diff --git a/chv-guest/generate_lockfile.py b/chv-guest/generate_lockfile.py new file mode 100644 index 0000000..2786e13 --- /dev/null +++ b/chv-guest/generate_lockfile.py @@ -0,0 +1,160 @@ +#!/usr/bin/env python3 +"""在构建宿主上生成 CHV 客机镜像的供应链 lockfile（to10 轨）。 + +事后校验见 check_lockfile.py（校验 BUILDINFO.versions）； +本脚本是事前/事中生成端：在构建宿主上跑，读 BUILDINFO +的 tag 与 contents（仅名字清单，不含版本号），逐项从本 +机 dpkg / pip 查真实版本，写出 lockfile.json： + + {"tag": ..., "image_sha256": ... | null, + "generated_at": UTC ISO 时间, + "versions": {"dpkg": ["name version", ...], + "pip": ["name version", ...]}, + "unresolved": ["查不到版本的名", ...]} + +绝不编造版本号：查不到的进 unresolved，不猜、不填占位 +版本。--verify 模式校验一份 lockfile：tag 与 BUILDINFO +一致、dpkg/pip 非空、每条都有非空版本、unresolved 为空； +任一不满足即退出非 0（unresolved 非空 = 清单不完整， +不是“生成成功”）。 + +用法（均在构建宿主上跑）： + python3 generate_lockfile.py BUILDINFO.json --output lockfile.json + python3 generate_lockfile.py --verify lockfile.json \ + --buildinfo BUILDINFO.json +本轨无构建宿主：脚本已落地且可跑（见回执里的 /tmp 合成 +BUILDINFO 实测），真实清单待构建宿主产出，绝不把合成 +数据写成生产清单。 +""" + +import json +import subprocess +import sys +from datetime import datetime, timezone + + +def _dpkg_version(name): + """dpkg 查版本，查不到回 None（不抛）。""" + try: + out = subprocess.run( + ["dpkg-query", "-W", "-f=${Version}", name], + capture_output=True, text=True, timeout=30) + except Exception: + return None + if out.returncode != 0: + return None + version = out.stdout.strip() + return version or None + + +def _pip_version(name): + """pip 查版本，查不到回 None（不抛）。""" + try: + out = subprocess.run( + [sys.executable, "-m", "pip", "show", name], + capture_output=True, text=True, timeout=30) + except Exception: + return None + if out.returncode != 0: + return None + for line in out.stdout.splitlines(): + if line.startswith("Version:"): + version = line.split(":", 1)[1].strip() + return version or None + return None + + +def generate(buildinfo_path, output_path): + with open(buildinfo_path) as fh: + buildinfo = json.load(fh) + names = buildinfo.get("contents") or [] + if not isinstance(names, list): + print("BUILDINFO.contents 不是列表", file=sys.stderr) + return 2 + dpkg, pip, unresolved = [], [], [] + for name in names: + version = _dpkg_version(name) + if version is not None: + dpkg.append("%s %s" % (name, version)) + continue + version = _pip_version(name) + if version is not None: + pip.append("%s %s" % (name, version)) + continue + unresolved.append(name) + lockfile = { + "tag": buildinfo.get("tag"), + "image_sha256": buildinfo.get("image_sha256") + or buildinfo.get("sha256"), + "generated_at": datetime.now(timezone.utc).isoformat(), + "versions": {"dpkg": sorted(dpkg), "pip": sorted(pip)}, + "unresolved": sorted(unresolved), + } + with open(output_path, "w") as fh: + json.dump(lockfile, fh, indent=2, sort_keys=True) + fh.write("\n") + if unresolved: + print("警告：%d 项查不到版本，已列入 unresolved：%s" + % (len(unresolved), ", ".join(sorted(unresolved))), + file=sys.stderr) + return 0 + + +def verify(lockfile_path, buildinfo_path): + with open(lockfile_path) as fh: + lockfile = json.load(fh) + if buildinfo_path: + with open(buildinfo_path) as fh: + buildinfo = json.load(fh) + if lockfile.get("tag") != buildinfo.get("tag"): + print("tag 不一致：lockfile %s != BUILDINFO %s" + % (lockfile.get("tag"), buildinfo.get("tag")), + file=sys.stderr) + return 1 + versions = lockfile.get("versions") or {} + for kind in ("dpkg", "pip"): + entries = versions.get(kind) + if not entries: + print("versions.%s 为空，供应链清单缺失" % kind, + file=sys.stderr) + return 1 + for entry in entries: + parts = str(entry).split() + if len(parts) < 2 or not parts[1]: + print("版本为空的条目：%s %r" % (kind, entry), + file=sys.stderr) + return 1 + unresolved = lockfile.get("unresolved") or [] + if unresolved: + print("清单不完整，unresolved 非空：%s" % sorted(unresolved), + file=sys.stderr) + return 1 + print("lockfile 校验通过：tag=%s dpkg=%d pip=%d" + % (lockfile.get("tag"), len(versions["dpkg"]), + len(versions["pip"]))) + return 0 + + +def main(argv): + if "--verify" in argv: + lockfile_path = argv[argv.index("--verify") + 1] + buildinfo_path = None + if "--buildinfo" in argv: + buildinfo_path = argv[argv.index("--buildinfo") + 1] + return verify(lockfile_path, buildinfo_path) + if len(argv) < 2: + print("用法：generate_lockfile.py BUILDINFO.json " + "--output lockfile.json | --verify lockfile.json " + "[--buildinfo BUILDINFO.json]", file=sys.stderr) + return 2 + output_path = None + if "--output" in argv: + output_path = argv[argv.index("--output") + 1] + if not output_path: + print("缺少 --output", file=sys.stderr) + return 2 + return generate(argv[1], output_path) + + +if __name__ == "__main__": + raise SystemExit(main(sys.argv)) diff --git a/cloud/chv/compat215.py b/cloud/chv/compat215.py index c89c777..5dd76f2 100644 --- a/cloud/chv/compat215.py +++ b/cloud/chv/compat215.py @@ -48,7 +48,6 @@ cloud.quota，测试钉住这些模块在本基线里不存在）： import os - # 合并后替换清单：每项的 local/replacement/test 字段由测试 # 逐项核对（local 函数真实存在、#215 模块当前不存在、 # test 命名的测试类真实存在），清单本身不得漂移。 diff --git a/cloud/chv/config.py b/cloud/chv/config.py index 0265d93..693cfe1 100644 --- a/cloud/chv/config.py +++ b/cloud/chv/config.py @@ -27,6 +27,33 @@ import os import re +def _to_int(value: object) -> int: + """动态配置值转 int（env/参数均未定型）：bool 拒绝， + int/float 按 int()，str 按 strip 后 int()，其它 ValueError。 + 旧行为是 int(value) 抛 TypeError/ValueError 后由调用方 + 转 ValueError；这里直接统一成 ValueError，文案由调用方定。""" + if isinstance(value, bool): + raise ValueError("布尔值不是整数") + if isinstance(value, int): + return value + if isinstance(value, float): + return int(value) + if isinstance(value, str): + return int(value.strip()) + raise ValueError("不是整数：%r" % (value,)) + + +def _to_float(value: object) -> float: + """动态配置值转 float，与 _to_int 同款口径。""" + if isinstance(value, bool): + raise ValueError("布尔值不是数字") + if isinstance(value, (int, float)): + return float(value) + if isinstance(value, str): + return float(value.strip()) + raise ValueError("不是数字：%r" % (value,)) + + def _parse_memory_bytes(value: str) -> int: """内存字符串转字节：2G/512M/1024K/2048（无单位=字节）。 非法值 ValueError，绝不静默按默认算。""" @@ -93,7 +120,7 @@ def _ssh_fragment(params: dict, env: dict) -> dict: else: raw_port = env.get("CHV_SSH_PORT") or 2222 try: - ssh_port = int(raw_port) + ssh_port = _to_int(raw_port) except (TypeError, ValueError): raise ValueError("CHV_SSH_PORT 必须是整数") if not 1 <= ssh_port <= 65535: @@ -134,11 +161,11 @@ def _quota_fragment(env: dict) -> dict: 磁盘/快照数上限不在这里，它们是磁盘行为，见 _disk_fragment。校验文案与旧代码逐字一致。""" try: - max_vms = int(env.get("CHV_MAX_VMS") or 4) + max_vms = _to_int(env.get("CHV_MAX_VMS") or 4) except ValueError: raise ValueError("CHV_MAX_VMS 必须是整数") try: - warm_pool_size = int(env.get("CHV_WARM_POOL_SIZE") or 0) + warm_pool_size = _to_int(env.get("CHV_WARM_POOL_SIZE") or 0) except ValueError: raise ValueError("CHV_WARM_POOL_SIZE 必须是整数") return {"max_vms": max_vms, "warm_pool_size": warm_pool_size} @@ -154,18 +181,18 @@ def _disk_fragment(env: dict) -> dict: raise ValueError( "CHV_DISK_MODE 只能是 copy 或 reuse：%s" % disk_mode) try: - disk_gb = float(env.get("CHV_DISK_GB")) \ + disk_gb = _to_float(env.get("CHV_DISK_GB")) \ if env.get("CHV_DISK_GB") else 50.0 except ValueError: raise ValueError("CHV_DISK_GB 必须是数字") if disk_gb <= 0: raise ValueError("CHV_DISK_GB 必须大于 0") try: - snapshot_max_count = int(env.get("CHV_SNAPSHOT_MAX_COUNT") - or 10) - snapshot_max_total_gb = float( + snapshot_max_count = _to_int(env.get("CHV_SNAPSHOT_MAX_COUNT") + or 10) + snapshot_max_total_gb = _to_float( env.get("CHV_SNAPSHOT_MAX_TOTAL_GB") or 20.0) - snapshot_retention_days = float( + snapshot_retention_days = _to_float( env.get("CHV_SNAPSHOT_RETENTION_DAYS") or 30.0) except ValueError: raise ValueError("CHV_SNAPSHOT_* 必须是数字") @@ -192,18 +219,18 @@ def _timeout_fragment(params: dict, env: dict) -> dict: 超时、客机就绪超时/间隔。ready_interval <= 0 时回落 5 秒（旧语义，非校验失败）。校验文案逐字一致。""" task_timeout_raw = env.get("CHV_TASK_TIMEOUT") - task_timeout = float(task_timeout_raw) \ + task_timeout = _to_float(task_timeout_raw) \ if task_timeout_raw else None try: - api_timeout = float( + api_timeout = _to_float( params.get("api_timeout") if params.get("api_timeout") is not None else env.get("CHV_API_TIMEOUT") or 30) - ready_timeout = float( + ready_timeout = _to_float( params.get("ready_timeout") if params.get("ready_timeout") is not None else env.get("CHV_READY_TIMEOUT") or 120) - ready_interval = float( + ready_interval = _to_float( params.get("ready_interval") if params.get("ready_interval") is not None else env.get("CHV_READY_INTERVAL") or 5) diff --git a/cloud/chv/disk.py b/cloud/chv/disk.py index f0404d2..68addf6 100644 --- a/cloud/chv/disk.py +++ b/cloud/chv/disk.py @@ -203,7 +203,7 @@ def check_snapshot_limits(state_dir: str, vm_name: str, max_count: int, raise ChvFailure( "snapshot_quota_exceeded", "快照数量已达上限 %d（VM %s）" % (max_count, vm_name)) - total = 0 + total: float = 0.0 for entry in kept: size = entry.get("size") if isinstance(size, (int, float)): diff --git a/cloud/chv/executor.py b/cloud/chv/executor.py index d19d6a9..3bdddc4 100644 --- a/cloud/chv/executor.py +++ b/cloud/chv/executor.py @@ -34,30 +34,35 @@ import shutil import subprocess import threading import time +from collections.abc import Callable -from cloud.chv.api_client import _unix_socket_api -from cloud.chv.api_client import normalize_api_result -from cloud.executor_base import Executor -from cloud.chv.config import _parse_memory_bytes -from cloud.chv.config import build_chv_config -from cloud.chv.disk import check_disk_quota -from cloud.chv.disk import check_snapshot_limits -from cloud.chv.disk import delete_task_disk -from cloud.chv.disk import provision_task_disk -from cloud.chv.disk import record_snapshot -from cloud.chv.disk import task_disk_path +from cloud.chv.api_client import _unix_socket_api, normalize_api_result +from cloud.chv.config import _parse_memory_bytes, build_chv_config +from cloud.chv.disk import ( + check_disk_quota, + check_snapshot_limits, + delete_task_disk, + provision_task_disk, + record_snapshot, + task_disk_path, +) from cloud.chv.errors import ChvFailure -from cloud.chv.hardening import ensure_secure_state_dir -from cloud.chv.hardening import parse_net_entries -from cloud.chv.hardening import socket_is_live -from cloud.chv.hardening import validate_egress_proxy -from cloud.chv.hardening import validate_net_entries -from cloud.chv.hardening import validate_snapshot_url +from cloud.chv.hardening import ( + ensure_secure_state_dir, + parse_net_entries, + socket_is_live, + validate_egress_proxy, + validate_net_entries, + validate_snapshot_url, +) from cloud.chv.ssh_transport import build_ssh_command as _build_ssh_cmd from cloud.chv.ssh_transport import known_hosts_path as _known_hosts_path -from cloud.chv.ssh_transport import parse_keyscan_output -from cloud.chv.ssh_transport import split_secret_env -from cloud.chv.ssh_transport import write_known_hosts +from cloud.chv.ssh_transport import ( + parse_keyscan_output, + split_secret_env, + write_known_hosts, +) +from cloud.executor_base import Executor _LOGGER = logging.getLogger("cloud.cloud_hypervisor") @@ -246,6 +251,57 @@ class CloudHypervisorExecutor(Executor): 开启嵌套虚拟化的实例，那是用户侧开通决策，不在本轨内。 """ + # to10：mypy 属性声明（仅注解、不赋值，运行时仍由 + # _copy_config_fields/_init_runtime 抄写）。没有这组声明， + # 动态 setattr 会让 mypy 认为全部配置属性不存在（基线 + # 115 个报错里近 90 个属此类），类型检查等于没跑。 + binary: str | None + kernel: str | None + rootfs: str | None + cpus: str + memory: str + state_dir: str + ssh_key: str | None + ssh_user: str + ssh_port: int + guest_host_py: str + cmdline: str + net: str | None + net_json: str | None + kvm_path: str + cpuinfo_path: str + ch_remote: str + model: str | None + max_vms: int + disk_mode: str + disk_gb: float + egress_proxy: str | None + allow_direct_egress: bool + task_timeout: float | None + snapshot_url: str | None + warm_pool_size: int + api_timeout: float + ready_timeout: float + ready_interval: float + host_key_mode: str + ssh_host: str | None + ssh_forward: bool + reuse_experiment: bool + single_tenant_project: str | None + snapshot_max_count: int + snapshot_max_total_gb: float + snapshot_retention_days: float + snapshot_dir: str | None + allow_remote_snapshot: bool + runner: "Callable[..., object]" + launcher: "Callable[..., object]" + api_client: "Callable[..., object]" + sleeper: "Callable[..., object]" + audit_callback: object + audit_failure_count: int + _vms: dict + _guest_env_delivered: dict + def __init__(self, binary: str | None = None, kernel: str | None = None, rootfs: str | None = None, @@ -253,7 +309,8 @@ class CloudHypervisorExecutor(Executor): cpus: str | int | None = None, memory: str | None = None, state_dir: str | None = None, - runner: object = None, launcher: object = None, + runner: Callable[..., object] | None = None, + launcher: Callable[..., object] | None = None, ssh_key: str | None = None, ssh_user: str | None = None, ssh_port: int | str | None = None, @@ -265,12 +322,13 @@ class CloudHypervisorExecutor(Executor): cpuinfo_path: str = "/proc/cpuinfo", env: dict | None = None, model: str | None = None, ch_remote: str | None = None, - api_client: object = None, - sleeper: object = None, api_timeout: float | None = None, + api_client: Callable[..., object] | None = None, + sleeper: Callable[..., object] | None = None, + api_timeout: float | None = None, ready_timeout: float | None = None, ready_interval: float | None = None, host_key_mode: str | None = None) -> None: - self._env = env if env is not None else os.environ + self._env: dict = env if env is not None else os.environ # type: ignore[assignment] # 配置解析已拆到 cloud/chv/config.py（参数优先于环境 # 变量，默认值与 #216 一致；显式空用户/越界端口的收紧 # 见 config.py 头注释）。字段名与原属性一一对应，旧 @@ -291,7 +349,15 @@ class CloudHypervisorExecutor(Executor): self._copy_config_fields(_cfg) self._init_runtime(runner, launcher, api_client, sleeper) - def _init_runtime(self, runner, launcher, api_client, sleeper) -> None: + def _init_runtime(self, runner: Callable[..., object] | None, + launcher: Callable[..., object] | None, + api_client: Callable[..., object] | None, + sleeper: Callable[..., object] | None) -> None: + # to10：runner/launcher/api_client/sleeper 均为可调用注入对象； + # 形参保持 object 以兼容旧测试的任意 lambda，调用点用 + # getattr 可调用检查后调用（见 _call_runner 等），mypy + # 的 object-not-callable 在 _call_runner 处以 Callable 局部 + # 变量处理。 """运行时注入：seed 投递集合、API 客户端、睡眠器、 runner/launcher 默认值（纯移动）。""" # feat-chv-secfix（H2）：本实例已 seed 投递密钥的 VM @@ -346,7 +412,11 @@ class CloudHypervisorExecutor(Executor): 幂等：传入的已是 chv- 前缀名时不再二次加前缀（清理 扫描 socket 文件名时传的就是 chv- 名，二次加前缀会 把 chv-dead 算成 chv-chv-dead、永远清不掉）。""" - raw = task_or_name.get("id") if isinstance(task_or_name, dict) \ + # to10：id 按契约是 str；cast 只给 mypy 用，运行时原样 + # 传给 re.sub（id 缺失时仍按旧行为 TypeError，不静默 + # 编成 "None"）。 + from typing import cast as _cast + raw = _cast(str, task_or_name.get("id")) if isinstance(task_or_name, dict) \ else str(task_or_name) safe = re.sub(r"[^A-Za-z0-9_.-]", "-", raw) if safe.startswith("chv-"): @@ -589,7 +659,7 @@ class CloudHypervisorExecutor(Executor): - egress_not_configured：CHV_EGRESS_PROXY 未设且未显式 允许直连（默认 fail-closed，VM 出口必须过白名单代理） """ - missing = [] # (kind, 文本) 列表，全部收集完再一次抛 + missing: list = [] # (kind, 文本) 列表，全部收集完再一次抛 self._preflight_host_checks(missing) self._preflight_file_checks(missing) self._preflight_net_checks(missing) @@ -637,6 +707,34 @@ class CloudHypervisorExecutor(Executor): self, "audit_failure_count", 0) + 1 _LOGGER.error("chv_audit_failed event=%s kind=%s error=%s", event, kind, exc) + # to10（ADR-14）：回调失败的事件再落盘一份—— + # 宿主日志可能轮转、计数只剩数字，落盘文件是 + # 唯一能事后找回“哪次事件丢了”的取证。落盘 + # 本身 best-effort，失败只记日志，绝不炸任务。 + self._spill_audit(event, task, detail, kind) + + def _spill_audit(self, event, task, detail, kind) -> None: + """审计失败事件落盘（to10，见 ADR-14 取舍）： + state_dir/chv-audit-spill.jsonl，每行一 JSON + {ts, event, kind, detail, task_id}；文件超 1MB 时 + 先轮转成 .old 再写。detail 沿用 _emit 契约，绝不 + 含密钥值。落盘不是 DB 替代：timeline 查询看不到它， + 需宿主人工查阅；轮转只留最新两代，旧事件会丢。""" + try: + path = os.path.join(self.state_dir, + "chv-audit-spill.jsonl") + if os.path.exists(path) \ + and os.path.getsize(path) > 1024 * 1024: + os.replace(path, path + ".old") + task_id = task.get("id") if isinstance(task, dict) \ + else str(task) + line = json.dumps({"ts": time.time(), "event": event, + "kind": kind, "detail": detail, + "task_id": task_id}) + with open(path, "a") as fh: + fh.write(line + "\n") + except Exception as exc: + _LOGGER.error("chv_audit_spill_failed error=%s", exc) # -- 命令构造（纯函数，不做文件校验） ------------------------------- def _build_vm_command(self, task): @@ -671,7 +769,7 @@ class CloudHypervisorExecutor(Executor): return _known_hosts_path(self.state_dir, self.vm_name(task_or_name)) - def _net_guest_ip(self): + def _net_guest_ip(self) -> str | None: """VmConfig net 首个带 ip 的条目（静态 tap 客机 IP）。""" if not self.net_json: return None @@ -691,8 +789,9 @@ class CloudHypervisorExecutor(Executor): 显式 CHV_SSH_HOST > net 静态客机 IP > 显式转发宣称下 的回环；三者皆无时报 ChvFailure（ssh_target_undeclared）， 绝不无条件假设 127.0.0.1:2222 能通。""" - if getattr(self, "ssh_host", None): - return self.ssh_host + _ssh_host = self.ssh_host + if _ssh_host: + return _ssh_host guest_ip = self._net_guest_ip() if guest_ip: return guest_ip @@ -704,6 +803,30 @@ class CloudHypervisorExecutor(Executor): "里给 net 条目配 ip（静态 tap 客机 IP）、或显式设 " "CHV_SSH_FORWARD=1 宣称宿主 NAT 已转发回环端口") + def _read_keypair_texts(self, paths: dict) -> tuple: + """读 key 对文本（to10 提取，provision_host_key 降行用）。""" + with open(paths["public"], "r") as fh: + public_text = fh.read() + with open(paths["private"], "r") as fh: + private_text = fh.read() + return public_text, private_text + + def _write_provisioned_known_hosts(self, paths: dict, + public_text: str) -> None: + """写预置 known_hosts（to10 提取）。主机部分跟 SSH + 目标走（客机 IP 时不再写死回环，否则钉死校验永远 + 对不上）；解析失败时回环仅作钉死写入，不放宽校验。""" + from cloud.chv.hostkeys import known_hosts_line + from cloud.chv.ssh_transport import write_known_hosts + try: + ssh_host = self.resolve_ssh_host() + except ChvFailure: + ssh_host = "127.0.0.1" + write_known_hosts( + paths["known_hosts"], + [known_hosts_line(self.ssh_port, public_text, + host=ssh_host)]) + def provision_host_key(self, task: dict, secret_env: dict | None = None) -> dict: """按 VM 预置 host key（补缺口轨强方案，见 hostkeys.py）： 启动前为该 VM 生成一次性 ed25519 host key 对，公钥 @@ -718,10 +841,13 @@ class CloudHypervisorExecutor(Executor): 脏标记存在时直接拒绝（清理失败即拒起，不再复用 旧 key 对）。成功后审计 host_key_generated（只含 VM 名与密钥名，不含任何 key 值）。""" + # to10：known_hosts 写入已抽到 _write_provisioned_known_hosts， + # 此处只留生成与 seed 写入的 import（删后不再有未用 import）。 from cloud.chv.hostkeys import ( - dirty_marker_path, generate_host_keypair, - known_hosts_line, write_seed) - from cloud.chv.ssh_transport import write_known_hosts + dirty_marker_path, + generate_host_keypair, + write_seed, + ) vm = self.vm_name(task) if os.path.exists(dirty_marker_path(self.state_dir, vm)): raise ChvFailure( @@ -732,20 +858,8 @@ class CloudHypervisorExecutor(Executor): paths = generate_host_keypair( self.state_dir, vm, runner=lambda cmd: self._call_runner(cmd)) - with open(paths["public"], "r") as fh: - public_text = fh.read() - with open(paths["private"], "r") as fh: - private_text = fh.read() - # feat-chv-fix：known_hosts 行的主机部分跟 SSH 目标走 - # （客机 IP 时不再写死回环，否则钉死校验永远对不上）。 - try: - ssh_host = self.resolve_ssh_host() - except ChvFailure: - ssh_host = "127.0.0.1" - write_known_hosts( - paths["known_hosts"], - [known_hosts_line(self.ssh_port, public_text, - host=ssh_host)]) + public_text, private_text = self._read_keypair_texts(paths) + self._write_provisioned_known_hosts(paths, public_text) seed_dir = write_seed(paths["seed_dir"], vm, private_text, public_text, secret_env=secret_env) @@ -819,7 +933,7 @@ class CloudHypervisorExecutor(Executor): from cloud.chv.hostkeys import GUEST_ENV_PATH delivered = self._guest_env_delivered.get( self.vm_name(task), set()) - missing = [name for name in secrets if name not in delivered] + missing: list = [name for name in secrets if name not in delivered] if missing: raise ChvFailure( "secret_env_not_delivered", @@ -835,6 +949,27 @@ class CloudHypervisorExecutor(Executor): guest_env_file=guest_env_file) # -- API 模式：VmConfig / API 调用 / 就绪轮询 ----------------------- + def _vm_config_disks(self, task_disk: str | None, + seed_disk: str | None) -> list: + """VmConfig 磁盘列表（to10 提取，含 L2 白名单校验）。 + 任务盘 + 可选 seed 只读盘（CIDATA ISO，cloud-init + NoCloud 数据源）；显式给的盘若不是基础镜像本身， + 须在 state_dir 内，否则 VmConfig 不能挂进宿主任意文件。""" + # 旧文案逐字保留：任务盘“任务磁盘须在…”，seed 盘 + # “seed 磁盘须在…”（label 本身含“磁盘”与空格）。 + for path, label in ((task_disk, "任务磁盘"), + (seed_disk, "seed 磁盘")): + if path and path != self.rootfs: + if not os.path.realpath(path).startswith( + os.path.realpath(self.state_dir) + os.sep): + raise ChvFailure( + "disk_path_not_allowed", + "%s须在 state_dir 内：%s" % (label, path)) + disks: list = [{"path": task_disk or self.rootfs}] + if seed_disk: + disks.append({"path": seed_disk, "readonly": True}) + return disks + def build_vm_config(self, task_disk: str | None = None, seed_disk: str | None = None) -> dict: """API 模式的 VmConfig（vm.create 请求体，纯函数）： cpus{boot_vcpus,max_vcpus}、memory{size: 字节}、 @@ -847,36 +982,14 @@ class CloudHypervisorExecutor(Executor): cpus = int(self.cpus) except ValueError: raise ValueError("CHV_CPU 必须是整数：%r" % (self.cpus,)) - # feat-chv-secfix（L2）：磁盘路径白名单。显式给的 - # task_disk 若不是基础镜像本身（reuse 模式的合法 - # 情形），须在 state_dir 内；seed 盘须在 state_dir - # 内——VmConfig 再也不能挂进宿主任意文件。 - if task_disk and task_disk != self.rootfs: - if not os.path.realpath(task_disk).startswith( - os.path.realpath(self.state_dir) + os.sep): - raise ChvFailure( - "disk_path_not_allowed", - "任务磁盘须在 state_dir 内：%s" % task_disk) - if seed_disk: - if not os.path.realpath(seed_disk).startswith( - os.path.realpath(self.state_dir) + os.sep): - raise ChvFailure( - "disk_path_not_allowed", - "seed 磁盘须在 state_dir 内：%s" % seed_disk) config = { "cpus": {"boot_vcpus": cpus, "max_vcpus": cpus}, "memory": {"size": _parse_memory_bytes(self.memory)}, "kernel": {"path": self.kernel}, "cmdline": {"args": self.cmdline}, - "disks": [{"path": task_disk or self.rootfs}], + "disks": self._vm_config_disks(task_disk, seed_disk), "rng": {"src": "/dev/urandom"}, } - # feat-chv-fix（P0-1）：seed 磁盘是 VmConfig 第二块 - # 只读盘（CIDATA 卷标 ISO，客机 cloud-init NoCloud - # 数据源），不是宿主目录——目录本身从不进 VmConfig。 - if seed_disk: - config["disks"].append({"path": seed_disk, - "readonly": True}) if self.net_json: try: net = json.loads(self.net_json) @@ -1230,7 +1343,7 @@ class CloudHypervisorExecutor(Executor): raise return secrets, net_entries - def _provision_seed_disk(self, task: dict, name: str, secrets) -> str: + def _provision_seed_disk(self, task: dict, name: str, secrets) -> str | None: """provisioned 模式的 seed 盘供给：生成→制盘→挂盘， 任一步失败即抛（调用方 stop_vm 清理），纯移动。""" seed_disk = None @@ -1248,8 +1361,10 @@ class CloudHypervisorExecutor(Executor): # 失败即 fail-fast（VM 尚未启动，此时拒绝最 # 安全，except 里 stop_vm 会清掉 seed）。 from cloud.chv.hostkeys import ( - build_seed_disk, delete_host_private_key, - seed_disk_path) + build_seed_disk, + delete_host_private_key, + seed_disk_path, + ) provisioned = self.provision_host_key( task, secret_env=secrets or None) seed_disk = seed_disk_path(self.state_dir, name) @@ -1498,8 +1613,8 @@ class CloudHypervisorExecutor(Executor): self.pause_vm(task_or_name) self.snapshot_vm(task_or_name, snapshot_url) self.stop_vm(task_or_name) - task = task_or_name if isinstance(task_or_name, dict) \ - else {"id": task_or_name} + # to10：删死赋值（task 建出后未用，F841；返回值直接 + # 用 task_or_name 算 VM 名，行为不变）。 return {"vm": self.vm_name(task_or_name), "state": "hibernated", "snapshot": snapshot_url} @@ -1616,7 +1731,8 @@ class CloudHypervisorExecutor(Executor): 非 live（无 api-socket 文件且无登记活进程）即删盘 并审计；删失败审计 vm_disk_delete_failed 并计入 failed。返回 {deleted, failed}。""" - deleted, failed = [], [] + deleted: list = [] + failed: list = [] try: entries = os.listdir(self.state_dir) except Exception: @@ -1693,7 +1809,7 @@ class CloudHypervisorExecutor(Executor): """崩溃残留扫描（H5）：私钥一律删、非 live VM 的 seed 同理删并审计；返回已清理 VM 名。worker 启动与 cleanup_stale_vms 时调它，不依赖 socket 是否存在。""" - cleaned = [] + cleaned: list = [] try: entries = os.listdir(self.state_dir) except Exception: @@ -1814,8 +1930,7 @@ class CloudHypervisorExecutor(Executor): extra_env=extra_env) result = self._call_runner(ssh_cmd, timeout=self.task_timeout) # 方法内 import：见模块头注释的循环导入说明。 - from cloud.sandbox import _parse_host_result - from cloud.sandbox import SimpleNamespaceShim + from cloud.sandbox import SimpleNamespaceShim, _parse_host_result parsed = _parse_host_result(task["id"], SimpleNamespaceShim(result)) parsed["executor"] = "cloud-hypervisor" diff --git a/cloud/chv/experimental_pool.py b/cloud/chv/experimental_pool.py index 5abccbb..ffb5df3 100644 --- a/cloud/chv/experimental_pool.py +++ b/cloud/chv/experimental_pool.py @@ -4,6 +4,7 @@ cloud/chv/pool.py 移到本文件，cloud/chv/__init__ 与 cloud/sandbox.py 不再 re-export 它，生产代码 import 本 模块即违规（见 docs/chv-architecture.md ADR-12 的禁 + import 清单与测试钉住）。实验用法只走 `from cloud.chv.experimental_pool import WarmPool`。 生产残留清理已彻底分文件：见 cloud/chv/orphans.py 的 @@ -53,18 +54,21 @@ Registry（本轨新增）： 仍可用，重启后孤儿改由 socket 扫描兜底）。 """ +from typing import TYPE_CHECKING + +if TYPE_CHECKING: + from cloud.chv.executor import CloudHypervisorExecutor + import os -from cloud.chv.orphans import _REGISTRY_FILENAME -from cloud.chv.orphans import _read_registry -from cloud.chv.orphans import _write_registry +from cloud.chv.orphans import _REGISTRY_FILENAME, _read_registry, _write_registry class WarmPool: """快照预热池（实验性，见模块头定案）。 wired_to_dispatch = False：生产 dispatch 不查池。 - + 目标：冷启动（preflight + 起 VMM + create + boot + SSH 就绪）是慢路径；预热池提前把干净 VM 起好并暂停，任务 @@ -81,7 +85,7 @@ class WarmPool: wired_to_dispatch = False - def __init__(self, executor: object, size: int | None = None, + def __init__(self, executor: "CloudHypervisorExecutor", size: int | None = None, snapshot_url: str | None = None, registry_path: str | None = None) -> None: self.executor = executor @@ -90,7 +94,7 @@ class WarmPool: self.snapshot_url = snapshot_url \ if snapshot_url is not None \ else getattr(executor, "snapshot_url", None) - self._entries = {} # 合成名 -> {"state", "vm", "task", ...} + self._entries: dict = {} # 合成名 -> {"state", "vm", "task", ...} self.registry_path = registry_path if self.registry_path is None: state_dir = getattr(executor, "state_dir", None) diff --git a/cloud/chv/hardening.py b/cloud/chv/hardening.py index c37f9b9..f71b1eb 100644 --- a/cloud/chv/hardening.py +++ b/cloud/chv/hardening.py @@ -20,7 +20,7 @@ from cloud.chv.errors import ChvFailure _NET_FIELDS = frozenset({"tap", "mac", "ip", "mask"}) -def ensure_secure_state_dir(state_dir: str) -> None: +def ensure_secure_state_dir(state_dir: str) -> str: """建 state_dir 并保证 0700（H6）：目录不存在即按 0700 建；已存在但组/其它位可访时 chmod 0700；失败即 ChvFailure（state_dir_insecure），绝不带着 /tmp 式 @@ -125,7 +125,7 @@ def _normalize_net_entry(entry: dict) -> dict: def _check_net_duplicates(normalized: list) -> None: """本次条目内 tap/mac/ip 重复即 net_config_invalid。""" - seen = {"tap": set(), "mac": set(), "ip": set()} + seen: dict[str, set[str]] = {"tap": set(), "mac": set(), "ip": set()} for entry in normalized: for kind in ("tap", "mac", "ip"): value = entry.get(kind) @@ -152,7 +152,7 @@ def _check_net_conflicts(normalized: list, existing_entries: list) -> None: % (kind, value, other.get("vm"))) -def validate_egress_proxy(egress_proxy: str) -> None: +def validate_egress_proxy(egress_proxy: str) -> str | None: """egress 代理地址校验（H1 代码门的一部分）：须为 http/https URL 且带主机名；否则 net_config_invalid… 实际 kind 用 config_invalid（preflight 的配置类失败）。 diff --git a/cloud/chv/hostkeys.py b/cloud/chv/hostkeys.py index 65ff0cd..34c40d0 100644 --- a/cloud/chv/hostkeys.py +++ b/cloud/chv/hostkeys.py @@ -32,6 +32,7 @@ import os import shlex +from collections.abc import Callable # seed 投递的客内密钥环境文件路径（feat-chv-secfix，H2）： # 只在 provisioned 模式的 seed user-data 里写出，0600 root， @@ -88,7 +89,8 @@ def seed_disk_path(state_dir: str, vm_name: str) -> str: return os.path.join(state_dir, vm_name, "seed.iso") -def build_seed_disk(seed_dir: str, seed_disk: str, runner: object = None) -> str: +def build_seed_disk(seed_dir: str, seed_disk: str, + runner: Callable[[list[str]], object] | None = None) -> str: """把 seed 目录做成 CIDATA 卷标 ISO（feat-chv-fix，P0-1 的 注入载体）：genisoimage -volid CIDATA -joliet -rock。 @@ -206,7 +208,7 @@ def write_seed(seed_dir: str, vm_name: str, private_key_text: str, def generate_host_keypair(state_dir: str, vm_name: str, - runner: object = None, + runner: Callable[[list[str]], object] | None = None, keygen: str = "ssh-keygen") -> dict: """为该 VM 生成（或复用本 VM 已有的）host key 对。 diff --git a/cloud/chv/ssh_transport.py b/cloud/chv/ssh_transport.py index 995c461..71899b1 100644 --- a/cloud/chv/ssh_transport.py +++ b/cloud/chv/ssh_transport.py @@ -52,7 +52,7 @@ import shlex _CHV_SECRET_ENV_NAMES = frozenset({"SUB2API_API_KEY"}) -def split_secret_env(extra_env: dict) -> tuple: +def split_secret_env(extra_env: dict | None) -> tuple[dict, dict]: """把 extra_env 拆成（密钥子集，非密钥子集）。""" secrets, plain = {}, {} for key, value in dict(extra_env or {}).items(): diff --git a/cloud/chv/worker_adapter.py b/cloud/chv/worker_adapter.py new file mode 100644 index 0000000..b7b32e5 --- /dev/null +++ b/cloud/chv/worker_adapter.py @@ -0,0 +1,196 @@ +"""CHV worker 适配模块（to10 轨，见 ADR-13）。 + +worker.py 里散落的 CHV 强相关接线抽到这里，worker 只留 +薄包装（方法名、签名、调用点不变，旧测试继续走 worker +方法）。本模块只依赖标准库 + cloud.chv 叶子模块 + +cloud.logging_config，不 import cloud.worker，绝无循环。 + +抽入本模块的（行为与错误文案逐字搬运）： +- call_goal_resume_breakpoint：通用断点调用（原 worker + 顶层函数 _call_goal_resume_breakpoint 的本体）。 +- make_chv_audit_callback：VM 生命周期事件进审计的回调 + （原 Worker.__init__ 里的闭包本体）。 +- reconcile_warm_pool：生产残留清理 facade 调用 + 审计 + （原 Worker._reconcile_warm_pool 的本体）。 +- cancel_watchdog_loop / quota_watchdog_loop：同步路径 + 两条看门狗循环（原 Worker._chv_*_watchdog 的本体， + self 依赖全部改成回调参数）。 +- watchdog_vm_name：看门狗启动时的 VM 名派生（原 + _run_attempt 里的 try/except 块）。 + +明确不抽的（见 ADR-13 诚实清单，worker 通用流程）： +cancel() 里的 cancel_vm 分支、cleanup_stale_containers +的 CHV 分支、网关守卫、_executor_name 的 CHV 分支、 +_run_attempt 的看门狗线程启动、VM 死亡写回——它们都 +嵌在 Docker/Local 共用的通用流程里，硬抽只会把通用 +流程撕成两半，行为不变的收益为零。 +""" + +import inspect +import threading +from collections.abc import Callable + +from cloud.chv.executor import CloudHypervisorExecutor +from cloud.logging_config import log_event + + +def is_chv_executor(executor: object) -> bool: + """是否为 CHV 执行器（worker 各处 isinstance 的同款判断）。""" + return isinstance(executor, CloudHypervisorExecutor) + + +def call_goal_resume_breakpoint(executor: object, task: dict, + model_provider: str | None = None, + extra_env: dict | None = None) -> object: + """通用断点调用：执行器提供 goal_resume_breakpoint_task + 时调用它。 + + Local 仍走原路径（调用方直接传 model_provider，不经过 + 这里，以免改生产语义）；这里只用于非 Local 执行器： + 签名接受 model_provider 时传它（CHV/Docker 均接受）， + 否则只传 task（旧假执行器单参）。feat-chv-secfix（H2）： + CHV 的 gateway_env 不再经 SSH env 前缀注入（密钥值进 + argv 的旧路径已废除），改走 provisioned seed 盘客内文件 + 投递；CHV 没有 Local 那样的进程 env 通道，seed 文件是 + 唯一通道，keyscan 模式下密钥投递 fail-closed。""" + method = getattr(executor, "goal_resume_breakpoint_task") + try: + params = inspect.signature(method).parameters + except (TypeError, ValueError): + return method(task) + kwargs: dict = {} + if "model_provider" in params: + kwargs["model_provider"] = model_provider + if "extra_env" in params and extra_env is not None: + kwargs["extra_env"] = extra_env + return method(task, **kwargs) + + +def make_chv_audit_callback(audit_fn: Callable) -> Callable: + """VM 生命周期事件进审计的回调（原 __init__ 闭包本体）。 + + audit_fn 即 Worker._audit（best-effort，失败不影响任务）； + 时间线展示 VM 事件时读审计表即可，不另起一套 VM 事件表； + 成本/配额仍走 worker 现有 quota 语义，本回调绝不记账。""" + + def _chv_audit(event, task, detail, kind=None): + try: + project_id = task.get("project_id") \ + if isinstance(task, dict) else None + audit_fn(project_id, "chv_vm_" + event, detail or "") + except Exception: + pass + + return _chv_audit + + +def reconcile_warm_pool(executor: object, audit_fn: Callable, + when: str) -> list: + """显式调生产残留清理 facade，返回已清理 VM 名。 + + 仅 CHV worker 有效（其它执行器返回 []）。实现是 + cloud.chv.orphans.reconcile_orphans_for_executor()： + 直接读 registry 文件做清理，不构造 WarmPool、不碰 + 实验接口。调用失败不炸主流程，但进审计（失败本身 + 也审计，不静默丢弃）。orphan 条目绝不发给任务， + reconcile 只做清理。""" + if not isinstance(executor, CloudHypervisorExecutor): + return [] + try: + from cloud.chv.orphans import reconcile_orphans_for_executor as _reconcile + cleaned = _reconcile(executor) or [] + except Exception as exc: + try: + audit_fn(None, "chv_warm_pool_reconcile_failed", + "when=%s error=%s" % (when, exc)) + except Exception: + pass + return [] + if cleaned: + try: + audit_fn(None, "chv_warm_pool_reconcile", + "when=%s cleaned=%s" % (when, ",".join(cleaned))) + except Exception: + pass + return cleaned + + +def cancel_watchdog_loop(task_id: str, vm_name: str | None, + stop_event: threading.Event, + cancel_requested: set, + get_task: Callable, + cancel_vm: Callable | None, + logger: object) -> None: + """CHV 同步路径的取消看门狗（feat-chv-fix，P1-2，Docker + 取消看门狗同款模式）：同步 SSH 阻塞时 worker 的 cancel() + 只能改库、清 _running 项，阻塞中的 run 本身不会醒； + 本线程每 1 秒查库/取消标志，命中时调 cancel_vm 真关 + VM，然后退出。查库失败时当没取消，不误杀。""" + while not stop_event.wait(1.0): + try: + if task_id in cancel_requested: + cancelled = True + else: + db_task = get_task(task_id) + cancelled = (db_task is not None and + db_task.get("status") == "interrupted") + except Exception: + cancelled = False + if not cancelled: + continue + try: + if cancel_vm is not None and vm_name: + cancel_vm(vm_name) + log_event(logger, "worker_chv_killed", + task_id=task_id, + message="cancel watchdog killed VM %s " + "(task interrupted/cancelled)" + % vm_name) + except Exception: + pass + return + + +def quota_watchdog_loop(task_id: str, vm_name: str | None, + stop_event: threading.Event, + quota_check: Callable | None, + mark_quota_hit: Callable, + cancel_vm: Callable | None, + logger: object, + interval: float = 30.0) -> None: + """CHV 同步路径的配额看门狗（feat-chv-fix，P1-2，Docker + 配额看门狗同款）：命中时记 _quota_hit_tasks 并调 + cancel_vm 真关 VM；_run_attempt 在同步返回后把结果 + 改写成 quota_exceeded，走统一配额中断收尾。""" + while not stop_event.wait(interval): + try: + over = bool(quota_check()) if quota_check else False + except Exception: + over = False + if not over: + continue + mark_quota_hit(task_id) + try: + if cancel_vm is not None and vm_name: + cancel_vm(vm_name) + log_event(logger, "worker_chv_killed", + task_id=task_id, + message="quota watchdog killed VM %s " + "(quota exceeded)" % vm_name) + except Exception: + pass + return + + +def watchdog_vm_name(executor: object, task: dict) -> str | None: + """看门狗启动时的 VM 名派生（原 _run_attempt 块）： + executor.vm_name(task) 抛错时回 None（不建看门狗）， + 绝不炸主派发流程。