host
5b2bdc9bbf
Docker goal_resume_breakpoint_task 是唯一漏传 env=self._docker_cli_env(extra_env) 的 goal 方法：密钥只按名 进 docker run，值靠 CLI 环境带；漏传导致容器内 key 为空， 断点后每轮瞬间报 Missing SUB2API_API_KEY、无限空转 （2026-10-29 长任务实弹：274 轮全失败）。 单测：断言断点方法调用 runner 时 CLI 环境含密钥值， 命令里密钥只按名出现不带值。
652 lines
27 KiB
Python
652 lines
27 KiB
Python
"""P1 执行面针对性单测（只测本路改动，不跑全量三套）。
|
||
|
||
运行（在 ~/workspace/codex-app-server-p1security 下）：
|
||
python3 -m unittest discover -s cloud/tests -p "test_p1_security_worker.py" -v
|
||
|
||
说明：只用注入的假 runner 与测试假值，不跑真实模型回合、不联网、
|
||
不登录任何服务；env 里的值只是占位用的假名，绝不是真实密钥。
|
||
本文件只归本路（执行面）所有，不碰 test_p1_security_server.py。
|
||
"""
|
||
|
||
import json
|
||
import os
|
||
import sqlite3
|
||
import sys
|
||
import tempfile
|
||
import threading
|
||
import time
|
||
import unittest
|
||
from pathlib import Path
|
||
from types import SimpleNamespace
|
||
from unittest import mock
|
||
|
||
try:
|
||
import cloud.sandbox as sandbox_mod
|
||
from cloud.cost import CostStore
|
||
from cloud.pr_exec import enqueue_pr_job
|
||
from cloud.sandbox import DockerExecutor, LocalExecutor
|
||
from cloud.server import init_db
|
||
from cloud.worker import Worker, _code_signature
|
||
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
|
||
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
|
||
import cloud.sandbox as sandbox_mod
|
||
from cloud.cost import CostStore
|
||
from cloud.pr_exec import enqueue_pr_job
|
||
from cloud.sandbox import DockerExecutor, LocalExecutor
|
||
from cloud.server import init_db
|
||
from cloud.worker import Worker, _code_signature
|
||
|
||
FAKE_ENV = {"XCM_API_KEY": "test-placeholder"}
|
||
|
||
|
||
def _completed_runner(cmd):
|
||
if "dispatch" in cmd:
|
||
name = cmd[cmd.index("--name") + 1]
|
||
else:
|
||
name = cmd[cmd.index("--task") + 1]
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout=json.dumps({"task": name, "status": "completed",
|
||
"text": "done"}),
|
||
stderr="")
|
||
|
||
|
||
def _usage_runner(usage, calls):
|
||
def fake(cmd):
|
||
calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout=json.dumps({"status": "completed", "text": "done",
|
||
"usage": usage}),
|
||
stderr="")
|
||
return fake
|
||
|
||
|
||
def _insert_task(db_path, task_id, created_by="tester", status="queued",
|
||
project_id="p1", **cols):
|
||
now = time.time()
|
||
conn = sqlite3.connect(db_path)
|
||
try:
|
||
conn.execute(
|
||
"INSERT INTO tasks(id, project_id, prompt, status,"
|
||
" idempotency_key, depends_on, priority, created_by,"
|
||
" created_at, updated_at, last_heartbeat)"
|
||
" VALUES(?, ?, ?, ?, NULL, '[]', 0, ?, ?, ?, NULL)",
|
||
(task_id, project_id, "do work", status, created_by, now, now),
|
||
)
|
||
for key, value in cols.items():
|
||
conn.execute(
|
||
"UPDATE tasks SET %s = ? WHERE id = ?" % key,
|
||
(value, task_id))
|
||
conn.commit()
|
||
finally:
|
||
conn.close()
|
||
|
||
|
||
def _insert_quota(db_path, user_id, micros):
|
||
conn = sqlite3.connect(db_path)
|
||
try:
|
||
conn.execute(
|
||
"CREATE TABLE IF NOT EXISTS user_quotas("
|
||
"user_id TEXT PRIMARY KEY, quota_usd_micros BIGINT,"
|
||
" updated_at DOUBLE PRECISION)")
|
||
conn.execute(
|
||
"INSERT INTO user_quotas(user_id, quota_usd_micros, updated_at)"
|
||
" VALUES(?, ?, ?)",
|
||
(user_id, micros, time.time()))
|
||
conn.commit()
|
||
finally:
|
||
conn.close()
|
||
|
||
|
||
def _insert_cost(db_path, task_id, model, input_tokens, output_tokens,
|
||
micros, source="actual"):
|
||
_insert_task(db_path, task_id, status="completed")
|
||
store = CostStore(db_path)
|
||
try:
|
||
store.record(task_id, "p1", model, input_tokens=input_tokens,
|
||
output_tokens=output_tokens, usd_micros=micros,
|
||
source=source)
|
||
finally:
|
||
try:
|
||
store._conn.close()
|
||
except Exception:
|
||
pass
|
||
|
||
|
||
def _status(db_path, task_id):
|
||
conn = sqlite3.connect(db_path)
|
||
try:
|
||
row = conn.execute(
|
||
"SELECT status FROM tasks WHERE id = ?", (task_id,)).fetchone()
|
||
finally:
|
||
conn.close()
|
||
return row[0] if row else None
|
||
|
||
|
||
def _error_text(db_path, task_id):
|
||
conn = sqlite3.connect(db_path)
|
||
try:
|
||
row = conn.execute(
|
||
"SELECT error_text FROM tasks WHERE id = ?",
|
||
(task_id,)).fetchone()
|
||
finally:
|
||
conn.close()
|
||
return row[0] if row else None
|
||
|
||
|
||
class P1WorkerTestCase(unittest.TestCase):
|
||
def setUp(self):
|
||
self.tmp = tempfile.TemporaryDirectory()
|
||
self.db_path = os.path.join(self.tmp.name, "control.db")
|
||
init_db(self.db_path)
|
||
self.calls = []
|
||
|
||
def tearDown(self):
|
||
self.tmp.cleanup()
|
||
|
||
def _worker(self, runner=None, **kwargs):
|
||
kwargs.setdefault("env", FAKE_ENV)
|
||
return Worker(self.db_path, "/fake/host.py",
|
||
runner=runner or _completed_runner, **kwargs)
|
||
|
||
# 1. 无配额行按默认 $10 额度（不再是不限额）
|
||
def test_missing_quota_row_uses_default_quota(self):
|
||
_insert_cost(self.db_path, "donor-a", "gpt-6.1-sol",
|
||
1, 0, 11000000)
|
||
_insert_cost(self.db_path, "donor-b", "gpt-6.1-sol",
|
||
1, 0, 9000000)
|
||
worker = self._worker()
|
||
over = worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1"}, None)
|
||
# donor-a 归 tester：已花 $11 > 默认 $10，挡住。
|
||
self.assertIsNotNone(over)
|
||
self.assertIn("quota_exceeded", over)
|
||
# donor-b 同样归 tester 会把总额推得更高，单独测另一用户：
|
||
_insert_task(self.db_path, "donor-c", created_by="rich",
|
||
status="completed")
|
||
store = CostStore(self.db_path)
|
||
try:
|
||
store.record("donor-c", "p1", "gpt-6.1-sol", input_tokens=1,
|
||
output_tokens=0, usd_micros=9000000,
|
||
source="actual")
|
||
finally:
|
||
try:
|
||
store._conn.close()
|
||
except Exception:
|
||
pass
|
||
under = worker._quota_exceeded(
|
||
{"created_by": "rich", "project_id": "p1"}, None)
|
||
self.assertIsNone(under)
|
||
|
||
# 2. 显式 NULL 配额行仍是不限额
|
||
def test_explicit_null_quota_row_stays_unlimited(self):
|
||
_insert_quota(self.db_path, "tester", None)
|
||
_insert_cost(self.db_path, "donor", "gpt-6.1-sol",
|
||
1, 0, 999000000)
|
||
worker = self._worker()
|
||
self.assertIsNone(worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1"}, None))
|
||
|
||
# 3. 属主为空拒绝新回合（不再返回 None 放行）
|
||
def test_ownerless_task_refused(self):
|
||
worker = self._worker()
|
||
for owner in (None, ""):
|
||
err = worker._quota_exceeded(
|
||
{"created_by": owner, "project_id": "p1"}, None)
|
||
self.assertIsNotNone(err)
|
||
