DEVX-66: fix: calibrate molecule weights from actual CI execution times

src/devx/molecule/distribute_molecule.py

@@ -135,24 +135,74 @@ def build_multi_role_pairs(
 # Heuristic weights for known heavy molecule scenarios.
 # These are estimated from CI run times — scenarios that pull large Docker
 # images or run complex Ansible playbooks take longer.
+#
+# Weights are calibrated from actual CI execution times (converge→destroy):
+#   nextcloud:           ~7.7m  → 15
+#   restore/default:     ~5.7m  → 11
+#   customer-apps:       ~5.5m  → 11
+#   zitadel/default:     ~5.0m  → 10
+#   docker_base/default: ~3.9m  →  8
+#   app_hardening/default: ~1.9m → 4
+#   app_container/default: ~1.8m → 3
+#   postgres-upgrade:    ~1.7m  →  3
+#   observability/default: ~1.7m → 3
+#   storage/default:     ~1.5m  →  3
+#   storage/object-storage: ~1.0m → 2
+#   vaultwarden:         ~0.7m  →  2
+#   simple-app:          ~0.7m  →  2
+#
+# Role-specific weights take priority over scenario-name weights.
+# The (role, scenario) tuple is checked first, then the scenario name
+# alone, then the default weight.
+_ROLE_SCENARIO_WEIGHTS: dict[tuple[str, str], int] = {
+    ("app_container", "nextcloud"): 15,
+    ("app_container", "customer-apps"): 11,
+    ("app_container", "vaultwarden"): 2,
+    ("app_container", "simple-app"): 2,
+    ("app_container", "postgres-upgrade"): 3,
+    ("app_container", "default"): 3,
+    ("restore", "default"): 11,
+    ("zitadel", "default"): 10,
+    ("docker_base", "default"): 8,
+    ("observability", "default"): 3,
+    ("app_hardening", "default"): 4,
+    ("storage", "default"): 3,
+    ("storage", "object-storage"): 2,
+}
+
+# Fallback weights by scenario name only (for single-role projects or
+# scenarios not in the role-specific table).
 _SCENARIO_WEIGHTS: dict[str, int] = {
-    "nextcloud": 10,
+    "nextcloud": 15,
+    "customer-apps": 11,
+    "restore": 11,
+    "zitadel": 10,
+    "docker-base": 8,
+    "postgresql": 3,
+    "postgres-upgrade": 3,
     "gitea": 8,
-    "vaultwarden": 7,
-    "zitadel": 7,
-    "postgresql": 6,
     "redis": 5,
     "backup": 5,
-    "docker-base": 4,
+    "vaultwarden": 2,
+    "simple-app": 2,
+    "object-storage": 2,
     "default": 3,
     "binary": 2,
 }
 _DEFAULT_SCENARIO_WEIGHT = 3
 
 
-def _scenario_weight(scenario: str) -> int:
-    """Estimate a weight for a scenario based on its name."""
+def _scenario_weight(scenario: str, role: str | None = None) -> int:
+    """Estimate a weight for a scenario based on its name and optionally its role.
+
+    Role-specific weights take priority over scenario-name-only weights.
+    """
     s = scenario.lower()
+    if role is not None:
+        r = role.lower()
+        key = (r, s)
+        if key in _ROLE_SCENARIO_WEIGHTS:
+            return _ROLE_SCENARIO_WEIGHTS[key]
     for key, weight in _SCENARIO_WEIGHTS.items():
         if key in s:
             return weight
@@ -181,11 +231,11 @@ def _lpt_distribute[T](items: list[T], weights: list[int], max_runners: int) ->
 def distribute_multi_role(pairs: list[MultiRoleTestPair], max_runners: int) -> list[list[MultiRoleTestPair]]:
     """Split *pairs* into *max_runners* balanced groups using LPT scheduling.
 
-    Each pair is weighted by scenario name heuristics (e.g. ``nextcloud`` is
-    heavier than ``binary``). Pairs are sorted by weight descending and
+    Each pair is weighted by role+scenario heuristics (e.g. ``nextcloud`` is
+    heavier than ``simple-app``). Pairs are sorted by weight descending and
     assigned to the runner with the least total weight.
     """
-    weights = [_scenario_weight(p.scenario) for p in pairs]
+    weights = [_scenario_weight(p.scenario, p.role) for p in pairs]
     return _lpt_distribute(pairs, weights, max_runners)
 
 

tests/unit/test_distribute_molecule.py

@@ -483,7 +483,7 @@ class TestCliMultiRole:
 
 class TestScenarioWeight:
     def test_known_heavy_scenario(self) -> None:
-        assert _scenario_weight("nextcloud") == 10
+        assert _scenario_weight("nextcloud") == 15
         assert _scenario_weight("gitea") == 8
 
     def test_known_light_scenario(self) -> None:
@@ -493,13 +493,32 @@ class TestScenarioWeight:
         assert _scenario_weight("unknown-scenario") == 3
 
     def test_case_insensitive(self) -> None:
-        assert _scenario_weight("NextCloud") == 10
+        assert _scenario_weight("NextCloud") == 15
         assert _scenario_weight("GITEA") == 8
 
     def test_substring_match(self) -> None:
-        assert _scenario_weight("nextcloud-with-redis") == 10
+        assert _scenario_weight("nextcloud-with-redis") == 15
         assert _scenario_weight("custom-gitea-setup") == 8
 
+    def test_role_specific_weight(self) -> None:
+        """Role+scenario pairs take priority over scenario-name-only weights."""
+        assert _scenario_weight("default", "restore") == 11
+        assert _scenario_weight("default", "zitadel") == 10
+        assert _scenario_weight("default", "docker_base") == 8
+        assert _scenario_weight("default", "app_hardening") == 4
+        assert _scenario_weight("default", "app_container") == 3
+        assert _scenario_weight("default", "storage") == 3
+        assert _scenario_weight("default", "observability") == 3
+
+    def test_role_specific_overrides_scenario_name(self) -> None:
+        """vaultwarden has a scenario-name weight of 2, but role-specific is also 2."""
+        assert _scenario_weight("vaultwarden", "app_container") == 2
+        assert _scenario_weight("vaultwarden") == 2
+
+    def test_customer_apps_weight(self) -> None:
+        assert _scenario_weight("customer-apps", "app_container") == 11
+        assert _scenario_weight("customer-apps") == 11
+
 
 class TestLptDistribute:
     def test_equal_weights_produce_even_split(self) -> None: