account for in-flight capacity in launch scheduling

agent/nix_builder_autoscaler/tests/test_scheduler.py

@@ -173,6 +173,39 @@ def test_launch_respects_max_slots():
     assert len(launching) == 0
 
 
+def test_launch_does_not_overprovision_when_in_flight_capacity_exists():
+    db, runtime, config, clock, metrics = _make_env(slot_count=4, max_slots=4)
+
+    db.create_reservation("x86_64-linux", "test1", None, 1200)
+    db.create_reservation("x86_64-linux", "test2", None, 1200)
+
+    # Tick 1 launches two slots for two pending reservations.
+    scheduling_tick(db, runtime, config, clock, metrics)
+    launching_after_first_tick = db.list_slots(SlotState.LAUNCHING)
+    assert len(launching_after_first_tick) == 2
+
+    # Tick 2 sees in-flight capacity and should not launch more.
+    scheduling_tick(db, runtime, config, clock, metrics)
+    launching_after_second_tick = db.list_slots(SlotState.LAUNCHING)
+    assert len(launching_after_second_tick) == 2
+    assert len(runtime.list_managed_instances()) == 2
+
+
+def test_in_flight_capacity_uses_max_leases_per_slot():
+    db, runtime, config, clock, metrics = _make_env(slot_count=3, max_slots=3, max_leases=2)
+    db.create_reservation("x86_64-linux", "test1", None, 1200)
+    db.create_reservation("x86_64-linux", "test2", None, 1200)
+
+    # One in-flight slot should represent capacity for two reservations.
+    db.update_slot_state("slot001", SlotState.LAUNCHING, instance_id="i-launching")
+
+    scheduling_tick(db, runtime, config, clock, metrics)
+
+    launching = db.list_slots(SlotState.LAUNCHING)
+    assert len(launching) == 1
+    assert len(runtime.list_managed_instances()) == 0
+
+
 def test_min_slots_maintained():
     db, runtime, config, clock, metrics = _make_env(min_slots=1)