self.assertIn("quota_exceeded", err)
|
||
|
||
# 4. 协调员子任务属主跟父任务走
|
||
def test_child_inherits_parent_owner(self):
|
||
_insert_quota(self.db_path, "tester", 100)
|
||
_insert_cost(self.db_path, "donor", "gpt-6.1-sol",
|
||
1, 0, 200)
|
||
_insert_task(self.db_path, "parent", created_by="tester")
|
||
_insert_task(self.db_path, "parent2", created_by="fresh")
|
||
worker = self._worker()
|
||
blocked = worker._quota_exceeded(
|
||
{"created_by": None, "parent_task_id": "parent",
|
||
"project_id": "p1"}, None)
|
||
self.assertIsNotNone(blocked)
|
||
self.assertIn("quota_exceeded", blocked)
|
||
# fresh 用户无配额行、未花钱：默认额度内放行。
|
||
self.assertIsNone(worker._quota_exceeded(
|
||
{"created_by": None, "parent_task_id": "parent2",
|
||
"project_id": "p1"}, None))
|
||
|
||
# 5. 查库失败 fail-closed（新回合拒绝，不自动放行）
|
||
def test_quota_db_failure_is_fail_closed(self):
|
||
worker = self._worker()
|
||
with mock.patch("cloud.worker._db_connect",
|
||
side_effect=RuntimeError("test db down")):
|
||
err = worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1"}, None)
|
||
self.assertIsNotNone(err)
|
||
self.assertTrue(err.startswith("quota_check_failed:"))
|
||
|
||
# 6. Docker 配额看门狗：命中杀容器、未命中不杀
|
||
def test_docker_quota_watchdog(self):
|
||
killed = []
|
||
|
||
def fake(cmd, timeout=None):
|
||
killed.append(list(cmd))
|
||
return SimpleNamespace(returncode=0, stdout="", stderr="")
|
||
|
||
executor = DockerExecutor(runner=fake)
|
||
worker = Worker(self.db_path, "/fake/host.py",
|
||
env=FAKE_ENV, executor=executor)
|
||
# 未命中：0.2 秒后停表，kill 不应被调。
|
||
stop = threading.Event()
|
||
threading.Timer(0.2, stop.set).start()
|
||
worker._docker_quota_watchdog(
|
||
"T1", "codex-test", stop, lambda: False, interval=0.05)
|
||
self.assertEqual(killed, [])
|
||
# 命中：立即杀容器并记命中标记。
|
||
stop2 = threading.Event()
|
||
worker._docker_quota_watchdog(
|
||
"T2", "codex-test", stop2, lambda: True, interval=0.05)
|
||
self.assertTrue(any("kill" in c for c in killed))
|
||
self.assertTrue(worker._take_quota_hit("T2"))
|
||
# 命中已消费，第二次取不到（不污染下一轮）。
|
||
self.assertFalse(worker._take_quota_hit("T2"))
|
||
|
||
# 7. usage_cost_usd 按实际成本累加，max_cost_usd 不再是死参数
|
||
def test_usage_cost_accumulates_and_budget_trips(self):
|
||
_insert_task(self.db_path, "T1")
|
||
worker = self._worker(
|
||
runner=_usage_runner(
|
||
{"input_tokens": 10000, "cached_input_tokens": 0,
|
||
"output_tokens": 1000, "total_tokens": 11000},
|
||
self.calls),
|
||
model="gpt-6.1-sol", max_cost_usd=0.01)
|
||
self.assertFalse(worker._budget_exceeded())
|
||
result = worker.run_one()
|
||
self.assertEqual(result["status"], "completed")
|
||
# 10000*$2 + 1000*$10 = $0.03，超过 $0.01 上限。
|
||
self.assertAlmostEqual(worker.usage_cost_usd, 0.03, places=6)
|
||
self.assertTrue(worker._budget_exceeded())
|
||
conn = sqlite3.connect(self.db_path)
|
||
try:
|
||
row = conn.execute(
|
||
"SELECT usd_micros, source FROM cost_records"
|
||
" WHERE task_id = 'T1'").fetchone()
|
||
finally:
|
||
conn.close()
|
||
self.assertEqual(row, (30000, "actual"))
|
||
|
||
# 8. 取消分支先记成本再返回（与配额中断收尾同款记账点）
|
||
def test_cancel_branch_records_cost(self):
|
||
worker = self._worker()
|
||
task = {"id": "T1", "project_id": "p1", "prompt": "do work"}
