Bug 1830760 - For manifests selected with very high confidence (e.g. manifests with modified tests), run on all configs

bugbug/models/testselect.py

@@ -280,7 +280,7 @@ def load_failing_together(task: str) -> dict[str, tuple[float, float]]:
 
 
 def select_configs(
-    groups: Collection[str],
+    group_confidences: dict[str, float],
     min_redundancy_confidence: float,
     max_configurations: int = 3,
 ) -> dict[str, list[str]]:
@@ -290,6 +290,12 @@ def select_configs(
     all_configs_by_group = pickle.loads(failing_together[b"$CONFIGS_BY_GROUP$"])
     config_costs = {config: _get_cost(config) for config in all_configs}
 
+    all_groups = group_confidences.keys()
+    high_confidence_groups = {
+        group for group in all_groups if group_confidences.get(group, 0.0) >= 0.99
+    }
+    groups = [group for group in all_groups if group not in high_confidence_groups]
+
     solver = pywraplp.Solver(
         "select_configs", pywraplp.Solver.CBC_MIXED_INTEGER_PROGRAMMING
     )
@@ -305,6 +311,14 @@ def select_configs(
         )
     }
 
+    # Configs used by high-confidence groups are already committed; fix their
+    # config_vars to 1 so the solver treats their fixed cost as already paid.
+    committed_configs = set()
+    for group in high_confidence_groups:
+        committed_configs |= set(all_configs_by_group.get(group, all_configs))
+    for config in committed_configs:
+        solver.Add(config_vars[config] == 1)
+
     equivalence_sets = _get_equivalence_sets(min_redundancy_confidence)
 
     for group in groups:
@@ -374,7 +388,10 @@ def select_configs(
         )
     )
 
-    configs_by_group: dict[str, list[str]] = {}
+    configs_by_group: dict[str, list[str]] = {
+        group: list(all_configs_by_group.get(group, all_configs))
+        for group in high_confidence_groups
+    }
     for group in groups:
         configs_by_group[group] = []
 
@@ -922,7 +939,14 @@ def eval_apply_transforms(
         if granularity == "label":
             selected = reduce_configs(selected, reduction)
         elif granularity == "group":
-            group_configs = select_configs(selected, reduction)
+            group_configs = select_configs(
+                {
+                    name: confidence
+                    for name, confidence in push["all_possibly_selected"].items()
+                    if name in selected
+                },
+                reduction,
+            )
 
     if minimum is not None and len(selected) < minimum:
         remaining = [

http_service/bugbug_http/models.py

@@ -373,7 +373,7 @@ def _analyze_patch(revs: list[bytes], branch: str | None) -> dict:
     for group in test_scheduling.find_manifests_for_paths(REPO_DIR, modified_paths):
         groups[group] = 1.0
 
-    config_groups = testselect.select_configs(groups.keys(), 0.9)
+    config_groups = testselect.select_configs(groups, 0.9)
 
     data = {
         "tasks": tasks,

tests/test_testselect.py

@@ -718,37 +718,28 @@ def test_select_configs(failing_together_config_group: LMDBDict) -> None:
     test_scheduling.close_failing_together_db("config_group")
 
     result = testselect.select_configs(
-        {
-            "group1",
-        },
+        {"group1": 0.0},
         1.0,
     )
     assert len(result) == 1
     assert set(result["group1"]) == {"linux2404-64-asan/debug", "linux2404-64/opt"}
 
     result = testselect.select_configs(
-        {
-            "group2",
-        },
+        {"group2": 0.0},
         1.0,
     )
     assert len(result) == 1
     assert set(result["group2"]) == {"linux2404-64/debug", "linux2404-64/opt"}
 
     result = testselect.select_configs(
-        {
-            "group3",
-        },
+        {"group3": 0.0},
         1.0,
     )
     assert len(result) == 1
     assert set(result["group3"]) == {"windows10/debug", "linux2404-64/opt"}
 
     result = testselect.select_configs(
-        {
-            "group1",
-            "group2",
-        },
+        {"group1": 0.0, "group2": 0.0},
         1.0,
     )
     assert len(result) == 2
@@ -759,33 +750,23 @@ def test_select_configs(failing_together_config_group: LMDBDict) -> None:
     }
 
     result = testselect.select_configs(
-        {
-            "group1",
-            "group3",
-        },
+        {"group1": 0.0, "group3": 0.0},
         1.0,
     )
     assert len(result) == 2
     assert set(result["group1"]) == {"linux2404-64/opt", "linux2404-64-asan/debug"}
     assert set(result["group3"]) == {"windows10/debug", "linux2404-64/opt"}
 
     result = testselect.select_configs(
-        {
-            "group2",
-            "group3",
-        },
+        {"group2": 0.0, "group3": 0.0},
         1.0,
     )
     assert len(result) == 2
     assert set(result["group2"]) == {"linux2404-64/opt", "linux2404-64/debug"}
     assert set(result["group3"]) == {"linux2404-64/opt", "windows10/debug"}
 
     result = testselect.select_configs(
-        {
-            "group1",
-            "group2",
-            "group3",
-        },
+        {"group1": 0.0, "group2": 0.0, "group3": 0.0},
         1.0,
     )
     assert len(result) == 3
@@ -796,3 +777,19 @@ def test_select_configs(failing_together_config_group: LMDBDict) -> None:
         "linux2404-64-asan/debug",
     }
     assert set(result["group3"]) == {"linux2404-64/opt", "windows10/debug"}
+
+    # A group selected with >= 0.99 confidence runs on all its configs.
+    all_configs = {
+        "linux2404-64-asan/debug",
+        "linux2404-64/debug",
+        "linux2404-64/opt",
+        "mac/debug",
+        "windows10/debug",
+    }
+    result = testselect.select_configs(
+        {"group1": 0.99, "group2": 0.0},
+        1.0,
+    )
+    assert len(result) == 2
+    assert set(result["group1"]) == all_configs
+    assert set(result["group2"]) == {"linux2404-64/opt", "linux2404-64/debug"}