executor 无 vm_name 时等同抛错 + （原代码里 AttributeError 同样被 except 吞掉回 None）。""" + get_vm_name = getattr(executor, "vm_name", None) + if not callable(get_vm_name): + return None + try: + return get_vm_name(task) + except Exception: + return None diff --git a/cloud/console-react/src/api/chv.test.ts b/cloud/console-react/src/api/chv.test.ts index 3a73d9e..82ef4b6 100644 --- a/cloud/console-react/src/api/chv.test.ts +++ b/cloud/console-react/src/api/chv.test.ts @@ -1,5 +1,5 @@ import { describe, expect, it } from "vitest"; -import { chvAttachmentBlocked, chvAvailability, chvQueueHint, chvVmBadge, chvVmStatusText, countChvWorkers, isChvExecutor, isWorkerAlive } from "./chv"; +import { chvAttachmentBlocked, chvAvailability, chvAvailabilityText, chvQueueHint, chvVmBadge, chvVmDisplay, chvVmStatusText, countChvWorkers, isChvExecutor, isWorkerAlive } from "./chv"; // 纯逻辑测试：只测 chv.ts，不渲染组件、不发请求、不读 localStorage。 // alive 规则与 WorkersSection 现有口径一致：alive 字段优先， @@ -103,3 +103,53 @@ describe("chvQueueHint", () => { expect(chvQueueHint("cloud-hypervisor", "queued", { state: "known" as const, online: 1, total: 1 })).toBe(""); }); }); + +describe("chvVmDisplay（to10 四态）", () => { + it("字段 present=true → dead", () => { + const d = chvVmDisplay({ executor: "cloud-hypervisor", chv_vm_dead: true }); + expect(d.state).toBe("dead"); + expect(d.text).toContain("VM 中途死亡"); + }); + it("字段 absent → not-reported，不猜 live", () => { + const d = chvVmDisplay({ executor: "cloud-hypervisor" }); + expect(d.state).toBe("not-reported"); + expect(d.text).toContain("不代表 VM 活着"); + }); + it("任务未加载 → unknown", () => { + const d = chvVmDisplay(null); + expect(d.state).toBe("unknown"); + expect(d.text).toContain("无法判断"); + }); + it("非 CHV → not-applicable，text 为空", () => { + const d = chvVmDisplay({ executor: "docker", chv_vm_dead: true }); + expect(d.state).toBe("not-applicable"); + expect(d.text).toBe(""); + }); + it("旧入口仍是 display 的 text（unknown/not-applicable 回空）", () => { + expect(chvVmStatusText({ executor: "cloud-hypervisor", chv_vm_dead: true })).toContain("VM 中途死亡"); + expect(chvVmStatusText(null)).toBe(""); + expect(chvVmStatusText({ executor: "docker" })).toBe(""); + }); +}); + +describe("chvAvailabilityText（to10 四处统一文案）", () => { + it("known 零在线：不拦提交、后端是最终权威", () => { + const text = chvAvailabilityText({ state: "known", online: 0, total: 2 }); + expect(text).toContain("在线 0 / 共 2"); + expect(text).toContain("不拦提交"); + expect(text).toContain("最终权威"); + }); + it("known 非零：只有在线数，无排队警告", () => { + const text = chvAvailabilityText({ state: "known", online: 1, total: 1 }); + expect(text).toContain("在线 1 / 共 1"); + expect(text).not.toContain("排队"); + }); + it("未知三原因各有不同文案，绝不写成 0", () => { + expect(chvAvailabilityText({ state: "unknown", reason: "aggregate-only" })).toContain("仅全局管理员可见"); + expect(chvAvailabilityText({ state: "unknown", reason: "load-failed" })).toContain("加载失败"); + expect(chvAvailabilityText({ state: "unknown", reason: "not-loaded" })).toContain("尚未加载"); + for (const reason of ["aggregate-only", "load-failed", "not-loaded"] as const) { + expect(chvAvailabilityText({ state: "unknown", reason })).not.toContain("在线 0"); + } + }); +}); diff --git a/cloud/console-react/src/api/chv.ts b/cloud/console-react/src/api/chv.ts index 31e32f8..ce10c25 100644 --- a/cloud/console-react/src/api/chv.ts +++ b/cloud/console-react/src/api/chv.ts @@ -54,27 +54,85 @@ export function chvAttachmentBlocked( return isChvExecutor(executor) && hasAttachments; } -// VM 只读状态文案：输入只有后端已返回的两个字段，输出 -// 三种情形之一，绝不编造第四种： -// - chv_vm_dead 为 true → 已报告死亡（来源：worker 终局 -// 写回的 interrupted_reason，见 server _chv_vm_dead_from_row）； -// - CHV 任务但无该字段 → 未报告死亡，且明写“后端无 live -// 状态，不代表 VM 活着”（server 没有 VM 登记表）； -// - 非 CHV 任务 → 空串（不显示 VM 行）。 -export function chvVmStatusText(task: { +// VM 只读状态：四种情形各有明确文案，绝不编造第五种， +// 绝不猜 live（to10 精确化，见 ADR-15）： +// - dead：CHV 任务且 chv_vm_dead 为 true（字段已 present； +// 来源：worker 终局写回的 interrupted_reason，见 server +// _chv_vm_dead_from_row）； +// - not-reported：CHV 任务但字段 absent（后端只在死亡时 +// 才返回该字段；absent 只表示“未报告”，且明写“后端无 +// live 状态字段，不代表 VM 活着”，server 没有 VM 登记表）； +// - unknown：任务详情本身未加载（task 为 null），无法判断； +// - not-applicable：非 CHV 任务，不显示 VM 行（text 为空）。 +export type ChvVmDisplay = + | { state: "dead"; text: string } + | { state: "not-reported"; text: string } + | { state: "unknown"; text: string } + | { state: "not-applicable"; text: string }; + +export function chvVmDisplay(task: { executor?: string | null; chv_vm_dead?: boolean; -}): string { +} | null | undefined): ChvVmDisplay { + if (task === null || task === undefined) { + return { + state: "unknown", + text: "任务详情未加载，无法判断 VM 状态（后端无 live 状态字段，详情加载后才显示）", + }; + } // 死亡文案同样先过执行器门：后端只对 CHV 任务派生该字段， // 万一某天非 CHV 行误带该字段，前端也不把 docker 任务 // 写成“VM 死亡”。 if (task.chv_vm_dead && isChvExecutor(task.executor)) { - return "VM 中途死亡，任务未完成、断点可续跑（后端只读字段 chv_vm_dead）"; + return { + state: "dead", + text: "VM 中途死亡，任务未完成、断点可续跑（后端只读字段 chv_vm_dead）", + }; } if (isChvExecutor(task.executor)) { - return "未报告 VM 死亡（后端无 live 状态字段，不代表 VM 活着）"; + return { + state: "not-reported", + text: "未报告 VM 死亡（后端无 live 状态字段，不代表 VM 活着）", + }; } - return ""; + return { state: "not-applicable", text: "" }; +} + +// 旧入口（保留给已测调用点）：即 chvVmDisplay 的 text， +// not-applicable 与 unknown 均回空串（调用点只在已加载的 +// CHV 行上用它，unknown 分支由详情页的新入口处理）。 +export function chvVmStatusText(task: { + executor?: string | null; + chv_vm_dead?: boolean; +} | null | undefined): string { + const display = chvVmDisplay(task); + return display.state === "not-applicable" || + display.state === "unknown" + ? "" + : display.text; +} + +// CHV 可用性统一文案（to10）：新建任务、Workers、环境 +// 三处此前各写各的，Workers 的未知分支甚至恒写“仅全局 +// 管理员可见”（加载失败/未加载时也是这句，是文案 bug）。 +// 四处统一走本函数：known 零在线时不拦提交、后端是最终 +// 权威；未知时按 reason 分三句，绝不把未知写成 0。 +export function chvAvailabilityText( + availability: ChvAvailability, +): string { + if (availability.state === "known") { + const base = `CHV worker 在线 ${availability.online} / 共 ${availability.total}（来自 /workers）`; + return availability.online === 0 + ? `${base}：当前无 CHV worker 在线，任务将一直排队，请先确认宿主 worker 已起；本提示不拦提交，worker 随后上线可认领，后端是最终权威。` + : base; + } + if (availability.reason === "aggregate-only") { + return "CHV worker 在线数未知：完整 worker 清单仅全局管理员可见，非管理员只能看到聚合数，无法按执行器分项。"; + } + if (availability.reason === "load-failed") { + return "CHV worker 在线数未知：/workers 加载失败，请到 Workers 区刷新后重试。"; + } + return "CHV worker 在线数未知：/workers 尚未加载，请到 Workers 区刷新。"; } // 列表行徽标：只有已报告死亡时才出现短徽标；未报告时 diff --git a/cloud/console-react/src/sections/DispatchSection.tsx b/cloud/console-react/src/sections/DispatchSection.tsx index 8c3a8a4..3f1efb7 100644 --- a/cloud/console-react/src/sections/DispatchSection.tsx +++ b/cloud/console-react/src/sections/DispatchSection.tsx @@ -2,7 +2,7 @@ import { useRef, useState } from "react"; import { useConsole } from "../context"; import { Card, Bar } from "../components/ui"; import { gatherAttachments, shortId } from "../api/client"; -import { chvAttachmentBlocked, chvAvailability, isChvExecutor } from "../api/chv"; +import { chvAttachmentBlocked, chvAvailability, chvAvailabilityText, isChvExecutor } from "../api/chv"; import { EFFORTS } from "../api/types"; export default function DispatchSection() { @@ -125,14 +125,8 @@ export default function DispatchSection() { {isChv ?
实验执行器，默认关闭：需后端已开启 CHV worker 且宿主有 KVM，否则任务无人认领、一直排队；带附件时建任务即被后端拒绝（错误文案原样显示在下方）。前端只传 executor 值，不绕过后端白名单；默认仍是 docker。
: null} {isChv ? ( -
- {chvAvail.state === "known" - ? <>CHV worker 在线 {chvAvail.online} / 共 {chvAvail.total}（来自 /workers）{chvAvail.online === 0 ? ：当前无 CHV worker 在线，任务将一直排队，请先确认宿主 worker 已起；本提示不拦提交，worker 随后上线可认领，后端是最终权威。 : null} - : chvAvail.reason === "aggregate-only" - ? "CHV worker 在线数未知：完整 worker 清单仅全局管理员可见，非管理员只能看到聚合数，无法按执行器分项。" - : chvAvail.reason === "load-failed" - ? "CHV worker 在线数未知：/workers 加载失败，请到 Workers 区刷新后重试。" - : "CHV worker 在线数未知：/workers 尚未加载，请到 Workers 区刷新。"} +
+ {chvAvailabilityText(chvAvail)} {" "}
) : null} @@ -171,15 +165,15 @@ export default function DispatchSection() { {isSide ? null : (<> 附件（放进任务工作区，图片与普通文件不区分）： - - - + + + {isChv ?
cloud-hypervisor 任务不支持附件：附件输入已禁用（后端对带附件的 CHV 任务会 400，错误文案原样显示在下方；切换回其他执行器可恢复附件输入）。{hasAttachments ? <>当前仍有已选附件，请先清空再提交。 : ""}
: null} )} - {msg || ctx.dispatchMsg} + {msg || ctx.dispatchMsg} ); diff --git a/cloud/console-react/src/sections/EnvSection.tsx b/cloud/console-react/src/sections/EnvSection.tsx index 247b61b..01100c1 100644 --- a/cloud/console-react/src/sections/EnvSection.tsx +++ b/cloud/console-react/src/sections/EnvSection.tsx @@ -2,7 +2,7 @@ import { useState } from "react"; import { useConsole } from "../context"; import { Card, Bar, DataListRaw } from "../components/ui"; import { shortId } from "../api/client"; -import { chvAvailability } from "../api/chv"; +import { chvAvailability, chvAvailabilityText } from "../api/chv"; export default function EnvSection() { const ctx = useConsole(); @@ -37,7 +37,7 @@ export default function EnvSection() { - {executor === "cloud-hypervisor" ?
实验执行器，默认关闭：环境创建白名单同样允许 cloud-hypervisor（后端环境接口 allowlist：local / docker / cloud-hypervisor，不含 cloud）；仍需后端有 KVM 宿主上的 CHV worker 才会认领，否则任务排队。前端只传值，不绕过后端白名单。{chvAvail.state === "known" ? <> CHV worker 在线 {chvAvail.online} / 共 {chvAvail.total}（来自 /workers）{chvAvail.online === 0 ? "：当前无 CHV worker 在线，任务将一直排队。" : ""} : " CHV worker 在线数未知：完整 worker 清单仅全局管理员可见。"}
: null} + {executor === "cloud-hypervisor" ?
实验执行器，默认关闭：环境创建白名单同样允许 cloud-hypervisor（后端环境接口 allowlist：local / docker / cloud-hypervisor，不含 cloud）；仍需后端有 KVM 宿主上的 CHV worker 才会认领，否则任务排队。前端只传值，不绕过后端白名单。 {chvAvailabilityText(chvAvail)}
: null} {ctx.envMsg} ); diff --git a/cloud/console-react/src/sections/TaskDetailPanel.tsx b/cloud/console-react/src/sections/TaskDetailPanel.tsx index aca839c..f826d75 100644 --- a/cloud/console-react/src/sections/TaskDetailPanel.tsx +++ b/cloud/console-react/src/sections/TaskDetailPanel.tsx @@ -4,7 +4,7 @@ import { Card } from "../components/ui"; import { shortId, fmt } from "../api/client"; import { STATUS_EXPLAIN } from "../api/types"; import { firstChar, durationOf, tokenOf, fmtCostUsd } from "../api/taskMeta"; -import { chvAvailability, chvQueueHint, chvVmStatusText } from "../api/chv"; +import { chvAvailability, chvQueueHint, chvVmDisplay } from "../api/chv"; import DetailActions from "./detail/DetailActions"; import BrowserPanel from "./detail/BrowserPanel"; import ReviewPanel from "./detail/ReviewPanel"; @@ -131,7 +131,7 @@ export default function TaskDetailPanel() {
turn
{shortId(t.turn_id)}
root turn
{shortId(t.root_turn_id)}
executor
{t.executor || ""}{t.executor === "cloud-hypervisor" ? "（实验执行器，默认关闭；仅展示，无 VM 控制按钮）" : ""}
- {t.executor === "cloud-hypervisor" ?
VM 状态
{chvVmStatusText(t)}
: null} + {t.executor === "cloud-hypervisor" ?