|
||
with worker._db_lock:
|
||
cost_id = worker._record_usage_cost(
|
||
task, "gpt-6.1-sol",
|
||
{"input_tokens": 10000, "cached_input_tokens": 0,
|
||
"output_tokens": 1000, "total_tokens": 11000},
|
||
"done", 0)
|
||
self.assertIsNotNone(cost_id)
|
||
self.assertAlmostEqual(worker.usage_cost_usd, 0.03, places=6)
|
||
# 无用量时回落估算，不累加 usage_cost_usd。
|
||
with worker._db_lock:
|
||
worker._record_usage_cost(
|
||
{"id": "T2", "project_id": "p1", "prompt": "do work"},
|
||
"gpt-6.1-sol", None, "done", 0)
|
||
self.assertAlmostEqual(worker.usage_cost_usd, 0.03, places=6)
|
||
|
||
# 9. 估算行不计入无价 token 上限
|
||
def test_estimated_rows_excluded_from_unpriced_cap(self):
|
||
_insert_cost(self.db_path, "donor-est", "gpt-5.6-terra",
|
||
100000, 0, None, source="estimated")
|
||
worker = self._worker()
|
||
old = os.environ.get("CLOUD_QUOTA_UNPRICED_TOKEN_CAP")
|
||
os.environ["CLOUD_QUOTA_UNPRICED_TOKEN_CAP"] = "100"
|
||
try:
|
||
# 10 万估算 token 在上限 100 下仍放行（估算不是真花费）。
|
||
self.assertIsNone(worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1"}, None))
|
||
_insert_cost(self.db_path, "donor-act", "gpt-5.6-terra",
|
||
75, 75, None, source="actual")
|
||
err = worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1"}, None)
|
||
self.assertIsNotNone(err)
|
||
self.assertIn("无价模型 token", err)
|
||
finally:
|
||
if old is None:
|
||
os.environ.pop("CLOUD_QUOTA_UNPRICED_TOKEN_CAP", None)
|
||
else:
|
||
os.environ["CLOUD_QUOTA_UNPRICED_TOKEN_CAP"] = old
|
||
|
||
# 10. cache_write 不计入无价上限（只 input+output 计）
|
||
def test_cache_write_excluded_from_unpriced_cap(self):
|
||
worker = self._worker()
|
||
old = os.environ.get("CLOUD_QUOTA_UNPRICED_TOKEN_CAP")
|
||
os.environ["CLOUD_QUOTA_UNPRICED_TOKEN_CAP"] = "100"
|
||
try:
|
||
usage = {"input_tokens": 10, "output_tokens": 5,
|
||
"cache_write_input_tokens": 9999999,
|
||
"total_tokens": 15}
|
||
self.assertIsNone(worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1",
|
||
"model": "gpt-5.6-terra"}, usage))
|
||
usage2 = {"input_tokens": 200, "output_tokens": 0,
|
||
"cache_write_input_tokens": 0,
|
||
"total_tokens": 200}
|
||
err = worker._quota_exceeded(
|
||
{"created_by": "tester", "project_id": "p1",
|
||
"model": "gpt-5.6-terra"}, usage2)
|
||
self.assertIsNotNone(err)
|
||
finally:
|
||
if old is None:
|
||
os.environ.pop("CLOUD_QUOTA_UNPRICED_TOKEN_CAP", None)
|
||
else:
|
||
os.environ["CLOUD_QUOTA_UNPRICED_TOKEN_CAP"] = old
|
||
|
||
# 11. 协调员子任务继承父任务模型/强度，分工显式值优先
|
||
def test_coordinator_child_inherits_model_effort(self):
|
||
worker = self._worker()
|
||
task = {"id": "coord", "project_id": "p1", "prompt": "coord",
|
||
"priority": 0, "created_by": "tester", "model": "gpt-6.1-sol",
|
||
"effort": "high", "executor": None, "environment_id": None}
|
||
children = worker._coord_sync_children(task, [
|
||
{"name": "a", "prompt": "do a", "model": None,
|
||
"effort": None, "executor": None},
|
||
{"name": "b", "prompt": "do b", "model": "gpt-5.6-terra",
|
||
"effort": "low", "executor": None},
|
||
])
|
||
self.assertEqual(len(children), 2)
|
||
conn = sqlite3.connect(self.db_path)
|
||
try:
|
||
rows = {r[0]: (r[1], r[2]) for r in conn.execute(
|
||
"SELECT coordinator_part, model, effort FROM tasks"
