smol-neuron-studio / engine /tests /test_runtime.py
lorcapoul's picture
Smol Neuron Studio
0624276 verified
Raw History Blame Contribute Delete
3.84 kB
import json
import threading
import time
import pytest
import app.main as main
from app.runtime import Busy, Turnstile, Warmup
# --- Turnos -------------------------------------------------------------------------
def test_turnos_en_orden_de_llegada():
t = Turnstile(max_waiting=5)
a, b, c = t.take(), t.take(), t.take()
assert [t.ahead(x) for x in (a, b, c)] == [0, 1, 2]
assert t.wait(a, 0) and not t.wait(b, 0)
t.release(a)
assert t.wait(b, 0) and t.ahead(c) == 1
t.release(b)
t.release(c)
assert t.waiting == 0
def test_quien_se_va_antes_de_su_turno_no_bloquea():
t = Turnstile(max_waiting=5)
a, b, c = t.take(), t.take(), t.take()
t.release(b) # se fue esperando
assert t.ahead(c) == 1
t.release(a)
assert t.wait(c, 0), "c pasa directo: b ya no está"
t.release(c)
d = t.take()
assert t.wait(d, 0)
def test_cola_llena():
t = Turnstile(max_waiting=1)
t.take() # atendido
t.take() # esperando
with pytest.raises(Busy):
t.take()
def test_espera_hasta_su_turno_entre_hilos():
t = Turnstile(max_waiting=5)
a, b = t.take(), t.take()
threading.Timer(0.2, t.release, args=[a]).start()
start = time.monotonic()
assert t.wait(b, 5)
assert time.monotonic() - start >= 0.15
# --- Arranque -------------------------------------------------------------------------
def test_warmup_carga_una_sola_vez_y_reintenta_si_falla():
calls = []
def load():
calls.append(1)
if len(calls) == 1:
raise RuntimeError("boom")
w = Warmup(load)
w.start().join()
assert not w.ready and "boom" in w.error
w.run()
assert w.ready and w.error is None
w.run()
assert len(calls) == 2
def test_info_responde_503_mientras_carga(client, monkeypatch):
gate = threading.Event()
slow = Warmup(lambda: gate.wait(5))
monkeypatch.setattr(main, "warmup", slow)
slow.start()
r = client.get("/info")
assert r.status_code == 503 and r.json()["detail"] == "Warming up" and r.headers["retry-after"] == "2"
gate.set()
slow.run()
r = client.get("/info")
assert r.status_code == 200
assert r.json()["limits"]["max_queue"] == main.MAX_QUEUE
# --- Cola y límites en /generate/stream ----------------------------------------------
BODY = {"prompt": "Name a color.", "max_new_tokens": 4,
"interventions": [{"layer": 22, "unit": 333, "mode": "add", "value": 20}]}
def read_events(r):
return [json.loads(line[6:]) for line in r.iter_lines() if line.startswith("data: ")]
def test_stream_avisa_su_lugar_en_la_cola(client, monkeypatch):
t = Turnstile(max_waiting=4)
monkeypatch.setattr(main, "turnstile", t)
busy = t.take() # alguien está generando
threading.Timer(1.5, t.release, args=[busy]).start()
with client.stream("POST", "/generate/stream", json=BODY) as r:
events = read_events(r)
assert events[0] == {"type": "queued", "ahead": 1}
assert events[-1]["type"] == "done"
assert t.waiting == 0, "el turno se libera al terminar"
def test_stream_con_cola_llena_responde_503(client, monkeypatch):
t = Turnstile(max_waiting=0)
monkeypatch.setattr(main, "turnstile", t)
t.take()
r = client.post("/generate/stream", json=BODY)
assert r.status_code == 503 and "waiting" in r.json()["detail"]
def test_stream_corta_al_pasar_el_tiempo_maximo(client, monkeypatch):
monkeypatch.setattr(main, "turnstile", Turnstile(max_waiting=4))
monkeypatch.setattr(main, "MAX_REQUEST_SECONDS", 0)
with client.stream("POST", "/generate/stream", json=BODY) as r:
events = read_events(r)
assert events[-1]["type"] == "error" and "Time limit" in events[-1]["detail"]
assert not any(e["type"] == "done" for e in events)
assert main.turnstile.waiting == 0