VM 状态
{chvVmDisplay(t).text}
: null}
主机
{t.worktree_host || ""}
模型
{t.model || "默认"}
强度
{t.effort || "默认"}
diff --git a/cloud/console-react/src/sections/WorkersSection.tsx b/cloud/console-react/src/sections/WorkersSection.tsx index 06f05c4..94babae 100644 --- a/cloud/console-react/src/sections/WorkersSection.tsx +++ b/cloud/console-react/src/sections/WorkersSection.tsx @@ -2,7 +2,7 @@ import { useState } from "react"; import { useConsole } from "../context"; import { Card, Bar, DataListRaw } from "../components/ui"; import { workerAgeText, shortId } from "../api/client"; -import { chvAvailability } from "../api/chv"; +import { chvAvailability, chvAvailabilityText } from "../api/chv"; // Workers are read-only by design (GET /workers + worker-side // heartbeat only — no start/stop/kill actions exist to surface), so @@ -29,10 +29,8 @@ export default function WorkersSection() { return ( -
- {chvAvail.state === "known" - ? <>cloud-hypervisor worker 在线 {chvAvail.online} / 共 {chvAvail.total}（实验，默认关闭）{chvAvail.online === 0 ? "：当前无 CHV worker 在线，CHV 任务将一直排队。" : ""} - : "cloud-hypervisor worker 在线数未知：完整 worker 清单仅全局管理员可见，非管理员只能看到聚合数，无法按执行器分项。"} +
+ {chvAvailabilityText(chvAvail)}
setQuery(e.target.value)} /> diff --git a/cloud/console.html b/cloud/console.html index a7203c5..502c18c 100644 --- a/cloud/console.html +++ b/cloud/console.html @@ -4,7 +4,7 @@ Codex 命令台 - +`)}}b("流已断开")}catch(at){if(at.name==="AbortError")return;M(at instanceof Error?at.message:String(at)),b("未开始")}finally{p(!1)}},[r]),ft=d.useCallback(()=>{var z;(z=H.current)==null||z.abort(),p(!1),b("已停止")},[]),Et=d.useCallback(async z=>{if(r)try{const L=await I($.taskBrowserControl(r.id),"POST",{holder:z});A(L),M("")}catch(L){M(L instanceof Error?L.message:String(L))}},[r]),st=d.useCallback(async z=>{if(!r||(g==null?void 0:g.holder)!=="user"||!Z.current)return;const L=Z.current.getBoundingClientRect(),at=Math.round((z.clientX-L.left)/L.width*1280),jt=Math.round((z.clientY-L.top)/L.height*800);try{await I($.taskBrowserInput(r.id),"POST",{type:"mouse_click",x:at,y:jt})}catch(Mt){M(Mt instanceof Error?Mt.message:String(Mt))}},[r,g]),Ht=d.useCallback(async z=>{if(!(!r||(g==null?void 0:g.holder)!=="user")){z.preventDefault();try{z.key.length===1?await I($.taskBrowserInput(r.id),"POST",{type:"type",text:z.key}):await I($.taskBrowserInput(r.id),"POST",{type:"key",key:z.key})}catch(L){M(L instanceof Error?L.message:String(L))}}},[r,g]);if(!r)return c.jsx("div",{className:"muted",children:"点任务行查看详情"});const Q=(g==null?void 0:g.holder)||"—",ct=g!==null;return c.jsxs("div",{children:[c.jsxs("div",{style:{display:"flex",gap:8,flexWrap:"wrap",alignItems:"center"},children:[c.jsxs("span",{children:["状态 ",c.jsx("b",{children:O})]}),c.jsxs("span",{className:"muted",children:["会话 ",ct?"活跃":"未激活（轮次结束即消失）"]}),c.jsxs("span",{className:"muted",children:["控制权 ",Q]}),g!=null&&g.url?c.jsx("span",{className:"muted",title:g.url,children:g.url.slice(0,80)}):null]}),c.jsxs("div",{style:{display:"flex",gap:8,flexWrap:"wrap",marginTop:8},children:[q?c.jsx("button",{type:"button",onClick:ft,children:"停止观看"}):c.jsx("button",{type:"button",onClick:()=>void it(),children:"开始观看"}),c.jsx("button",{type:"button",onClick:()=>void F(),children:"刷新状态"}),Q!=="user"?c.jsx("button",{type:"button",onClick:()=>void Et("user"),disabled:!ct,children:"接管"}):c.jsx("button",{type:"button",onClick:()=>void Et("agent"),children:"交还给 Agent"})]}),N?c.jsx("div",{className:"muted",style:{marginTop:8},children:N}):null,Q==="user"?c.jsx("div",{className:"muted",style:{marginTop:8},children:"你已接管：点画面即点击，键盘输入直达页面。交还后 Agent 续跑前会先快照确认你的改动。"}):null,c.jsx("div",{style:{marginTop:8},children:h?c.jsx("img",{ref:Z,src:h,alt:"浏览器实时画面",tabIndex:0,onClick:z=>void st(z),onKeyDown:z=>void Ht(z),style:{width:"100%",maxWidth:1280,cursor:Q==="user"?"crosshair":"default",border:"1px solid var(--border, #ccc)"}}):c.jsx("div",{className:"muted",children:ct?"点“开始观看”接入画面流（接管后点画面即点击）":"浏览器会话未激活：任务本轮未开浏览器，或轮次已结束。"})})]})}function zg(){const i=Gt(),[r,h]=d.useState("uncommittedChanges"),[f,g]=d.useState(""),[A,O]=d.useState(""),[b,N]=d.useState("");return c.jsxs(c.Fragment,{children:[c.jsx("h3",{children:"评审队列"}),c.jsxs("select",{"aria-label":"评审目标",value:r,onChange:M=>h(M.target.value),children:[c.jsx("option",{value:"uncommittedChanges",children:"未提交改动（工作区）"}),c.jsx("option",{value:"baseBranch",children:"与分支的差异"}),c.jsx("option",{value:"commit",children:"指定提交"}),c.jsx("option",{value:"custom",children:"自定义评审指令"})]}),c.jsx("input",{className:"stacked",placeholder:"与分支的差异用：分支名（如 main）",style:{width:"100%"},value:f,onChange:M=>g(M.target.value)}),c.jsx("input",{className:"stacked",placeholder:"指定提交用：提交 sha",style:{width:"100%"},value:A,onChange:M=>O(M.target.value)}),c.jsx("textarea",{placeholder:"自定义评审指令用：评审指令正文",value:b,onChange:M=>N(M.target.value)}),c.jsxs(zt,{children:[c.jsx("button",{className:"primary",onClick:()=>void i.postReview({target_type:r,branch:f.trim()||void 0,sha:A.trim()||void 0,instructions:b.trim()||void 0}),children:"登记评审需求"}),c.jsx("button",{onClick:i.refreshReviews,children:"刷新队列"}),c.jsx("span",{className:"muted",children:i.reviewMsg})]}),c.jsx("div",{className:"muted",children:"本基线只登记评审需求（requested），不自动跑评审；评审是否执行、结果如何，以评审队列里的记录为准。"}),i.reviewError?c.jsx("div",{className:"muted",children:i.reviewError}):i.reviews.length===0?c.jsx("div",{className:"muted",children:"点任务行查看详情后，点“刷新队列”查看评审记录"}):c.jsx(Sn,{children:i.reviews.map(M=>{let q=M.target_type||"";return M.branch&&(q+="："+M.branch),M.sha&&(q+="："+String(M.sha).slice(0,12)),M.instructions&&(q+="："+String(M.instructions).slice(0,80)),c.jsxs("li",{children:[c.jsxs("span",{className:"row-main",children:[c.jsx("b",{children:"评审"})," ",se(M.id)," ｜ ",q," ｜ ",c.jsx("span",{className:"st-"+(M.status||""),children:M.status}),M.created_at?" ｜ "+Ge(M.created_at):"",c.jsx("br",{}),M.findings?String(M.findings).slice(0,1e3):c.jsx("span",{className:"muted",children:"（暂无 findings：评审尚未执行或尚未写回）"})]}),c.jsx("button",{onClick:()=>i.submitFindings(M.id),children:"写 findings"})]},M.id)})})]})}function Dg(){const i=Gt();return i.currentTask?i.currentTask.coordinator_parts?i.coordChildren.length===0?c.jsx("div",{className:"muted",children:"（子任务尚未创建）"}):c.jsx(Sn,{children:i.coordChildren.map(r=>c.jsx("li",{children:c.jsxs("span",{className:"row-main",children:["分工 ",c.jsx("b",{children:r.coordinator_part||""})," ｜ 状态 ",c.jsx("span",{className:"st-"+r.status,children:r.status})," ｜ ",se(r.id)," ｜ ",String(r.prompt||"").slice(0,60)]})},r.id))}):c.jsx("div",{className:"muted",children:"（非协调员任务）"}):c.jsx("div",{className:"muted",children:"点任务行查看详情后，这里显示协调员的子任务"})}function Rg(){const i=Gt();return i.currentTask?i.sideThreads.length===0?c.jsx("div",{className:"muted",children:"（暂无旁聊；列表“开旁聊”可在此任务下新建，不建 worktree）"}):c.jsx(Sn,{children:i.sideThreads.map(r=>c.jsx("li",{onClick:()=>i.showDetail(r.id),style:{cursor:"pointer"},children:c.jsxs("span",{className:"row-main",children:["旁聊 ｜ 状态 ",c.jsx("span",{className:"st-"+r.status,children:r.status})," ｜ ",se(r.id)," ｜ ",String(r.prompt||"").slice(0,60)]})},r.id))}):c.jsx("div",{className:"muted",children:"点任务行查看详情后，这里显示旁聊"})}function kg(){const i=Gt();if(!i.currentTask)return c.jsx("div",{className:"muted",children:"点任务行查看详情后，这里显示子代理链路"});if(i.agents.length===0)return c.jsxs("div",{className:"muted",children:["（暂无子代理）",i.agentsNote?c.jsx("span",{className:"err",children:i.agentsNote}):null]});const r={main:"主任务",delegate:"委派子任务",model:"模型子代理"},h={running:"进行中",completed:"已完成",failed:"失败",interrupted:"已中断",unknown:"未知"};return c.jsx(Sn,{children:i.agents.map((f,g)=>c.jsx("li",{children:c.jsxs("span",{className:"row-main",children:[c.jsxs("b",{children:[r[f.kind||""]||f.kind||"","："]}),f.name||""," ",c.jsx("span",{className:"st-"+(f.status||""),children:h[f.status||""]||f.status||"?"})," ｜ 过程调用 "+(f.calls?f.calls.length:0)+" 条",f.updated_ts?" ｜ 最近活动 "+Ge(f.updated_ts/1e3):"",f.prompt?c.jsxs(c.Fragment,{children:[c.jsx("br",{}),c.jsxs("span",{className:"muted",children:["任务：",f.prompt]})]}):null,f.calls&&f.calls.length?c.jsxs(c.Fragment,{children:[c.jsx("br",{}),f.calls.slice(-50).map((A,O)=>c.jsx("div",{children:xo(A)},O))]}):c.jsxs(c.Fragment,{children:[c.jsx("br",{}),c.jsx("span",{className:"muted",children:"（暂无过程调用）"})]})]})},g))})}function wg(){const i=Gt();return c.jsxs(c.Fragment,{children:[c.jsxs(zt,{children:[c.jsx("button",{onClick:i.refreshApprovals,children:"刷新审批队列"}),c.jsx("span",{className:"muted",children:i.approvalMsg})]}),c.jsx("div",{className:"muted",children:"待审批项需人工点记录允许或记录拒绝；超时默认拒绝。当前版本只留回执，不拦截实时行为（实时请求的事前拦截待 host 接入后生效，在此之前实时行为不变）。"}),c.jsx(ul,{empty:"（暂无审批项）",items:i.approvals.map(r=>c.jsxs("span",{className:"row-main",children:["[",r.kind,"｜",r.status,"] ",r.summary,["approved","denied","expired"].includes(r.status||"")?"｜回执："+(r.decision||r.status)+" by "+String(r.decided_by||"").slice(0,8)+" "+Ge(r.decided_at):"",r.status==="pending"?c.jsxs(c.Fragment,{children:[" ",c.jsx("button",{onClick:()=>i.decideApproval(r.id,"approved"),children:"记录允许"})," ",c.jsx("button",{onClick:()=>i.decideApproval(r.id,"denied"),children:"记录拒绝"})]}):null]},r.id))})]})}function Ug(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState("");return c.jsxs(c.Fragment,{children:[c.jsx("input",{placeholder:"目标任务 id（与本任务同项目；对方运行中排队执行、已终态则续跑复活，绝不打断当前轮）",style:{width:"100%"},value:r,onChange:A=>h(A.target.value)}),c.jsx("textarea",{placeholder:"消息正文（以本任务名义发出；对方 agent 会看到发送方任务 id，可照此回消息）",value:f,onChange:A=>g(A.target.value)}),c.jsxs(zt,{children:[c.jsx("button",{onClick:async()=>{await i.postPeer("queued",r,f),g("")},children:"排队发给对方"}),c.jsx("button",{className:"primary",onClick:async()=>{await i.postPeer("direct",r,f),g("")},children:"直接发送"}),c.jsx("span",{className:"muted",children:i.peerMsg})]}),c.jsx(ul,{empty:"（本任务尚未发出任务间消息）",items:i.peerMessages.map((A,O)=>c.jsxs("span",{className:"row-main",children:["[发给 ",String(A.task_id).slice(0,8),"｜",A.mode,"｜",A.status,"] ",String(A.body||"").slice(0,120)]},O))})]})}function Hg(){const i=Gt(),[r,h]=d.useState(""),f=i.currentTask?["completed","failed","interrupted"].includes(i.currentTask.status):!1;return c.jsxs(c.Fragment,{children:[c.jsx("div",{className:"muted",children:i.threadText||"点任务行查看详情后，这里显示线程元数据"}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"新线程名",value:r,onChange:g=>h(g.target.value)}),c.jsx("button",{disabled:!f,onClick:()=>i.threadAction("rename",r),children:"重命名"}),c.jsx("button",{disabled:!f,onClick:()=>i.threadAction("archive",""),children:"归档"}),c.jsx("button",{disabled:!f,onClick:()=>i.threadAction("unarchive",""),children:"取消归档"}),c.jsx("button",{disabled:!f,onClick:()=>i.threadAction("compact",""),children:"压缩"}),c.jsx("span",{className:"muted",children:i.threadGateNote}),c.jsx("span",{className:"muted",children:i.threadMsg})]}),c.jsx(ul,{empty:"暂无其他线程",items:i.threadList.map((g,A)=>c.jsxs("span",{className:"row-main",children:[g.name||g.thread_name||g.id||""," [",g.id||"","]",g.archived?" [已归档]":"",g.cwd?" "+g.cwd:""]},A))}),c.jsx("div",{className:"muted",children:"限制：Docker/异机任务线程操作一律 409；merged/merge-failed 不可改；turn/steer 与 review/start 不在本批。"})]})}function Bg(i){return/503|Service Unavailable/.test(i)?"上游模型服务暂时不可用（503）：模型一轮都没跑出来，不是任务逻辑失败；等上游恢复或换模型重试":/429|rate limit/i.test(i)?"上游限流（429）：模型拒绝了本次调用，可稍后重试":/timeout|timed out/i.test(i)?"上游调用超时：模型响应未在限时内返回":/401|403|Invalid API key|unauthorized/i.test(i)?"上游鉴权失败：网关 key 或权限有问题":/ECONNREFUSED|ENOTFOUND|network|fetch failed/i.test(i)?"连不上上游模型服务：网络或网关地址有问题":"任务失败（原因见原始错误文本）"}function qg(i){return i===null||i<0?"未知":i<60?Math.floor(i)+" 秒前":i<3600?Math.floor(i/60)+" 分钟前":(i/3600).toFixed(1)+" 小时前"}const Yg=[["overview","概览"],["collab","协作"],["review","评审与审批"],["threads","线程"],["browser","浏览器"]];function Gg(){const i=Gt(),r=i.currentTask,[h,f]=d.useState("overview");if(!r)return c.jsx(Jt,{id:"nav-detail",title:"任务详情",children:c.jsx("div",{className:"muted",children:"点任务行查看详情"})});const[g,A]=lv(r),[O,b]=av(r),N=fi(i.workers,i.workersSummary,i.workersError),M=qc(r.executor,r.status,N),q=ev(r),p=i.costSummary,H=r.plan?"