`GET /executions` solo leía `executions.jsonl`, donde nunca se escribe una ejecución en `awaiting_approval` (esas viven solo en el checkpointer). La página de Aprobaciones del dashboard hace `[e for e in list_executions() if e.status=="awaiting_approval"]`, así que nunca encontraba aprobaciones pendientes. Ahora `list_executions` reconstruye las no-terminales desde `execution_index.json` + `orchestrator.snapshot()` y las fusiona con las del JSONL. Añade `test_list_executions_incluye_la_awaiting`. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
137 lines
4.8 KiB
Python
137 lines
4.8 KiB
Python
"""Tests del router /executions: invoke, get, list, approve, reject."""
|
|
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
from fastapi.testclient import TestClient
|
|
|
|
from agentforge_core.api import deps
|
|
from agentforge_core.main import create_app
|
|
|
|
FIXTURES = Path(__file__).parent.parent / "fixtures"
|
|
|
|
|
|
@pytest.fixture(autouse=True)
|
|
def patch_settings(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
monkeypatch.setenv("DATA_DIR", str(tmp_path))
|
|
monkeypatch.setenv("AGENTS_DIR", str(FIXTURES / "agents"))
|
|
monkeypatch.setenv("POLICIES_DIR", str(FIXTURES / "policies"))
|
|
monkeypatch.setenv("LLM_PROVIDER", "mock")
|
|
for fn in (
|
|
deps.get_settings,
|
|
deps.get_registry,
|
|
deps.get_policy_store,
|
|
deps.get_orchestrator,
|
|
deps.get_llm_provider,
|
|
deps.get_guardrail_engine,
|
|
):
|
|
fn.cache_clear()
|
|
|
|
|
|
@pytest.fixture
|
|
def client() -> TestClient:
|
|
return TestClient(create_app())
|
|
|
|
|
|
def test_invoke_devuelve_execution_completada(client: TestClient) -> None:
|
|
r = client.post("/agents/incident_analyzer/invoke", json={"input": "degradación MOS pool SBC"})
|
|
assert r.status_code == 200
|
|
body = r.json()
|
|
assert body["status"] == "completed"
|
|
assert body["agent_name"] == "incident_analyzer"
|
|
assert body["agent_version"] == "v1"
|
|
assert body["needs_human_for"] is None
|
|
|
|
|
|
def test_invoke_agente_inexistente_404(client: TestClient) -> None:
|
|
r = client.post("/agents/inexistente/invoke", json={"input": "x"})
|
|
assert r.status_code == 404
|
|
|
|
|
|
def test_invoke_riesgo_alto_pausa_hitl(client: TestClient) -> None:
|
|
r = client.post("/agents/incident_analyzer/invoke", json={"input": "caída registros sip"})
|
|
assert r.status_code == 200
|
|
body = r.json()
|
|
assert body["status"] == "awaiting_approval"
|
|
assert body["needs_human_for"]
|
|
assert {a["id"] for a in body["needs_human_for"]} == {"act-1"}
|
|
|
|
|
|
def test_get_execution_existe_tras_invoke(client: TestClient) -> None:
|
|
trace_id = client.post(
|
|
"/agents/incident_analyzer/invoke", json={"input": "degradación MOS pool SBC"}
|
|
).json()["trace_id"]
|
|
r = client.get(f"/executions/{trace_id}")
|
|
assert r.status_code == 200
|
|
assert r.json()["trace_id"] == trace_id
|
|
|
|
|
|
def test_get_execution_inexistente_404(client: TestClient) -> None:
|
|
r = client.get("/executions/00000000-0000-0000-0000-000000000000")
|
|
assert r.status_code == 404
|
|
|
|
|
|
def test_list_executions_incluye_la_completada(client: TestClient) -> None:
|
|
client.post("/agents/incident_analyzer/invoke", json={"input": "degradación MOS pool SBC"})
|
|
r = client.get("/executions")
|
|
assert r.status_code == 200
|
|
rows = r.json()
|
|
assert len(rows) == 1
|
|
assert rows[0]["status"] == "completed"
|
|
assert rows[0]["n_proposed_actions"] == 1
|
|
|
|
|
|
def test_list_executions_incluye_la_awaiting(client: TestClient) -> None:
|
|
"""Una ejecución pausada en HITL no se escribe en el JSONL; aún así debe listarse
|
|
(la página de Aprobaciones del dashboard depende de ello)."""
|
|
trace_id = client.post(
|
|
"/agents/incident_analyzer/invoke", json={"input": "caída registros sip"}
|
|
).json()["trace_id"]
|
|
rows = client.get("/executions").json()
|
|
awaiting = [r for r in rows if r["trace_id"] == trace_id]
|
|
assert len(awaiting) == 1
|
|
assert awaiting[0]["status"] == "awaiting_approval"
|
|
|
|
|
|
def test_approve_resume_completa_ejecucion(client: TestClient) -> None:
|
|
invoked = client.post(
|
|
"/agents/incident_analyzer/invoke", json={"input": "caída registros sip"}
|
|
).json()
|
|
trace_id = invoked["trace_id"]
|
|
action_ids = [a["id"] for a in invoked["needs_human_for"]]
|
|
r = client.post(
|
|
f"/executions/{trace_id}/approve",
|
|
json={"approved_action_ids": action_ids, "comment": "OK adelante"},
|
|
)
|
|
assert r.status_code == 200
|
|
body = r.json()
|
|
assert body["status"] == "completed"
|
|
assert {a["id"] for a in body["final_output"]["approved_actions"]} == {"act-1"}
|
|
|
|
|
|
def test_reject_marca_failed(client: TestClient) -> None:
|
|
trace_id = client.post(
|
|
"/agents/incident_analyzer/invoke", json={"input": "caída registros sip"}
|
|
).json()["trace_id"]
|
|
r = client.post(f"/executions/{trace_id}/reject", json={"reason": "no procede ahora"})
|
|
assert r.status_code == 200
|
|
body = r.json()
|
|
assert body["status"] == "failed"
|
|
assert "rejected_by_human" in body["error"]
|
|
|
|
|
|
def test_approve_sobre_estado_no_pausado_da_409(client: TestClient) -> None:
|
|
trace_id = client.post(
|
|
"/agents/incident_analyzer/invoke", json={"input": "degradación MOS pool SBC"}
|
|
).json()["trace_id"]
|
|
r = client.post(f"/executions/{trace_id}/approve", json={"approved_action_ids": []})
|
|
assert r.status_code == 409
|
|
|
|
|
|
def test_approve_trace_id_desconocido_da_404(client: TestClient) -> None:
|
|
r = client.post(
|
|
"/executions/11111111-1111-1111-1111-111111111111/approve",
|
|
json={"approved_action_ids": []},
|
|
)
|
|
assert r.status_code == 404
|