|
||
" WHERE parent_task_id = 'coord'").fetchall()}
|
||
finally:
|
||
conn.close()
|
||
self.assertEqual(rows["a"], ("gpt-6.1-sol", "high"))
|
||
self.assertEqual(rows["b"], ("gpt-5.6-terra", "low"))
|
||
|
||
# 12. 断点续跑：Local/Docker/popen 三路都不传 --goal-text
|
||
def test_goal_resume_breakpoint_omits_goal_text(self):
|
||
calls = []
|
||
|
||
def fake(cmd):
|
||
calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
task = {"id": "T1", "project_id": "p1", "prompt": "orig goal",
|
||
"verify_command": "true", "goal_max_rounds": 3}
|
||
local = LocalExecutor("/fake/host.py", runner=fake)
|
||
local.goal_resume_task(task, None)
|
||
self.assertIn("goal-resume", calls[-1])
|
||
self.assertNotIn("--goal-text", calls[-1])
|
||
docker_calls = []
|
||
|
||
def dfake(cmd, timeout=None):
|
||
docker_calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
docker = DockerExecutor(runner=dfake, env=FAKE_ENV)
|
||
with mock.patch("cloud.sandbox.shutil.which",
|
||
return_value="/usr/bin/docker"):
|
||
docker.goal_resume_task(task, None)
|
||
self.assertIn("goal-resume", docker_calls[-1])
|
||
self.assertNotIn("--goal-text", docker_calls[-1])
|
||
with mock.patch.object(sandbox_mod.subprocess, "Popen") as popen:
|
||
popen.return_value = SimpleNamespace(pid=9999)
|
||
local.popen_task(
|
||
{"id": "T1", "prompt": "resume msg",
|
||
"_resume_kind": "breakpoint"})
|
||
# popen_task 默认 kind=dispatch；显式测 goal-resume：
|
||
local.popen_task(
|
||
{"id": "T1", "prompt": "resume msg",
|
||
"_resume_kind": "breakpoint"}, kind="goal-resume")
|
||
cmd = popen.call_args[0][0]
|
||
self.assertIn("goal-resume", cmd)
|
||
self.assertNotIn("--goal-text", cmd)
|
||
|
||
# 12b. 断点续跑透传 model_provider：首轮 thread 在该
|
||
# provider 下创建，续跑命令必须带同一 --model-provider，
|
||
# 否则 thread/resume 因 provider 表里没有它而失败
|
||
# （2026-10-29 probe15 实弹：sub2api not found）。
|
||
def test_goal_resume_breakpoint_forwards_model_provider(self):
|
||
docker_calls = []
|
||
|
||
def dfake(cmd, timeout=None):
|
||
docker_calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
task = {"id": "T1", "prompt": "orig goal"}
|
||
docker = DockerExecutor(runner=dfake, env=FAKE_ENV)
|
||
with mock.patch("cloud.sandbox.shutil.which",
|
||
return_value="/usr/bin/docker"):
|
||
docker.goal_resume_breakpoint_task(
|
||
task, model_provider="sub2api")
|
||
cmd = docker_calls[-1]
|
||
self.assertIn("goal-resume", cmd)
|
||
self.assertNotIn("--goal-text", cmd)
|
||
self.assertIn("--model-provider", cmd)
|
||
self.assertIn("sub2api", cmd)
|
||
# 不传 provider 时不带该 flag（透传语义，不是强制默认值）。
|
||
docker_calls.clear()
|
||
with mock.patch("cloud.sandbox.shutil.which",
|
||
return_value="/usr/bin/docker"):
|
||
docker.goal_resume_breakpoint_task(task)
|
||
self.assertNotIn("--model-provider", docker_calls[-1])
|
||
|
||
calls = []
|
||
|
||
def fake(cmd):
|
||
calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
local = LocalExecutor("/fake/host.py", runner=fake)
|
||
local.goal_resume_breakpoint_task(task, model_provider="sub2api")
|
||
self.assertIn("--model-provider", calls[-1])
|
||
self.assertIn("sub2api", calls[-1])
|
||
|
||
# 12c. worker 本地断点调用点透传网关 provider：run_one
|
||
# 认领 goal_resume_pending 任务后，本地断点命令必须带
|
||
# 网关解析出的 --model-provider（Docker 调用点同理，
|
||
# 沙箱级已覆盖命令形状）。
|
||
def test_worker_local_breakpoint_forwards_gateway_provider(self):
|
||
def fake(cmd):
|
||
self.calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
worker = self._worker(runner=fake)
|
||
worker._ensure_gateway = lambda task: ("sub2api", {})
|
||
_insert_task(self.db_path, "T1", goal_mode=1,
|
||
goal="orig goal", goal_max_rounds=3,
|
||
goal_resume_pending=1,
|
||
model="gpt-5.6-terra")
|
||
worker.run_one()
|
||
self.assertEqual(_status(self.db_path, "T1"), "completed")
|
||
self.assertTrue(any(
|
||
"--model-provider" in cmd and "sub2api" in cmd
|
||
for cmd in self.calls))
|
||
|
||
# 12d. Docker 断点续跑必须把密钥环境透传给 runner：
|
||
# 密钥类变量只按名进 docker run（-e SUB2API_API_KEY 不带值），
|
||
# 值靠 _docker_cli_env 放进 CLI 子进程环境；漏传 env 时容器
|
||
# 内 key 为空，每轮瞬间报 Missing SUB2API_API_KEY（2026-10-29
|
||
# 长任务实弹：断点后 274 轮全失败、无限空转）。
|
||
def test_docker_breakpoint_forwards_secret_env_to_runner(self):
|
||
seen = {}
|
||
|
||
def dfake(cmd, timeout=None, env=None):
|
||
seen["cmd"] = list(cmd)
|
||
seen["env"] = env
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
task = {"id": "T1", "prompt": "orig goal"}
|
||
docker = DockerExecutor(runner=dfake, env=FAKE_ENV)
|
||
with mock.patch("cloud.sandbox.shutil.which",
|
||
return_value="/usr/bin/docker"):
|
||
docker.goal_resume_breakpoint_task(
|
||
task, model_provider="sub2api",
|
||
extra_env={"SUB2API_API_KEY": "secret-placeholder",
|
||
"SUB2API_BASE_URL": "https://example.test"})
|
||
# 命令里密钥只按名出现，不带值（不进 argv）。
|
||
self.assertIn("SUB2API_API_KEY", seen["cmd"])
|
||
self.assertNotIn("SUB2API_API_KEY=secret-placeholder",
|
||
seen["cmd"])
|
||
# runner 的 CLI 环境里必须有值，否则容器内 key 为空。
|
||
self.assertIsNotNone(seen["env"])
|
||
self.assertEqual(seen["env"].get("SUB2API_API_KEY"),
|
||
"secret-placeholder")
|
||
|
||
# 13. 开新目标：goal-resume 带 --goal-text（旧行为保留）
|
||
def test_goal_resume_new_goal_keeps_goal_text(self):
|
||
calls = []
|
||
|
||
def fake(cmd):
|
||
calls.append(list(cmd))
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="goal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
|
||
task = {"id": "T1", "project_id": "p1", "prompt": "new goal"}
|
||
local = LocalExecutor("/fake/host.py", runner=fake)
|
||
local.goal_resume_task(task, "new goal")
|
||
self.assertIn("--goal-text", calls[-1])
|
||
self.assertIn("new goal", calls[-1])
|
||
|
||
# 14. goal 轮次上限：sandbox loud fail + worker 认领后 failed
|
||
def test_goal_max_rounds_cap(self):
|
||
task = {"id": "T1", "project_id": "p1", "prompt": "g",
|
||
"verify_command": "true", "goal_max_rounds": 101}
|
||
local = LocalExecutor("/fake/host.py", runner=_completed_runner)
|
||
with self.assertRaises(ValueError):
|
||
local.goal_task(task)
|
||
_insert_task(self.db_path, "T1", verify_command="true",
|
||
goal_max_rounds=101)
|
||
worker = self._worker()
|
||
result = worker.run_one()
|
||
self.assertEqual(result["status"], "failed")
|
||
self.assertIn("上限", result["error"])
|
||
self.assertEqual(_status(self.db_path, "T1"), "failed")
|
||
self.assertEqual(self.calls, [])
|
||
|
||
# 15. PR 非 local：显式 needs_local_worker，行状态持久化，
|
||
# 任务本身不动，标记过的行仍可被 local worker 认领
|