计划模式（只读）":r.goal_mode?r.no_time_limit?"goal 模式（不限轮次，模型宣告完成；任务不限时）":"goal 模式（不限轮次，模型宣告完成）":r.verify_command?"goal 模式（旧语义，验收命令）":null;return c.jsxs("div",{className:"detail-stack",id:"nav-detail",children:[c.jsx("nav",{className:"detail-tabs","aria-label":"任务详情页签",children:Yg.map(([Z,F])=>c.jsx("button",{type:"button",className:h===Z?"active":"",onClick:()=>f(Z),children:F},Z))}),h==="overview"?c.jsxs(c.Fragment,{children:[c.jsxs(Jt,{title:"任务详情",children:[c.jsxs("div",{style:{display:"flex",alignItems:"center",gap:8,flexWrap:"wrap"},children:[c.jsx("span",{className:"issue-first",title:"任务首字："+q,"aria-hidden":"true",children:q}),c.jsxs("span",{children:["状态 ",c.jsx("b",{className:"st-"+r.status,children:r.status}),"（",eg[r.status]||"未知状态","）"]}),c.jsxs("span",{className:"muted",children:["更新 ",Ge(r.updated_at)]})]}),c.jsxs("div",{className:"detail-cards",children:[c.jsxs("div",{className:"detail-card",title:A,children:[c.jsx("div",{className:"detail-card-label",children:"耗时"}),c.jsx("div",{className:"detail-card-value",children:g})]}),c.jsxs("div",{className:"detail-card",title:b,children:[c.jsx("div",{className:"detail-card-label",children:"token"}),c.jsx("div",{className:"detail-card-value",children:O})]}),c.jsxs("div",{className:"detail-card",title:"LLM 请求数 = GET /tasks/{id}/cost 的 llm_requests，即 cost_records 的 completions 之和：host.py 每收到一次 thread/tokenUsage/updated（一次上游补全）计 1，worker 每轮终局记账时累加；估算行（无真实用量）计 0；部署前已记账的老任务没有 completions 列，显示 0 而非回填数",children:[c.jsx("div",{className:"detail-card-label",children:"LLM 请求数"}),c.jsx("div",{className:"detail-card-value",children:p===void 0?"加载中…":p===null?"—":(p.llm_requests??0).toLocaleString()}),p===void 0?c.jsx("div",{className:"muted",children:"成本加载中…"}):p===null?c.jsx("div",{className:"muted",children:"成本明细加载失败，LLM 请求数不可用"}):c.jsxs("div",{className:"muted",children:["上游补全次数之和｜ 轮数 ",p.turns??0," ｜ fallback ",p.fallback_count??0]})]}),c.jsxs("div",{className:"detail-card",title:"成本明细：GET /tasks/{id}/cost 聚合；美元只计有价行，无价 token 单列；缓存是输入 token 的子集，不另计",children:[c.jsx("div",{className:"detail-card-label",children:"成本"}),c.jsx("div",{className:"detail-card-value",children:p===void 0?"加载中…":p===null?"—":jg(p.usd_micros)}),p===void 0?c.jsx("div",{className:"muted",children:"成本加载中…"}):p===null?c.jsx("div",{className:"muted",children:"成本明细加载失败（“查成本”按钮仍可手动拉取原始 JSON）"}):c.jsxs("div",{className:"muted",children:["token ",(p.total_tokens??0).toLocaleString(),"（实际 ",(p.actual_tokens??0).toLocaleString()," / 估算 ",(p.estimated_tokens??0).toLocaleString(),"，缓存 ",(p.cached_input_tokens??0).toLocaleString(),"，无价 ",(p.unpriced_actual_tokens??0).toLocaleString(),"）｜ fallback ",p.fallback_count??0]})]})]})]}),r.status==="running"||r.status==="queued"?(()=>{const Z=Date.now()/1e3,F=r.last_heartbeat||r.updated_at||null,it=F?Z-F:null;return c.jsxs(Jt,{title:"LLM 进度",children:[c.jsxs("div",{children:["状态 ",r.status==="running"?"进行中":"排队中"," · 已用 ",g,it!==null?" · 上次活动 "+qg(it):""]}),M?c.jsx("div",{className:"muted",children:M}):null,it!==null&&it>180?c.jsx("div",{className:"muted",children:"超过 3 分钟无活动：可能在等上游模型响应（上游 503 时 worker 会重试后失败，失败后这里会显示失败原因；无活动不等于已断连，服务端没有“LLM 已连接”字段，不作此断言）"}):c.jsx("div",{className:"muted",children:"首轮无输出时属正常：模型第一轮回复到达前 trace 里没有消息；回复到达后会在会话区出现，成本里的 LLM 请求数随之增加"})]})})():null,r.error_text?c.jsxs(Jt,{title:"失败原因",children:[c.jsx("div",{children:Bg(r.error_text)}),c.jsx("pre",{style:{whiteSpace:"pre-wrap",overflowWrap:"anywhere"},children:r.error_text.slice(0,1e3)}),c.jsxs("div",{className:"muted",children:["完整错误文本仍在下方“操作与结果”里；模型 ",r.model||"默认"," · executor ",r.executor||""]})]}):null,c.jsx(Jt,{title:"基本信息",children:c.jsxs("dl",{className:"detail-kv",children:[c.jsxs("div",{children:[c.jsx("dt",{children:"任务"}),c.jsx("dd",{children:r.id})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"turn"}),c.jsx("dd",{children:se(r.turn_id)})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"root turn"}),c.jsx("dd",{children:se(r.root_turn_id)})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"executor"}),c.jsxs("dd",{children:[r.executor||"",r.executor==="cloud-hypervisor"?"（实验执行器，默认关闭；仅展示，无 VM 控制按钮）":""]})]}),r.executor==="cloud-hypervisor"?c.jsxs("div",{children:[c.jsx("dt",{children:"VM 状态"}),c.jsx("dd",{role:"status","aria-label":"VM 状态",children:Sg(r).text})]}):null,c.jsxs("div",{children:[c.jsx("dt",{children:"主机"}),c.jsx("dd",{children:r.worktree_host||""})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"模型"}),c.jsx("dd",{children:r.model||"默认"})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"强度"}),c.jsx("dd",{children:r.effort||"默认"})]}),r.child_model||r.child_effort?c.jsxs("div",{children:[c.jsx("dt",{children:"子任务默认"}),c.jsx("dd",{children:(r.child_model||"跟随本任务")+" / "+(r.child_effort||"跟随本任务")})]}):null,c.jsxs("div",{children:[c.jsx("dt",{children:"权限"}),c.jsx("dd",{children:r.approval||"默认"})]}),c.jsxs("div",{children:[c.jsx("dt",{children:"项目"}),c.jsx("dd",{children:se(r.project_id)})]}),H?c.jsxs("div",{children:[c.jsx("dt",{children:"模式"}),c.jsx("dd",{children:H})]}):null,r.kind==="side"?c.jsxs("div",{children:[c.jsx("dt",{children:"类型"}),c.jsx("dd",{children:"旁聊（无 worktree/分支/容器）"})]}):null,r.forked_from?c.jsxs("div",{children:[c.jsx("dt",{children:"分支自"}),c.jsx("dd",{children:se(r.forked_from)})]}):null,r.coordinator_parts?c.jsxs("div",{children:[c.jsx("dt",{children:"协调员"}),c.jsxs("dd",{children:[r.coordinator_parts.length," 个子任务（",r.coordinator_parts.map(Z=>Z.name).join("、"),"）"]})]}):null,r.parent_task_id&&r.kind==="side"?c.jsxs("div",{children:[c.jsx("dt",{children:"父会话"}),c.jsxs("dd",{children:["父任务 ",se(r.parent_task_id)," ",c.jsx("button",{type:"button",onClick:()=>i.showDetail(r.parent_task_id),children:"跳转到父任务"})]})]}):null,r.parent_task_id&&r.kind!=="side"?c.jsxs("div",{children:[c.jsx("dt",{children:"协调子任务"}),c.jsxs("dd",{children:["父任务 ",se(r.parent_task_id),r.coordinator_part?"（"+r.coordinator_part+"）":""]})]}):null,r.attachments&&r.attachments.length?c.jsxs("div",{children:[c.jsx("dt",{children:"附件"}),c.jsx("dd",{children:r.attachments.join("、")})]}):null,r.pr_url?c.jsxs("div",{children:[c.jsx("dt",{children:"PR"}),c.jsx("dd",{children:c.jsxs("a",{href:r.pr_url,target:"_blank",rel:"noopener noreferrer",children:["PR #",r.pr_number]})})]}):null]})}),c.jsx(Jt,{title:"prompt",children:c.jsx("div",{style:{whiteSpace:"pre-wrap",overflowWrap:"anywhere"},children:r.prompt||"（无 prompt）"})}),c.jsxs(Jt,{title:"操作与结果",children:[c.jsx(Mg,{}),c.jsx("pre",{children:i.detailBody||"（暂无 result/error 输出）"}),i.costBox?c.jsx("pre",{children:i.costBox}):null]})]}):h==="collab"?c.jsxs(c.Fragment,{children:[c.jsx(Jt,{title:"协调子任务",children:c.jsx(Dg,{})}),c.jsx(Jt,{title:"旁聊",children:c.jsx(Rg,{})}),c.jsx(Jt,{title:"子代理链路",children:c.jsx(kg,{})}),c.jsx(Jt,{title:"给其他任务发消息",children:c.jsx(Ug,{})})]}):h==="review"?c.jsxs(c.Fragment,{children:[c.jsx(Jt,{title:"评审队列",children:c.jsx(zg,{})}),c.jsx(Jt,{title:"运行时审批队列",children:c.jsx(wg,{})})]}):h==="browser"?c.jsx(c.Fragment,{children:c.jsx(Jt,{title:"浏览器远控",children:c.jsx(Og,{})})}):c.jsx(c.Fragment,{children:c.jsx(Jt,{title:"线程",children:c.jsx(Hg,{})})})]})}function Qg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState(""),[A,O]=d.useState([]),[b,N]=d.useState(null),[M,q]=d.useState(!1),[p,H]=d.useState(""),[Z,F]=d.useState(""),it=d.useCallback(async()=>{if(!i.keySaved){O([]);return}try{const Q=await I($.userKeys);O(Q.keys||[]),F("")}catch(Q){F(Q instanceof Error?Q.message:"加载 key 列表失败")}},[i.keySaved]);d.useEffect(()=>{it()},[it]);const ft=async()=>{H(""),F(""),q(!1);try{const Q=await I($.userKeys,"POST",{label:f.trim()||"api-key"});N(Q),g(""),H("新 key 已生成，只显示一次：复制后请妥善保存。"),await it()}catch(Q){F(Q instanceof Error?Q.message:"生成 key 失败")}},Et=async()=>{if(b)try{await navigator.clipboard.writeText(b.api_key),q(!0)}catch{F("复制失败：请手动选中上方 key 文本复制")}},st=()=>{b&&(i.saveKeyInput(b.api_key),N(null),H("已存为当前浏览器 key（原有 key 未被撤销，需要时可单独撤销）。"))},Ht=async Q=>{const ct=(()=>{try{return nl().slice(-4)}catch{return""}})(),z=Q.id==="legacy"?"（主 key：撤销后当前浏览器保存的 key 也可能随之失效，需重新登录拿新 key）":Q.hint&&Q.hint===ct?"（这就是本浏览器当前在用的 key，撤销后本页会立刻未连接）":"";if(window.confirm(`撤销 key「${Q.label}」（后 4 位 ${Q.hint||"无"}）吗？${z}撤销不可恢复。`)){H(""),F("");try{await I($.userKey(Q.id),"DELETE"),H(`已撤销 key「${Q.label}」。`),await it()}catch(L){F(L instanceof Error?L.message:"撤销失败")}}};return c.jsxs(Jt,{id:"nav-key",title:"API Key",children:[c.jsxs(zt,{children:[c.jsx("input",{type:"password",placeholder:"粘贴你的 X-API-Key（与调 API 用的同一个）",autoComplete:"off",value:r,onChange:Q=>h(Q.target.value)}),c.jsx("button",{className:"primary",onClick:()=>{i.saveKeyInput(r),h("")},children:"保存到本浏览器"}),c.jsx("button",{onClick:i.clearKeyInput,children:"清除"}),c.jsx("span",{className:"muted",children:i.keyState})]}),c.jsx("div",{children:c.jsx("span",{className:"muted",children:i.quotaText})}),!i.keySaved&&c.jsx("div",{children:c.jsx("span",{className:"muted",children:"先粘贴并保存已有 key，再生成或管理 key（生成接口本身也要用已有 key 鉴权）。"})}),i.keySaved&&c.jsxs(c.Fragment,{children:[c.jsxs(zt,{children:[c.jsx("input",{placeholder:"新 key 备注（可选，默认 api-key，如 harness2-temp）",autoComplete:"off",value:f,onChange:Q=>g(Q.target.value)}),c.jsx("button",{className:"primary",onClick:()=>void ft(),children:"生成新 key"}),c.jsx("button",{onClick:()=>void it(),children:"刷新列表"})]}),b&&c.jsxs("div",{children:[c.jsx("div",{children:c.jsxs("span",{className:"muted",children:["新 key（",b.label,"，后 4 位 ",b.hint,"）：只显示一次，关闭后无法再查看。"]})}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",readOnly:!0,value:b.api_key,onFocus:Q=>Q.target.select()}),c.jsx("button",{className:"primary",onClick:()=>void Et(),children:M?"已复制":"复制"}),c.jsx("button",{onClick:st,children:"存为当前浏览器 key"}),c.jsx("button",{onClick:()=>N(null),children:"关闭（不再显示）"})]})]}),p&&c.jsx("div",{children:c.jsx("span",{className:"muted",children:p})}),Z&&c.jsx("div",{children:c.jsx("span",{children:Z})}),c.jsxs("div",{className:"api-key-list",children:[A.length===0&&c.jsx("span",{className:"muted",children:"暂无 key（或列表加载失败，见上方报错）。"}),A.map(Q=>c.jsxs("div",{children:[c.jsx("span",{children:Q.label})," ",c.jsxs("span",{className:"muted",children:["后 4 位 ",Q.hint||"无"," · 创建 ",Ge(Q.created_at)," · 最近使用 ",Q.last_used_at?Ge(Q.last_used_at):"未使用过"]})," ",Q.revoked_at?c.jsxs("span",{className:"muted",children:["已撤销（",Ge(Q.revoked_at),"）"]}):c.jsx("button",{onClick:()=>void Ht(Q),children:"撤销"})]},Q.id))]})]})]})}function Vg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState("");return c.jsxs(Jt,{id:"adminCard",title:c.jsxs(c.Fragment,{children:["管理区 ",c.jsx("span",{className:"muted",children:"（仅全局管理员）"})]}),children:[c.jsx("div",{className:"muted",children:i.adminGate}),i.adminBody?c.jsxs(c.Fragment,{children:[c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"要查配额的用户 id",value:i.adminUserId,onChange:A=>i.setAdminUserId(A.target.value)}),c.jsx("button",{onClick:i.adminQuota,children:"查该用户配额"})]}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"要设置配额的用户 id",value:r,onChange:A=>h(A.target.value)}),c.jsx("input",{type:"text",placeholder:"美元配额（留空 = 不限额）",value:f,onChange:A=>g(A.target.value)}),c.jsx("button",{className:"primary",onClick:()=>i.adminSetQuota(r,f),children:"设置配额"})]}),c.jsxs(zt,{children:[c.jsx("button",{onClick:i.adminSweep,children:"手动触发 sweep（清理陈旧 running 任务）"}),c.jsx("button",{onClick:i.adminBackup,children:"手动触发备份"})]}),c.jsx("div",{className:"muted",children:i.adminResult})]}):null]})}function Xg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState("contributor"),A=i.projects.find(O=>O.id===i.projectId);return c.jsxs(Jt,{title:"项目成员",children:[A?c.jsxs("div",{className:"muted",children:[A.name," ｜ ",A.id," ｜ 仓库 ",A.repo_url||"未绑定"," ｜ 我的角色 ",A.my_role||"未知"]}):c.jsx("div",{className:"muted",children:"先选项目"}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"成员 user_id 或用户名",value:r,onChange:O=>h(O.target.value)}),c.jsxs("select",{"aria-label":"邀请角色",value:f,onChange:O=>g(O.target.value),children:[c.jsx("option",{value:"viewer",children:"viewer（只读）"}),c.jsx("option",{value:"contributor",children:"contributor（可建任务）"}),c.jsx("option",{value:"maintainer",children:"maintainer（可批合并）"}),c.jsx("option",{value:"owner",children:"owner（可管项目）"})]}),c.jsx("button",{onClick:async()=>{await i.inviteMember(r,f),h("")},children:"邀请成员"}),c.jsx("button",{onClick:i.refreshMembers,children:"刷新成员"})]}),i.memberError?c.jsx("div",{className:"muted",children:i.memberError}):i.members.length===0?c.jsx("div",{className:"muted",children:"先选项目"}):c.jsx(Sn,{children:i.members.map(O=>{const b=(O.username||"")+(O.username?"（"+String(O.user_id).slice(0,8)+"）":String(O.user_id).slice(0,8));return c.jsxs("li",{children:[c.jsx("span",{className:"row-main",children:b}),c.jsx("select",{"aria-label":"成员角色",value:O.role||"viewer",onChange:N=>void i.changeMemberRole(O.user_id,N.target.value),children:lg.map(N=>c.jsx("option",{value:N,children:N},N))}),c.jsx("button",{onClick:()=>void i.removeMember(O.user_id),children:"移除"})]},O.user_id)})}),c.jsx("span",{className:"muted",children:i.memberMsg})]})}function Lg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState(""),[A,O]=d.useState("");return c.jsxs("div",{className:"project-nav",children:[c.jsxs(Jt,{title:"项目",children:[c.jsx(zt,{children:c.jsx("button",{onClick:i.refreshProjects,children:"刷新项目"})}),i.projects.length===0?c.jsx("div",{className:"muted",children:"暂无项目"}):c.jsx("ul",{className:"project-nav-list",children:i.projects.map(b=>c.jsxs("li",{className:"project-nav-item"+(b.id===i.projectId?" selected":""),onClick:()=>i.selectProject(b.id),title:b.id,children:[c.jsx("span",{className:"project-nav-name",children:b.name}),c.jsxs("span",{className:"project-nav-meta",children:[se(b.id),b.my_role?" · "+b.my_role:""]})]},b.id))}),c.jsx("h3",{children:"创建项目"}),c.jsx(zt,{children:c.jsx("input",{type:"text",placeholder:"新项目名",value:r,onChange:b=>h(b.target.value)})}),c.jsx(zt,{children:c.jsxs("select",{"aria-label":"从 GitHub 仓库选",value:f,onChange:b=>{g(b.target.value),b.target.value&&O(b.target.value)},children:[c.jsx("option",{value:"",children:"（从 GitHub App 安装仓库里选，或手动输入）"}),i.githubRepos.map(b=>c.jsx("option",{value:b.clone_url,children:b.repo_full_name},b.clone_url))]})}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"repo_url（可选）",value:A,onChange:b=>O(b.target.value)}),c.jsx("button",{className:"primary",onClick:async()=>{await i.createProject(r,A),h(""),O(""),g("")},children:"创建项目"})]}),c.jsx("span",{className:"muted",children:i.projectMsg})]}),c.jsx(Xg,{})]})}function Zg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState(""),[A,O]=d.useState(""),b=fi(i.workers,i.workersSummary,i.workersError);return c.jsxs(Jt,{id:"nav-env",title:c.jsxs(c.Fragment,{children:["环境 ",c.jsx("span",{className:"muted",children:"（选、建、管：当前项目的基准检出）"})]}),children:[c.jsxs(zt,{children:[c.jsx("button",{onClick:i.refreshEnvs,children:"刷新"}),c.jsx("span",{className:"muted",children:i.envCount})]}),i.envs.length===0?c.jsx("div",{className:"muted",children:"先选项目"}):c.jsx(Sn,{children:i.envs.map(N=>c.jsxs("li",{children:[c.jsxs("span",{className:"row-main",children:[N.name,"（",se(N.id),"） / 分支 ",N.branch||"默认分支"," / 执行器 ",N.executor||"不限"," / 状态 ",N.status,N.status==="failed"&&N.error_text?" / 失败原因："+N.error_text:""]}),N.status==="archived"?null:c.jsx("button",{onClick:()=>void i.archiveEnv(N.id),children:"归档"})]},N.id))}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"新环境名",value:r,onChange:N=>h(N.target.value)}),c.jsx("input",{type:"text",placeholder:"分支（可选，空=默认分支）",value:f,onChange:N=>g(N.target.value)}),c.jsxs("select",{"aria-label":"新环境执行器",value:A,onChange:N=>O(N.target.value),children:[c.jsx("option",{value:"",children:"执行器：不限"}),c.jsx("option",{value:"local",children:"local"}),c.jsx("option",{value:"docker",children:"docker"}),c.jsx("option",{value:"cloud-hypervisor",children:"cloud-hypervisor（实验，默认关闭）"})]}),c.jsx("button",{onClick:async()=>{await i.createEnv(r,f,A),h(""),g("")},children:"新建环境"})]}),A==="cloud-hypervisor"?c.jsxs("div",{className:"muted","aria-live":"polite",children:["实验执行器，默认关闭：环境创建白名单同样允许 cloud-hypervisor（后端环境接口 allowlist：local / docker / cloud-hypervisor，不含 cloud）；仍需后端有 KVM 宿主上的 CHV worker 才会认领，否则任务排队。前端只传值，不绕过后端白名单。 ",To(b)]}):null,c.jsx("span",{className:"muted",children:i.envMsg})]})}function Kg(){const i=Gt(),[r,h]=d.useState("");return c.jsxs(Jt,{title:"仓库与环境上下文",children:[c.jsx("div",{className:"muted",children:i.envRepoLine||"先选项目"}),c.jsx("div",{className:"muted",children:i.envCount}),i.envRepoBar?c.jsxs(zt,{children:[c.jsxs("select",{"aria-label":"补仓库",value:r,onChange:f=>h(f.target.value),children:[c.jsx("option",{value:"",children:"（从 GitHub App 安装仓库里选）"}),i.githubRepos.map(f=>c.jsx("option",{value:f.clone_url,children:f.repo_full_name},f.clone_url))]}),c.jsx("button",{onClick:()=>void i.saveEnvRepo(r),children:"保存仓库"})]}):null,c.jsx("span",{className:"muted",children:i.envRepoMsg})]})}function Jg(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState("all"),A=M=>{const q=Yc(M);return M.alive!==void 0&&M.alive!==null?!!M.alive:q.age!==null&&q.age<=120},O=r.trim().toLowerCase(),b=(i.workers||[]).filter(M=>!(O&&!((M.worker_id||M.id||"")+" "+(M.executor||"")).toLowerCase().includes(O)||f==="online"&&!A(M)||f==="offline"&&A(M))),N=fi(i.workers,i.workersSummary,i.workersError);return c.jsxs(Jt,{id:"nav-workers",title:"Workers",children:[c.jsx(zt,{children:c.jsx("button",{onClick:i.refreshWorkers,children:"刷新"})}),c.jsx("div",{className:"muted","aria-live":"polite",children:To(N)}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"搜索 worker_id / 执行器","aria-label":"搜索 Worker",value:r,onChange:M=>h(M.target.value)}),c.jsxs("select",{"aria-label":"在线状态筛选",value:f,onChange:M=>g(M.target.value),children:[c.jsx("option",{value:"all",children:"在线状态：全部"}),c.jsx("option",{value:"online",children:"仅在线"}),c.jsx("option",{value:"offline",children:"仅离线"})]}),i.workers?c.jsxs("span",{className:"muted",children:["显示 ",b.length," / 共 ",i.workers.length," 个"]}):null]}),i.workersError?c.jsx("div",{className:"muted",children:i.workersError}):i.workers===null?c.jsxs("div",{className:"muted",children:[i.workersSummary||"未加载","（完整清单仅全局管理员可见，搜索/筛选仅对完整清单生效）"]}):i.workers.length===0?c.jsx("div",{className:"muted",children:"无 worker 注册"}):b.length===0?c.jsx("div",{className:"muted",children:"没有符合筛选条件的 worker"}):c.jsx(Sn,{children:b.map((M,q)=>{const p=M.running===void 0||M.running===null?"?":M.running,H=M.capacity===void 0||M.capacity===null?"?":M.capacity,Z=Yc(M),F=A(M),it=F?" / 在线":" / 离线（占用为上次上报值）";return c.jsx("li",{className:F?"is-online":"is-offline",children:c.jsxs("span",{className:"row-main",children:[M.worker_id||M.id||"?"," / ",M.executor||""," / 占用 ",p,"/",H,it," / 最后心跳 ",Z.text]})},q)})})]})}function Fg(){const i=Gt();return c.jsxs(Jt,{title:"健康与配额",children:[c.jsxs(zt,{children:[c.jsx("span",{className:"muted",children:i.healthText}),c.jsx("button",{onClick:i.refreshHealth,children:"刷新"})]}),c.jsx("div",{children:c.jsx("span",{className:"muted",children:i.quotaText})}),c.jsx("div",{className:"muted",children:i.workersSummary})]})}function Wg(){const i=Gt();return c.jsxs(Jt,{id:"nav-skills",title:"技能与 MCP",children:[c.jsx("div",{className:"muted",children:"反映宿主租户 home，容器内 home 可能不同；重载只在 schema 枚举里，当前 pin 的二进制拒绝时原样 502；OAuth 只发起登录，不代表登录完成。"}),c.jsxs(zt,{children:[c.jsx("button",{onClick:i.refreshSkills,children:"刷新"}),c.jsx("span",{className:"muted",children:i.skillMsg})]}),c.jsx("h3",{children:"技能"}),i.skillError?c.jsx("div",{className:"muted",children:i.skillError}):c.jsx(ul,{empty:"暂无技能",items:i.skills.map((r,h)=>c.jsxs("span",{className:"row-main",children:[r.name||r.id||"",r.description?" - "+r.description:"",r.detail?" - "+r.detail:""]},h))}),c.jsx("h3",{children:"应用"}),c.jsx(ul,{empty:"暂无应用",items:i.apps.map((r,h)=>c.jsxs("span",{className:"row-main",children:[r.name||r.id||"",r.description?" - "+r.description:""]},h))}),c.jsx("h3",{children:"MCP 服务"}),c.jsx(ul,{empty:"暂无 MCP 服务",items:i.mcpServers.map((r,h)=>{const f=r.detail||r.details||r.state||"";return c.jsxs("span",{className:"row-main",children:[r.name||r.id||""," | 状态 ",r.status||"",f?" | "+f:""]},h)})})]})}function $g(){const i=Gt(),[r,h]=d.useState("");return c.jsxs(Jt,{title:"MCP 操作",children:[c.jsxs(zt,{children:[c.jsx("button",{onClick:i.reloadMcp,children:"重载配置"}),c.jsx("span",{className:"muted",children:i.mcpMsg})]}),c.jsxs(zt,{children:[c.jsx("input",{type:"text",placeholder:"MCP 服务名",value:r,onChange:f=>h(f.target.value)}),c.jsx("button",{onClick:()=>void i.mcpOauth(r),children:"发起 OAuth 登录"}),c.jsx("span",{className:"muted",children:i.mcpOauthMsg})]}),i.mcpOauthBox?c.jsx("pre",{children:i.mcpOauthBox}):null]})}const Wh=160;function Ig(i){return i?i%86400===0?"（每天）":i%3600===0?"（每小时）":i%60===0?"（每 "+i/60+" 分钟）":"":""}function Pg({detail:i}){const r=Gt(),h=f=>{if(f){r.showDetail(f);try{window.location.hash="#view-tasks"}catch{}}};return c.jsxs("div",{className:"card",children:[c.jsxs("h2",{children:["自动化详情：",i.name]}),c.jsxs("div",{className:"muted",children:["状态 ",i.status,i.due?"｜到期":""," ｜ 每 ",i.interval_seconds," 秒 ｜ 下次 ",Ge(i.next_run_at),"｜ 上次执行 ",Ge(i.last_run_at)," ｜ 创建人 ",i.created_by?se(i.created_by):""," ｜ 创建于 ",Ge(i.created_at),i.cancelled_at?" ｜ 已取消于 "+Ge(i.cancelled_at):"",i.heartbeat_binding?" ｜ 心跳绑定：已绑定":""]}),c.jsx("div",{className:"muted",style:{whiteSpace:"pre-wrap",marginTop:8},children:i.prompt||""}),i.repo_url?c.jsxs("div",{className:"muted",children:["仓库 ",i.repo_url]}):null,c.jsxs(zt,{children:[c.jsx("button",{onClick:r.closeAutomationDetail,children:"关闭详情"}),c.jsxs("span",{className:"muted",children:["编辑请在下面列表里点“编辑”。",r.autoDetailMsg]})]}),c.jsxs("div",{className:"data-group",children:[c.jsxs("b",{children:["最近产出(",(i.runs||[]).length,")"]}),c.jsx(ul,{empty:"(无)",items:(i.runs||[]).map((f,g)=>c.jsxs("span",{className:"row-main",children:[f.status,f.created_at?"｜"+Ge(f.created_at):"",f.task_id?c.jsxs(c.Fragment,{children:[" ｜ 任务 ",c.jsx("button",{onClick:()=>h(f.task_id),children:se(f.task_id)})]}):"",f.detail?"｜"+String(f.detail).slice(0,120):""]},g))})]}),i.runs_note?c.jsx("div",{className:"muted",children:i.runs_note}):null]})}function tp(){const i=Gt(),[r,h]=d.useState(""),[f,g]=d.useState(""),[A,O]=d.useState(""),[b,N]=d.useState(""),[M,q]=d.useState(""),[p,H]=d.useState(""),[Z,F]=d.useState(""),[it,ft]=d.useState(""),[Et,st]=d.useState(""),[Ht,Q]=d.useState({}),ct=z=>{q(z.id),H(z.name||""),F(z.prompt||""),ft(z.interval_seconds?String(z.interval_seconds):""),st(z.repo_url||"")};return c.jsxs(c.Fragment,{children:[c.jsx("input",{placeholder:"自动化名称(同项目内启用中的名称唯一)",style:{width:"100%"},value:r,onChange:z=>h(z.target.value)}),c.jsx("textarea",{placeholder:"定时任务内容(创建只登记定义与到期状态；执行由后续执行器经 claim 抢单后建任务，不另起调度器)",value:f,onChange:z=>g(z.target.value)}),c.jsxs(zt,{children:[c.jsx("input",{placeholder:"间隔秒数(默认 86400，最小 60)",style:{width:"40%"},value:A,onChange:z=>O(z.target.value)}),c.jsx("input",{placeholder:"仓库(可选)",style:{width:"40%"},value:b,onChange:z=>N(z.target.value)}),c.jsx("button",{className:"primary",onClick:async()=>{await i.createAutomation(r,f,A,b),h(""),g(""),O(""),N("")},children:"创建自动化"}),c.jsx("button",{onClick:i.refreshAutomations,children:"刷新自动化"}),c.jsx("span",{className:"muted",children:i.autoMsg})]}),c.jsx(ul,{empty:"(暂无自动化)",items:i.automations.map(z=>{const L=z.prompt||"",at=!!Ht[z.id],jt=L.length>Wh,Mt=L?at||!jt?L:L.slice(0,Wh)+"…":"（无内容）";return c.jsxs("div",{children:[c.jsxs("div",{className:"row-main",children:["[",z.status,z.due?"｜到期":"",z.run_once?"｜一次性":"","] ",z.name,"｜每 ",z.interval_seconds," 秒",Ig(z.interval_seconds),"｜上次 ",Ge(z.last_run_at)||"未运行","｜下次 ",Ge(z.next_run_at),c.jsx("button",{onClick:()=>void i.loadAutomationDetail(z.id),children:"详情"}),z.status==="active"?c.jsxs(c.Fragment,{children:[" ",c.jsx("button",{onClick:()=>ct(z),children:"编辑"})," ",c.jsx("button",{onClick:()=>i.cancelAutomation(z.id),children:"取消"})]}):null]}),z.repo_url?c.jsxs("div",{className:"muted",children:["仓库：",z.repo_url]}):null,z.status==="cancelled"&&z.cancelled_at?c.jsxs("div",{className:"muted",children:["已取消于 ",Ge(z.cancelled_at)]}):null,c.jsxs("div",{className:"muted",style:{whiteSpace:"pre-wrap",overflowWrap:"anywhere"},children:[Mt,jt?c.jsx("button",{onClick:()=>Q(vt=>({...vt,[z.id]:!vt[z.id]})),children:at?"