File size: 9,382 Bytes
dfb775d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 | """Hands-free CPU autostart, gated on the mindX SEA agent's decision.
`autostart_cpu_training` has two gates: `MINDXTRAIN_AUTOSTART` arms
autonomous mode, and the StrategicEvolutionAgent's decision file is the
actual decider. Tests patch `coach.api._SPAWN` with a fake so the launch
path runs without a subprocess, and point `MINDXTRAIN_SEA_DECISION` at a
temp file so no real mindX install is touched.
"""
from __future__ import annotations
import json
import time
from collections.abc import Iterator
from pathlib import Path
from typing import Any
import pytest
from fastapi.testclient import TestClient
from mindxtrain.operator import runs as _runs
from mindxtrain.operator.app import app
from mindxtrain.operator.coach import api as coach_api
from mindxtrain.operator.coach import run_metrics as rm
client = TestClient(app)
@pytest.fixture(autouse=True)
def _restore(monkeypatch: pytest.MonkeyPatch) -> Iterator[None]:
"""Restore the real spawn fn + clear all autostart/SEA env each test."""
original = coach_api._SPAWN
for var in (
"MINDXTRAIN_AUTOSTART",
"MINDXTRAIN_AUTOSTART_RECIPE",
"MINDXTRAIN_SEA_DECISION",
"MINDXTRAIN_MINDX_ROOT",
):
monkeypatch.delenv(var, raising=False)
yield
coach_api._SPAWN = original
# Drain any runs the fake spawn left "running" — otherwise the
# idempotency check in autostart_cpu_training trips on stale runs
# carried into the next test by the module-global registry.
for run in coach_api._REGISTRY.list_runs():
if run.status in ("pending", "running"):
coach_api._REGISTRY.publish(
run.id,
_runs.StatusEvent(
run_id=run.id, status="cancelled", message="test teardown",
),
)
def _fake_spawn(run: _runs.Run, _cfg: Any, _plan: Any) -> None:
"""Stand-in spawn — marks the run running without a subprocess."""
coach_api._REGISTRY.publish(
run.id,
_runs.StatusEvent(run_id=run.id, status="running", message="fake"),
)
def _write_decision(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch, **fields: Any,
) -> Path:
"""Write a SEA decision file and point the operator at it."""
record: dict[str, Any] = {
"recommend": True,
"recipe": None,
"reason": "test recommendation",
"decided_by": "strategic_evolution_agent",
"ts": time.time(),
"ttl_s": 3600,
}
record.update(fields)
path = tmp_path / "training_recommendation.json"
path.write_text(json.dumps(record), encoding="utf-8")
monkeypatch.setenv("MINDXTRAIN_SEA_DECISION", str(path))
return path
async def _cleanup_samplers() -> None:
"""Stop any per-run metrics samplers the launch path started."""
for tid in list(rm._TASKS):
await rm.stop_metrics_sampler(tid)
# ---- env-driven gates ----------------------------------------------------
def test_autostart_disabled_by_default() -> None:
assert coach_api.autostart_enabled() is False
@pytest.mark.parametrize("val", ["1", "true", "TRUE", "yes", "on"])
def test_autostart_enabled_recognises_truthy(
monkeypatch: pytest.MonkeyPatch, val: str,
) -> None:
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", val)
assert coach_api.autostart_enabled() is True
def test_autostart_recipe_defaults_to_smoke() -> None:
assert coach_api.autostart_recipe() == "mindx_fallback_qwen3_1_5b_cpu_smoke"
def test_autostart_recipe_honours_override(
monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setenv("MINDXTRAIN_AUTOSTART_RECIPE", "qwen3_8b_sft_lora")
assert coach_api.autostart_recipe() == "qwen3_8b_sft_lora"
# ---- SEA decision gate ---------------------------------------------------
def test_sea_gate_closed_when_no_decision_file(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
monkeypatch.setenv("MINDXTRAIN_SEA_DECISION", str(tmp_path / "missing.json"))
gate = coach_api.sea_training_gate()
assert gate["open"] is False
assert gate["available"] is False
def test_sea_gate_open_on_fresh_recommendation(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
_write_decision(tmp_path, monkeypatch, recommend=True, reason="corpus grew")
gate = coach_api.sea_training_gate()
assert gate["open"] is True
assert "corpus grew" in gate["reason"]
def test_sea_gate_closed_when_sea_says_no(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
_write_decision(tmp_path, monkeypatch, recommend=False, reason="loss still high")
gate = coach_api.sea_training_gate()
assert gate["open"] is False
assert gate["available"] is True
assert "loss still high" in gate["reason"]
def test_sea_gate_closed_when_recommendation_is_stale(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
_write_decision(
tmp_path, monkeypatch,
recommend=True, ts=time.time() - 10_000, ttl_s=60,
)
gate = coach_api.sea_training_gate()
assert gate["open"] is False
assert "stale" in gate["reason"]
def test_sea_gate_handles_corrupt_file(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
bad = tmp_path / "training_recommendation.json"
bad.write_text("{not json", encoding="utf-8")
monkeypatch.setenv("MINDXTRAIN_SEA_DECISION", str(bad))
gate = coach_api.sea_training_gate()
assert gate["open"] is False # unreadable → treated as no decision
# ---- autostart_cpu_training behaviour ------------------------------------
@pytest.mark.asyncio
async def test_autostart_noop_when_disabled() -> None:
"""No env flag → returns None, launches nothing."""
