from unittest.mock import patch

from scripts.deploy.gcp.deploy_brain import main
from tests.deploy.gcloud_fake import BRAIN_URL, gcloud_responder


@patch("subprocess.run")
def test_deploy_brain_calls_gcloud(mock_run):
    # `preflight_text_serving` lit un service en --format=json avant de
    # construire : un stdout unique casserait sur json.loads.
    mock_run.side_effect = gcloud_responder(brain_url=BRAIN_URL)

    # Exécute sans crash
    main()

    # Récupère toutes les commandes appelées
    calls = [call[0][0] for call in mock_run.call_args_list]

    # On aplatit les listes de commandes passées pour la vérification
    flat_calls = [" ".join(cmd) for cmd in calls]

    # Vérifie qu'on a appelé gcloud builds submit
    assert any("builds submit" in c for c in flat_calls)

    # Vérifie qu'on a appelé gcloud run deploy
    assert any("run deploy" in c for c in flat_calls)

    # Vérifie qu'on a décrit le service pour obtenir l'URL
    assert any("run services describe" in c for c in flat_calls)

    # Vérifie qu'on a mis à jour le service web
    assert any("run services update" in c for c in flat_calls)


@patch("subprocess.run")
def test_deploy_brain_sets_explicit_scaling_bounds(mock_run):
    mock_run.side_effect = gcloud_responder(brain_url=BRAIN_URL)

    main()

    flat_calls = [" ".join(call[0][0]) for call in mock_run.call_args_list]
    deploy_call = next(c for c in flat_calls if "run deploy" in c)
    # Explicit scale-to-zero (no idle GPU billing) + cost ceiling aligned with
    # restore_brain_service default (3).
    assert "--min-instances=0" in deploy_call
    assert "--max-instances=3" in deploy_call