ref(dynamic-sampling): Count unnamed root transactions instead of dropping them

src/sentry/dynamic_sampling/per_org/queries.py

@@ -355,7 +355,9 @@ def get_eap_transaction_volumes(
 
     end_time = datetime.now(UTC)
     start_time = end_time - time_interval
-    transaction_counts_by_project: defaultdict[int, list[tuple[str, float]]] = defaultdict(list)
+    transaction_counts_by_project: defaultdict[int, defaultdict[str, float]] = defaultdict(
+        lambda: defaultdict(float)
+    )
 
     orderby = [
         DynamicSamplingQueryFields.DSC_PROJECT_ID,
@@ -372,7 +374,7 @@ def get_eap_transaction_volumes(
                 projects=config.projects,
                 organization=config.organization,
             ),
-            "query_string": f"{DynamicSamplingQueryFilters.IS_SEGMENT} {DynamicSamplingQueryFields.DSC_PROJECT_ID}:[{root_project_filter}] has:{DynamicSamplingQueryFields.DSC_TRANSACTION}",
+            "query_string": f"{DynamicSamplingQueryFilters.IS_SEGMENT} {DynamicSamplingQueryFields.DSC_PROJECT_ID}:[{root_project_filter}]",
             "selected_columns": [
                 DynamicSamplingQueryFields.DSC_PROJECT_ID,
                 DynamicSamplingQueryFields.DSC_TRANSACTION,
@@ -391,20 +393,26 @@ def get_eap_transaction_volumes(
             "sampling_mode": SAMPLING_MODE_HIGHEST_ACCURACY,
         }
     ):
-        transaction = row.get(DynamicSamplingQueryFields.DSC_TRANSACTION)
         total = _get_aggregate_float(row, DynamicSamplingQueryFields.COUNT)
         if total <= 0:
             continue
 
+        # A root span with no transaction name and one named "" are the same unnamed
+        # transaction, but EAP returns them as separate groups. Coalescing to "" keeps
+        # them a single class in the rebalancing model instead of two, one of which
+        # would carry the misleading name "None".
+        transaction = row.get(DynamicSamplingQueryFields.DSC_TRANSACTION) or ""
+
         project_id = _get_aggregate_int(row, DynamicSamplingQueryFields.DSC_PROJECT_ID)
-        transaction_counts = transaction_counts_by_project[project_id]
-        transaction_counts.append((str(transaction), total))
+        transaction_counts_by_project[project_id][transaction] += total
 
     return [
         ProjectTransactionCounts(
             project_id=project_id,
             org_id=config.organization.id,
-            transaction_counts=transaction_counts,
+            transaction_counts=sorted(
+                transaction_counts.items(), key=lambda item: (-item[1], item[0])
+            ),
         )
         for project_id, transaction_counts in sorted(transaction_counts_by_project.items())
     ]

tests/sentry/dynamic_sampling/per_org/test_queries.py

@@ -2,6 +2,7 @@
 
 from dataclasses import replace
 from datetime import UTC, datetime, timedelta
+from typing import Any
 from unittest.mock import patch
 
 from sentry_protos.snuba.v1.trace_item_attribute_pb2 import ExtrapolationMode
@@ -429,7 +430,7 @@ def test_get_eap_transaction_volumes(self) -> None:
                     project=project,
                     start_ts=timestamp + timedelta(seconds=5),
                 ),
-                # missing dsc.transaction — excluded by the has:sentry.dsc.transaction filter
+                # missing dsc.transaction — counted as the unnamed transaction ""
                 self.create_span(
                     {
                         "is_segment": True,
@@ -465,7 +466,7 @@ def test_get_eap_transaction_volumes(self) -> None:
             ProjectTransactionCounts(
                 org_id=organization.id,
                 project_id=project.id,
-                transaction_counts=[("checkout", 3), ("product", 1)],
+                transaction_counts=[("checkout", 3), ("", 1), ("product", 1)],
             ),
             ProjectTransactionCounts(
                 org_id=organization.id,
@@ -669,6 +670,53 @@ def segment(transaction, offset):
             )
         ]
 
+    def test_get_eap_transaction_volumes_coalesces_empty_dsc_transaction(self) -> None:
+        """
+        A root span with an empty ``sentry.dsc.transaction`` and one with the attribute
+        absent are the same unnamed transaction, but EAP returns them as two groups. Both
+        reach the rebalancing model as a single ``""`` class holding their summed count.
+        """
+        organization = self.create_organization()
+        project = self.create_project(organization=organization)
+        timestamp = before_now(minutes=15)
+
+        def segment(transaction: str | None, offset: int) -> dict[str, Any]:
+            dsc_tags = {} if transaction is None else {"dsc.transaction": transaction}
+            return self.create_span(
+                {
+                    "is_segment": True,
+                    "sentry_tags": {
+                        "transaction": str(transaction),
+                        "dsc.project_id": str(project.id),
+                        **dsc_tags,
+                    },
+                },
+                organization=organization,
+                project=project,
+                start_ts=timestamp + timedelta(seconds=offset),
+            )
+
+        self.store_spans(
+            [
+                segment("checkout", 0),
+                segment("checkout", 1),
+                # Root transaction name set to the empty string.
+                segment("", 2),
+                # Root transaction name absent entirely.
+                segment(None, 3),
+            ]
+        )
+
+        volumes = get_eap_transaction_volumes(self.get_config(organization))
+
+        assert volumes == [
+            ProjectTransactionCounts(
+                org_id=organization.id,
+                project_id=project.id,
+                transaction_counts=[("", 2), ("checkout", 2)],
+            )
+        ]
+
     def test_get_eap_transaction_volumes_project_over_cap_does_not_starve_other_projects(
         self,
     ) -> None: