Files
auto_control/core/stf_client.py
T
butubb 3a38d9a57f fix: adb 命令超时 + STF 释放失败检查,杜绝设备卡死/占用悬空
- adb 所有命令加 30s 超时:connect 到不可达地址不再无限挂起(worker 不再卡 connecting)
- STF release 检查响应状态并重试:504 等失败不再被静默吞掉,释放失败会真实上报
- release_all_mine 返回 (released, failed);web /api/release 透出失败列表
2026-08-08 16:30:42 +08:00

229 lines
9.5 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""OpenSTF REST API 客户端封装。
负责:设备列表查询、占用、远程 ADB 隧道建立、释放。
不负责实际 UI 操作(那是 DeviceWorker 的事)。
健壮性设计:
- 所有请求带 timeout,避免 STF 网关 504 时长时间卡死
- remoteConnect 带重试,STF provider 偶尔慢启动
- 友好错误分类:设备离线 / 网络不通 / STF 内部错误,便于上层决策
- 占用冲突自动重试(STF 偶发 400 "device already in use")
"""
import time
import requests
from config import STF_URL, STF_TOKEN
from core.logger import get_logger
_log = get_logger("core.stf")
# 请求超时(秒)。connect 超时 + read 超时
_TIMEOUT = (5, 15)
# remoteConnect 重试次数(STF provider 慢启动时需要重试)
_RC_RETRIES = 3
# remoteConnect 重试间隔
_RC_DELAY = 2
class STFError(Exception):
"""STF 相关错误基类。"""
def __init__(self, message, code=""):
super().__init__(message)
self.code = code # 错误码:offline / network / conflict / server / unknown
class DeviceOfflineError(STFError):
"""设备离线/不可用。"""
class STFNetworkError(STFError):
"""STF 服务不可达。"""
class DeviceConflictError(STFError):
"""设备已被占用。"""
def _headers():
return {"Authorization": f"Bearer {STF_TOKEN}"}
class STFClient:
"""OpenSTF REST API 封装。"""
def __init__(self, base_url=STF_URL, token=STF_TOKEN):
self.base_url = base_url
self.headers = _headers()
# ================== 设备查询 ==================
def list_all_devices(self):
"""返回 STF 上所有设备(含状态)。"""
try:
resp = requests.get(f"{self.base_url}/api/v1/devices",
headers=self.headers, timeout=_TIMEOUT)
resp.raise_for_status()
return resp.json().get("devices", [])
except requests.exceptions.ConnectionError as e:
raise STFNetworkError(f"STF 服务不可达: {e}", "network") from e
except requests.exceptions.Timeout as e:
raise STFNetworkError(f"STF 请求超时: {e}", "network") from e
except requests.exceptions.RequestException as e:
_log.error(f"获取设备列表失败: {e}")
raise STFError(f"获取设备列表失败: {e}", "server") from e
def list_free_devices(self):
"""返回可占用的空闲设备。"""
return [d for d in self.list_all_devices()
if d.get("present") and d.get("ready")
and not d.get("using") and d.get("owner") is None]
def list_my_devices(self):
"""返回当前账户已占用的设备。"""
try:
resp = requests.get(f"{self.base_url}/api/v1/user/devices",
headers=self.headers, timeout=_TIMEOUT)
resp.raise_for_status()
return resp.json().get("devices", [])
except requests.exceptions.RequestException as e:
_log.error(f"获取已占用设备失败: {e}")
return []
# ================== 占用/释放 ==================
def occupy(self, serial, retries=2):
"""占用设备。返回 (ok, msg)。
冲突(已被占用)时自动重试,偶发 400 "already in use" 可能是脏状态。
"""
for attempt in range(1, retries + 1):
try:
r = requests.post(f"{self.base_url}/api/v1/user/devices/{serial}",
headers=self.headers, timeout=_TIMEOUT)
if r.status_code in (200, 201):
return True, r.text[:200]
if r.status_code == 400 and "already" in r.text.lower():
# 已被占用(可能是自己之前占用没释放干净),尝试先释放再占
_log.warning(f"{serial} 占用冲突,尝试清理后重试 ({attempt}/{retries})")
try:
requests.delete(f"{self.base_url}/api/v1/user/devices/{serial}",
headers=self.headers, timeout=_TIMEOUT)
except Exception:
pass
time.sleep(1)
continue
return False, f"HTTP {r.status_code}: {r.text[:200]}"
except requests.exceptions.RequestException as e:
_log.warning(f"{serial} 占用请求异常 ({attempt}/{retries}): {e}")
if attempt < retries:
time.sleep(1)
return False, str(e)
return False, "占用冲突重试耗尽"
def release(self, serial, retries=3):
"""释放设备占用。返回 (ok, msg)。
STF 释放可能因设备响应超时返回 504(如设备过载/卡死),此时设备实际未释放。
必须检查响应状态并重试,不能静默吞掉——否则占用会一直悬着。
"""
self.remote_disconnect(serial)
last_status = None
last_body = ""
for attempt in range(1, retries + 1):
try:
r = requests.delete(f"{self.base_url}/api/v1/user/devices/{serial}",
headers=self.headers, timeout=_TIMEOUT)
if r.status_code in (200, 201, 202, 204):
return True, "released"
last_status = r.status_code
last_body = r.text[:150]
_log.warning(f"释放 {serial} 失败 HTTP {r.status_code}: {last_body} ({attempt}/{retries})")
except requests.exceptions.RequestException as e:
_log.warning(f"释放 {serial} 请求异常 ({attempt}/{retries}): {e}")
last_status = -1
last_body = str(e)
if attempt < retries:
time.sleep(2)
return False, f"STF 释放失败(HTTP {last_status}): {last_body}"
def release_all_mine(self):
"""释放当前账户占用的所有设备(清理用)。返回 (released, failed)。"""
released, failed = [], []
for d in self.list_my_devices():
serial = d["serial"]
ok, msg = self.release(serial)
if ok:
released.append(serial)
_log.info(f"已释放 {serial}")
else:
failed.append(serial)
_log.error(f"释放 {serial} 失败: {msg}")
return released, failed
# ================== 远程 ADB 隧道 ==================
def remote_connect(self, serial):
"""建立远程 ADB 隧道,返回 remoteConnectUrl。
STF provider 慢启动或设备掉线时会 504,这里带重试。
设备真离线时快速失败,不长时间卡住 worker。
"""
last_err = None
for attempt in range(1, _RC_RETRIES + 1):
try:
conn = requests.post(
f"{self.base_url}/api/v1/user/devices/{serial}/remoteConnect",
headers=self.headers, timeout=_TIMEOUT,
)
# 504 = STF 网关等 provider 响应超时,通常是设备掉线或 provider 卡死
if conn.status_code == 504:
last_err = f"STF 网关超时(504),设备可能掉线"
_log.warning(f"{serial} remoteConnect 504 ({attempt}/{_RC_RETRIES})")
if attempt < _RC_RETRIES:
time.sleep(_RC_DELAY)
continue
raise DeviceOfflineError(
f"{serial} remoteConnect 超时,设备可能掉线或 provider 卡死",
"offline",
)
conn.raise_for_status()
data = conn.json()
if not data.get("success"):
raise STFError(f"远程连接失败: {data.get('description')}", "server")
url = (data.get("remoteConnectUrl") or data.get("remoteAdbUrl")
or data.get("remote_adb_url") or data.get("adbUrl")
or data.get("url"))
if not url:
raise STFError("STF 返回成功但无 remoteConnectUrl", "server")
return url
except DeviceOfflineError:
raise
except requests.exceptions.ConnectionError as e:
last_err = str(e)
_log.warning(f"{serial} remoteConnect 网络异常 ({attempt}/{_RC_RETRIES}): {e}")
if attempt < _RC_RETRIES:
time.sleep(_RC_DELAY)
except requests.exceptions.Timeout as e:
last_err = str(e)
_log.warning(f"{serial} remoteConnect 超时 ({attempt}/{_RC_RETRIES}): {e}")
if attempt < _RC_RETRIES:
time.sleep(_RC_DELAY)
except requests.exceptions.RequestException as e:
last_err = str(e)
_log.error(f"{serial} remoteConnect 请求异常: {e}")
raise STFError(f"远程连接请求失败: {e}", "server") from e
# 重试耗尽
raise DeviceOfflineError(
f"{serial} remoteConnect 重试 {_RC_RETRIES} 次失败: {last_err}",
"offline",
)
def remote_disconnect(self, serial):
"""断开远程 ADB 隧道。"""
try:
requests.post(
f"{self.base_url}/api/v1/user/devices/{serial}/remoteDisconnect",
headers=self.headers, timeout=_TIMEOUT,
)
except requests.exceptions.RequestException:
pass # 断开失败不影响主流程