Fix ai-service Dockerfile: bake in drug_entities.json, override its path

apps/ai-service/tests/test_budget.py

@@ -0,0 +1,59 @@
+"""`rag/budget.py::RequestBudget` — F-08."""
+from __future__ import annotations
+
+import time
+
+import pytest
+
+from rag.budget import RequestBudget, RequestBudgetExhausted
+from rag.ports import AnswerGenerationUnavailable
+
+
+def test_budget_exhausted_is_a_subclass_of_answer_generation_unavailable():
+    """Deliberate: every existing `except AnswerGenerationUnavailable:`
+    fail-open/fail-closed handler in the codebase must catch this with zero
+    changes, since it predates F-08 and already encodes the right behaviour
+    for "the provider is unavailable to us right now"."""
+    assert issubclass(RequestBudgetExhausted, AnswerGenerationUnavailable)
+
+
+def test_fresh_budget_has_budget():
+    budget = RequestBudget.start(max_wall_clock_ms=20_000, max_calls=5)
+    assert budget.has_budget() is True
+
+
+def test_require_spends_one_call():
+    budget = RequestBudget.start(max_wall_clock_ms=20_000, max_calls=2)
+    budget.require()
+    assert budget.calls_remaining == 1
+    budget.require()
+    assert budget.calls_remaining == 0
+    assert budget.has_budget() is False
+
+
+def test_require_raises_once_calls_are_exhausted():
+    budget = RequestBudget.start(max_wall_clock_ms=20_000, max_calls=1)
+    budget.require()
+    with pytest.raises(RequestBudgetExhausted):
+        budget.require()
+
+
+def test_require_raises_once_the_deadline_has_passed():
+    budget = RequestBudget.start(max_wall_clock_ms=0, max_calls=100)
+    time.sleep(0.01)
+    assert budget.has_budget() is False
+    with pytest.raises(RequestBudgetExhausted):
+        budget.require()
+
+
+def test_a_failed_require_does_not_spend_a_call():
+    """`require()` raises before decrementing when there's no budget left —
+    `calls_remaining` must not go negative, which would otherwise make a
+    budget that's already exhausted look like it has "negative debt" instead
+    of cleanly `0`."""
+    budget = RequestBudget.start(max_wall_clock_ms=20_000, max_calls=1)
+    budget.require()
+    for _ in range(3):
+        with pytest.raises(RequestBudgetExhausted):
+            budget.require()
+    assert budget.calls_remaining == 0