||
def test_pr_non_local_needs_local_worker(self):
|
||
conn = sqlite3.connect(self.db_path)
|
||
try:
|
||
job_id = enqueue_pr_job(conn, "T1", "p1", "test-host")
|
||
finally:
|
||
conn.close()
|
||
executor = DockerExecutor(runner=lambda cmd, timeout=None:
|
||
SimpleNamespace(returncode=0, stdout="",
|
||
stderr=""))
|
||
worker = Worker(self.db_path, "/fake/host.py", env=FAKE_ENV,
|
||
executor=executor, host_name="test-host")
|
||
results = worker._run_pr_jobs()
|
||
self.assertEqual(len(results), 1)
|
||
self.assertEqual(results[0]["status"], "needs_local_worker")
|
||
self.assertEqual(results[0]["pr_job_id"], job_id)
|
||
conn = sqlite3.connect(self.db_path)
|
||
try:
|
||
row = conn.execute(
|
||
"SELECT status FROM pr_jobs WHERE id = ?",
|
||
(job_id,)).fetchone()
|
||
self.assertEqual(row[0], "needs_local_worker")
|
||
# 标记过的行仍可被 local worker 认领，不会被搁置。
|
||
from cloud.pr_exec import claim_pr_job
|
||
claimed = claim_pr_job(conn, "test-host")
|
||
self.assertIsNotNone(claimed)
|
||
self.assertEqual(claimed["id"], job_id)
|
||
finally:
|
||
conn.close()
|
||
|
||
# 16. 多文件签名 + 签名缺失/已变时认领 fail-closed
|
||
def test_code_signature_multi_file_and_claim_refusal(self):
|
||
sig = _code_signature("/fake/host.py")
|
||
self.assertIsNotNone(sig)
|
||
self.assertEqual(len(sig), 6)
|
||
for entry in sig:
|
||
self.assertEqual(len(entry), 2)
|
||
_insert_task(self.db_path, "T1")
|
||
worker = self._worker()
|
||
worker._code_signature = ("stale-signature",)
|
||
result = worker.run_one()
|
||
self.assertFalse(result["dispatched"])
|
||
self.assertEqual(result["reason"], "already_claimed")
|
||
self.assertEqual(_status(self.db_path, "T1"), "queued")
|
||
worker._code_signature = None
|
||
result = worker.run_one()
|
||
self.assertFalse(result["dispatched"])
|
||
self.assertEqual(_status(self.db_path, "T1"), "queued")
|
||
|
||
# 17. goal 首轮撞已存在 → 断点续跑一次，不直接 failed
|
||
def test_goal_already_exists_falls_back_to_breakpoint(self):
|
||
calls = []
|
||
|
||
def fake(cmd):
|
||
calls.append(list(cmd))
|
||
if "goal-resume" in cmd:
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="round 1: turn=completed verify_rc=0 "
|
||
"verify_out=\ngoal met after rounds\n"
|
||
"usage input=10 cached=0 cache_write=0 output=5 "
|
||
"reasoning=0 total=15 completions=1\n",
|
||
stderr="")
|
||
if "goal" in cmd:
|
||
return SimpleNamespace(
|
||
returncode=0,
|
||
stdout="host 任务已存在：T1 已存在",
|
||
stderr="")
|
||
return _completed_runner(cmd)
|
||
|
||
_insert_task(self.db_path, "T1", verify_command="true",
|
||
goal_max_rounds=2)
|
||
worker = self._worker(runner=fake)
|
||
result = worker.run_one()
|
||
self.assertEqual(result["status"], "completed")
|
||
# 注意：cmd 是元素列表，"goal" in cmd 只匹配恰好等于
|
||
# "goal" 的元素，goal-resume 命令里并没有 "goal" 元素，
|
||
# 所以按“任一元素含 goal 子串”过滤。
|
||
goal_cmds = [c for c in calls
|
||
if any("goal" in e for e in c)]
|
||
self.assertTrue(any("goal-resume" in c for c in goal_cmds))
|
||
resume_cmd = [c for c in goal_cmds if "goal-resume" in c][0]
|
||
self.assertNotIn("--goal-text", resume_cmd)
|
||
self.assertEqual(_status(self.db_path, "T1"), "completed")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
unittest.main()
|