = "production"
+    mock_record_1.date = "2024-01-01"
+    mock_record_1.api_key = "key-1"
+    mock_record_1.model = "gpt-4"
+    mock_record_1.model_group = "gpt-4"
+    mock_record_1.custom_llm_provider = "openai"
+    mock_record_1.mcp_namespaced_tool_name = None
+    mock_record_1.endpoint = "/chat/completions"
+    mock_record_1.spend = 25.0
+    mock_record_1.prompt_tokens = 500
+    mock_record_1.completion_tokens = 200
+    mock_record_1.cache_read_input_tokens = 0
+    mock_record_1.cache_creation_input_tokens = 0
+    mock_record_1.api_requests = 10
+    mock_record_1.successful_requests = 9
+    mock_record_1.failed_requests = 1
+
+    mock_record_2 = MagicMock()
+    mock_record_2.request_id = None
+    mock_record_2.tag = "staging"
+    mock_record_2.date = "2024-01-01"
+    mock_record_2.api_key = "key-2"
+    mock_record_2.model = "gpt-3.5-turbo"
+    mock_record_2.model_group = "gpt-3.5-turbo"
+    mock_record_2.custom_llm_provider = "openai"
+    mock_record_2.mcp_namespaced_tool_name = None
+    mock_record_2.endpoint = "/chat/completions"
+    mock_record_2.spend = 5.0
+    mock_record_2.prompt_tokens = 300
+    mock_record_2.completion_tokens = 100
+    mock_record_2.cache_read_input_tokens = 0
+    mock_record_2.cache_creation_input_tokens = 0
+    mock_record_2.api_requests = 5
+    mock_record_2.successful_requests = 5
+    mock_record_2.failed_requests = 0
+
+    mock_table = MagicMock()
+    mock_table.count = AsyncMock(return_value=2)
+    mock_table.find_many = AsyncMock(return_value=[mock_record_1, mock_record_2])
+    mock_prisma.db.litellm_dailytagspend = mock_table
+    mock_prisma.db.litellm_verificationtoken = MagicMock()
+    mock_prisma.db.litellm_verificationtoken.find_many = AsyncMock(return_value=[])
+
+    result = await get_daily_activity(
+        prisma_client=mock_prisma,
+        table_name="litellm_dailytagspend",
+        entity_id_field="tag",
+        entity_id=None,
+        entity_metadata_field=None,
+        start_date="2024-01-01",
+        end_date="2024-01-01",
+        model=None,
+        api_key=None,
+        page=1,
+        page_size=1000,
+        metadata_metrics_func=None,  # No custom func — matches the fix
+    )
+
+    # Metadata totals must reflect actual spend, NOT be zero
+    assert result.metadata.total_spend == 30.0  # 25.0 + 5.0
+    assert result.metadata.total_api_requests == 15  # 10 + 5
+    assert result.metadata.total_successful_requests == 14  # 9 + 5
+    assert result.metadata.total_failed_requests == 1
+    assert result.metadata.total_tokens == 1100  # (500+200) + (300+100)
+
+    # Verify breakdown still works
+    assert len(result.results) == 1
+    daily = result.results[0]
+    assert "production" in daily.breakdown.entities
+    assert "staging" in daily.breakdown.entities
+    assert daily.breakdown.entities["production"].metrics.spend == 25.0
+    assert daily.breakdown.entities["staging"].metrics.spend == 5.0
+
+
 @pytest.mark.asyncio
 async def test_aggregated_activity_preserves_metadata_for_deleted_keys():
     """Test that the full aggregation pipeline should preserve metadata for deleted keys."""