assert await coach_api.autostart_cpu_training() is None
@pytest.mark.asyncio
async def test_autostart_noop_when_sea_gate_closed(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
"""Autostart armed but SEA has not recommended → no launch."""
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
monkeypatch.setenv("MINDXTRAIN_SEA_DECISION", str(tmp_path / "missing.json"))
coach_api._SPAWN = _fake_spawn
assert await coach_api.autostart_cpu_training() is None
@pytest.mark.asyncio
async def test_autostart_noop_when_sea_declines(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
"""SEA decision present but recommend=false → no launch."""
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
_write_decision(tmp_path, monkeypatch, recommend=False)
coach_api._SPAWN = _fake_spawn
assert await coach_api.autostart_cpu_training() is None
@pytest.mark.asyncio
async def test_autostart_launches_when_sea_recommends(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
"""Both gates open → the smoke recipe is launched and a Run comes back."""
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
_write_decision(tmp_path, monkeypatch, recommend=True)
coach_api._SPAWN = _fake_spawn
try:
run = await coach_api.autostart_cpu_training()
assert run is not None
assert run.recipe == "mindx_fallback_qwen3_1_5b_cpu_smoke"
assert run.status == "running"
finally:
await _cleanup_samplers()
@pytest.mark.asyncio
async def test_autostart_uses_sea_chosen_recipe(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
"""SEA's `recipe` field overrides the MINDXTRAIN_AUTOSTART_RECIPE default."""
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
_write_decision(tmp_path, monkeypatch, recommend=True, recipe="qwen3_8b_sft_lora")
coach_api._SPAWN = _fake_spawn
try:
run = await coach_api.autostart_cpu_training()
assert run is not None
assert run.recipe == "qwen3_8b_sft_lora"
finally:
await _cleanup_samplers()
@pytest.mark.asyncio
async def test_autostart_idempotent_when_run_active(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
"""A run already pending/running → autostart skips (no stacked trainers)."""
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
_write_decision(tmp_path, monkeypatch, recommend=True)
coach_api._SPAWN = _fake_spawn
existing = coach_api._REGISTRY.create("qwen3_8b_sft_lora", tmp_path / "r")
assert existing.status == "pending" # counts as active
assert await coach_api.autostart_cpu_training() is None
# ---- /coach/api/sea-decision endpoint ------------------------------------
def test_sea_decision_endpoint_reports_closed_gate(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
monkeypatch.setenv("MINDXTRAIN_SEA_DECISION", str(tmp_path / "missing.json"))
r = client.get("/coach/api/sea-decision")
assert r.status_code == 200
body = r.json()
assert body["open"] is False
assert body["autostart_enabled"] is False
assert "decision_path" in body
def test_sea_decision_endpoint_reports_open_gate(
monkeypatch: pytest.MonkeyPatch, tmp_path: Path,
) -> None:
monkeypatch.setenv("MINDXTRAIN_AUTOSTART", "1")
_write_decision(tmp_path, monkeypatch, recommend=True, reason="dream corpus +340")
r = client.get("/coach/api/sea-decision")
body = r.json()
assert body["open"] is True
assert body["autostart_enabled"] is True
assert "dream corpus +340" in body["reason"]
assert body["decision"]["decided_by"] == "strategic_evolution_agent"
|