收起":"展开"}):null]}),M===z.id?c.jsxs("div",{children:[c.jsx("input",{placeholder:"自动化名称",style:{width:"100%"},value:p,onChange:vt=>H(vt.target.value)}),c.jsx("textarea",{placeholder:"定时任务内容",value:Z,onChange:vt=>F(vt.target.value)}),c.jsxs(zt,{children:[c.jsx("input",{placeholder:"间隔秒数(最小 60，留空保持原值)",style:{width:"40%"},value:it,onChange:vt=>ft(vt.target.value)}),c.jsx("input",{placeholder:"仓库(可选，清空则清除)",style:{width:"40%"},value:Et,onChange:vt=>st(vt.target.value)}),c.jsx("button",{className:"primary",onClick:async()=>{const vt=i.automations.find(B=>B.id===M)||z;await i.updateAutomation(vt,p,Z,it,Et)&&q("")},children:"保存"}),c.jsx("button",{onClick:()=>q(""),children:"取消"})]})]}):null]},z.id)})}),i.autoDetailMsg&&!i.autoDetail?c.jsx("div",{className:"muted",children:i.autoDetailMsg}):null,i.autoDetail?c.jsx(Pg,{detail:i.autoDetail},i.autoDetail.id):null]})}function ep(){var g,A,O,b;const i=Gt(),r=i.triage,h=N=>{if(N){i.showDetail(N);try{window.location.hash="#view-tasks"}catch{}}},f=N=>{if(N){i.loadAutomationDetail(N);try{window.location.hash="#view-automation"}catch{}}};return i.triageError?c.jsx("div",{className:"muted",children:i.triageError}):r?c.jsxs(c.Fragment,{children:[c.jsxs("div",{className:"data-group",children:[c.jsxs("b",{children:["到期自动化(",((g=r.counts)==null?void 0:g.automations_due)||0,")"]}),c.jsx(ul,{empty:"(无)",items:(r.automations_due||[]).map((N,M)=>c.jsx("span",{className:"row-main",children:N.id?c.jsxs("button",{onClick:()=>f(N.id),children:[N.name,"｜每 ",N.interval_seconds," 秒"]}):c.jsxs(c.Fragment,{children:[N.name,"｜每 ",N.interval_seconds," 秒"]})},M))})]}),c.jsxs("div",{className:"data-group",children:[c.jsxs("b",{children:["失败/中断任务(",((A=r.counts)==null?void 0:A.failed_tasks)||0,")"]}),c.jsx(ul,{empty:"(无)",items:(r.failed_tasks||[]).map((N,M)=>c.jsx("span",{className:"row-main",children:c.jsxs("button",{onClick:()=>h(N.id),children:[se(N.id),"｜",N.status,"｜",String(N.prompt||"").slice(0,80)]})},M))})]}),c.jsxs("div",{className:"data-group",children:[c.jsxs("b",{children:["待处理后续消息(",((O=r.counts)==null?void 0:O.pending_followups)||0,")"]}),c.jsx(ul,{empty:"(无)",items:(r.pending_followups||[]).map((N,M)=>c.jsx("span",{className:"row-main",children:c.jsxs("button",{onClick:()=>h(N.task_id),children:["任务 ",se(N.task_id),"｜",String(N.body||"").slice(0,80)]})},M))})]}),c.jsxs("div",{className:"data-group",children:[c.jsxs("b",{children:["最近自动化产出(",((b=r.counts)==null?void 0:b.recent_runs)||0,")"]}),c.jsx(ul,{empty:"(无)",items:(r.recent_runs||[]).map((N,M)=>c.jsxs("span",{className:"row-main",children:[se(N.automation_id),"｜",N.status,N.task_id?c.jsxs(c.Fragment,{children:["｜任务 ",c.jsx("button",{onClick:()=>h(N.task_id),children:se(N.task_id)})]}):"",N.task_status?"｜"+N.task_status:""]},M))})]})]}):c.jsx("div",{className:"muted",children:"(先选项目)"})}function lp(){return c.jsx(Jt,{id:"nav-automations",title:"自动化",children:c.jsx(tp,{})})}function ap(){const i=Gt();return c.jsxs(Jt,{id:"nav-triage",title:"分诊收件箱",children:[c.jsx(zt,{children:c.jsx("button",{onClick:i.refreshTriage,children:"刷新分诊"})}),c.jsx(ep,{})]})}const So=["pending","in_progress","blocked","completed","cancelled"],Bc={pending:"待办",in_progress:"进行中",blocked:"受阻",completed:"已完成",cancelled:"已取消"},np=["low","medium","high"],$h={low:"低",medium:"中",high:"高"};function up(i){return i?new Date(i*1e3).toLocaleString():""}function ip(){const i=Gt(),[r,h]=d.useState([]),[f,g]=d.useState({}),[A,O]=d.useState(0),[b,N]=d.useState(""),[M,q]=d.useState(""),[p,H]=d.useState(""),[Z,F]=d.useState(""),[it,ft]=d.useState(""),[Et,st]=d.useState(""),[Ht,Q]=d.useState("medium"),[ct,z]=d.useState(""),[L,at]=d.useState(""),jt=d.useCallback(async()=>{if(!i.projectId){h([]),g({}),O(0);return}try{let B=$.projectTodos(i.projectId);const dt=new URLSearchParams;p&&dt.set("status",p),Z.trim()&&dt.set("q",Z.trim());const Ct=dt.toString();Ct&&(B+="?"+Ct);const V=await I(B);h(V.todos||[]),g(V.counts||{}),O(V.total||0),q("")}catch(B){q("待办加载失败："+((B==null?void 0:B.message)||String(B)))}},[i.projectId,p,Z]);d.useEffect(()=>{jt()},[jt]);const Mt=async()=>{if(!i.projectId){N("先选项目");return}if(!it.trim()){N("标题必填");return}N("");try{const B={title:it.trim(),description:Et,priority:Ht};ct.trim()&&(B.task_id=ct.trim()),L&&(B.due_at=Math.floor(new Date(L).getTime()/1e3)),await I($.projectTodos(i.projectId),"POST",B),ft(""),st(""),z(""),at(""),N("已创建"),await jt()}catch(B){N("创建失败："+((B==null?void 0:B.message)||String(B)))}},vt=async(B,dt)=>{if(i.projectId)try{await I($.projectTodo(i.projectId,B.id),"PATCH",dt),await jt()}catch(Ct){q("待办更新失败："+((Ct==null?void 0:Ct.message)||String(Ct)))}},bt=async B=>{if(i.projectId&&window.confirm("删除待办“"+B.title+"”？删除后不可恢复（取消请改状态为已取消）。"))try{await I($.projectTodo(i.projectId,B.id),"DELETE"),await jt()}catch(dt){q("待办删除失败："+((dt==null?void 0:dt.message)||String(dt)))}};return c.jsxs(Jt,{id:"nav-todos",title:"项目待办",children:[c.jsx("div",{className:"muted",children:i.projectId?"共 "+A+" 项 ｜ "+So.map(B=>Bc[B]+" "+(f[B]||0)).join(" ｜ "):"先选项目"}),M?c.jsx("div",{className:"err",children:M}):null,c.jsxs(zt,{children:[c.jsxs("select",{"aria-label":"状态筛选",value:p,onChange:B=>H(B.target.value),children:[c.jsx("option",{value:"",children:"状态：全部"}),So.map(B=>c.jsx("option",{value:B,children:Bc[B]},B))]}),c.jsx("input",{type:"text",placeholder:"搜索标题/说明",value:Z,onChange:B=>F(B.target.value)}),c.jsx("button",{onClick:()=>void jt(),children:"刷新"})]}),c.jsx("input",{type:"text",placeholder:"待办标题（必填）",style:{width:"100%"},value:it,onChange:B=>ft(B.target.value)}),c.jsx("textarea",{placeholder:"说明（可选）",value:Et,onChange:B=>st(B.target.value)}),c.jsxs(zt,{children:[c.jsx("select",{"aria-label":"优先级",value:Ht,onChange:B=>Q(B.target.value),children:np.map(B=>c.jsxs("option",{value:B,children:["优先级：",$h[B]]},B))}),c.jsx("input",{type:"text",placeholder:"关联任务 id（可选）",value:ct,onChange:B=>z(B.target.value)}),c.jsx("input",{type:"datetime-local","aria-label":"截止时间",value:L,onChange:B=>at(B.target.value)}),c.jsx("button",{className:"primary",onClick:()=>void Mt(),children:"新建待办"}),c.jsx("span",{className:"muted",children:b})]}),c.jsx(ul,{empty:i.projectId?"（暂无待办）":"（先选项目）",items:r.map(B=>c.jsxs("span",{className:"row-main",children:[c.jsx("b",{children:B.title})," ｜ ",c.jsx("span",{className:"st-"+B.status,children:Bc[B.status]||B.status})," ｜ 优先级 "+($h[B.priority]||B.priority),B.task_id?" ｜ 任务 "+se(B.task_id):"",B.due_at?" ｜ 截止 "+up(B.due_at):"",B.completed_at?" ｜ 完成于 "+new Date(B.completed_at*1e3).toLocaleString():"",B.description?c.jsxs(c.Fragment,{children:[c.jsx("br",{}),c.jsx("span",{className:"muted",children:B.description})]}):null,c.jsx("br",{}),c.jsx("select",{"aria-label":"改状态",value:B.status,onChange:dt=>void vt(B,{status:dt.target.value}),children:So.map(dt=>c.jsx("option",{value:dt,children:Bc[dt]},dt))})," ",c.jsx("button",{onClick:()=>void vt(B,{status:"completed"}),children:"完成"})," ",c.jsx("button",{onClick:()=>void bt(B),children:"删除"})]},B.id))}),c.jsx("div",{className:"muted",children:"状态与 Muse 待办同款五态；完成时服务端记完成时间，改回其它状态时清掉；已取消是软关闭，删除才是硬删除。"})]})}const _o=[["tasks","任务工作台","任务"],["todos","项目待办","任务"],["resources","环境与 Workers","管理"],["automation","技能与自动化","管理"],["admin","系统与 Key","管理"]],cp=Object.fromEntries(_o.map(([i,r])=>[i,r]));function Ih(){const i=window.location.hash.replace("#view-","");return _o.some(([r])=>r===i)?i:"tasks"}function sp({view:i,setView:r}){return c.jsxs("aside",{className:"shell-side",children:[c.jsx("div",{className:"shell-brand",children:"Codex 命令台"}),c.jsx("nav",{className:"shell-nav","aria-label":"导航",children:["任务","管理"].map(h=>c.jsxs("div",{className:"shell-nav-group",children:[c.jsx("div",{className:"shell-nav-group-label",children:h}),_o.filter(([,,f])=>f===h).map(([f,g])=>c.jsx("button",{type:"button",className:i===f?"active":"",onClick:()=>r(f),children:g},f))]},h))}),c.jsx("div",{id:"nav-projects",className:"shell-project-nav",children:c.jsx(Lg,{})}),c.jsxs("div",{className:"shell-stats",children:[c.jsx("div",{className:"shell-stats-label",children:"概览"}),c.jsx(vg,{})]})]})}function rp(){const i=Gt();return c.jsx("span",{className:i.keySaved?"dash-status ok":"dash-status off",children:i.keySaved?"● 已连接":"○ 未连接"})}function op(){const i=hg();fg(b=>i.refreshAll(b),1e4),dg(()=>i.refreshSlow(),6e4);const[r,h]=d.useState(()=>Ih()),[f,g]=d.useState(!0),A=b=>{h(b);try{window.history.replaceState(null,"","#view-"+b)}catch{}};d.useEffect(()=>{const b=()=>h(Ih());return window.addEventListener("hashchange",b),()=>window.removeEventListener("hashchange",b)},[]);const O=()=>{A("tasks"),g(!0),setTimeout(()=>{var b;return(b=document.getElementById("sec-dispatch"))==null?void 0:b.scrollIntoView({behavior:"smooth"})},0)};return c.jsx(tv.Provider,{value:i,children:c.jsxs("div",{className:"shell",children:[c.jsx(sp,{view:r,setView:A}),c.jsxs("div",{className:"shell-main",children:[c.jsxs("header",{className:"shell-header",children:[c.jsx("h1",{children:cp[r]}),c.jsx(rp,{}),c.jsx("span",{className:"muted shell-hide-sm",children:i.healthText}),c.jsx("span",{className:"muted shell-hide-sm",children:i.quotaText}),c.jsxs("span",{className:"muted shell-stamp",children:["自动刷新 · 每 10 秒 · 已刷新 ",i.refreshStamp]})]}),r==="tasks"?c.jsxs("div",{className:"workspace",children:[c.jsxs("section",{className:"workspace-rail","aria-label":"任务列表",children:[c.jsx("h2",{children:"任务列表"}),c.jsx(bg,{onNew:O})]}),c.jsxs("section",{className:"workspace-center",children:[c.jsx("div",{className:"composer-toggle",children:c.jsx("button",{type:"button",onClick:()=>g(!f),children:f?"收起新建任务":"新建任务"})}),f?c.jsx("section",{id:"sec-dispatch",children:c.jsx(xg,{})}):null,c.jsxs("section",{className:"card workspace-conv",children:[c.jsx("h2",{children:"对话"}),c.jsx(Ag,{})]})]}),c.jsx("section",{className:"workspace-inspector","aria-label":"任务详情",children:c.jsx(Gg,{})})]}):c.jsx("main",{className:"page",children:r==="todos"?c.jsx(c.Fragment,{children:c.jsx(ip,{})}):r==="resources"?c.jsxs(c.Fragment,{children:[c.jsx(Zg,{}),c.jsx(Kg,{}),c.jsx(Jg,{}),c.jsx(Fg,{})]}):r==="automation"?c.jsxs(c.Fragment,{children:[c.jsx(Wg,{}),c.jsx($g,{}),c.jsx(lp,{}),c.jsx(ap,{})]}):c.jsxs(c.Fragment,{children:[c.jsx(Vg,{}),c.jsx(Qg,{})]})})]})]})})}Py.createRoot(document.getElementById("root")).render(c.jsx(Ly.StrictMode,{children:c.jsx(op,{})})); diff --git a/cloud/tests/test_chv_arch95.py b/cloud/tests/test_chv_arch95.py index f33d078..706b543 100644 --- a/cloud/tests/test_chv_arch95.py +++ b/cloud/tests/test_chv_arch95.py @@ -225,6 +225,9 @@ class DockerSyncGuardTest(unittest.TestCase): try: ex = DockerExecutor(runner=lambda cmd, timeout=None: FakeResult(returncode, stdout, stderr)) + # to10：goal 解析是纯函数测试，绝不依赖本机 docker + # 二进制（无 docker 的机器上 _check_docker 会直接炸）。 + ex._check_docker = lambda: None ex._call_runner = lambda cmd, timeout=None, env=None: \ FakeResult(returncode, stdout, stderr) return ex.goal_task({"id": "t1", "goal": "do it", diff --git a/cloud/tests/test_chv_to10.py b/cloud/tests/test_chv_to10.py new file mode 100644 index 0000000..2c44806 --- /dev/null +++ b/cloud/tests/test_chv_to10.py @@ -0,0 +1,388 @@ +"""to10 轨新增测试：覆盖率缺口分支（非刷行）。 + +只测此前 coverage 报告里未覆盖的分支：api_client 的假 +socket 错误路径、hostkeys 的 keygen/seed 失败路径、 +orphans 的 registry 失败路径、keyscan 过滤、compat 的 +fail-closed 分支、config 的类型转换与 assemble 冲突。 +不删旧测试、不改分母；每项都是行为断言，不是 grep。 +""" + +import os +import socket +import tempfile +import unittest + +from cloud.chv import api_client +from cloud.chv import compat215 +from cloud.chv import config as chv_config +from cloud.chv import hostkeys +from cloud.chv import orphans +from cloud.chv import ssh_transport + + +class _FakeSocket: + """按 recv 调用序列供数据的假 socket。""" + + def __init__(self, chunks, timeout_at=None): + self._chunks = list(chunks) + self._timeout_at = timeout_at + self._calls = 0 + self.closed = False + + def recv(self, _size): + self._calls += 1 + if self._timeout_at is not None \ + and self._calls >= self._timeout_at: + raise socket.timeout("fake timeout") + if not self._chunks: + return b"" + return self._chunks.pop(0) + + def close(self): + self.closed = True + + +class ApiClientTest(unittest.TestCase): + def test_header_timeout_raises(self): + with self.assertRaises(RuntimeError): + api_client._read_header(_FakeSocket([], timeout_at=1)) + + def test_header_early_close_raises(self): + with self.assertRaises(RuntimeError): + api_client._read_header(_FakeSocket([])) + + def test_chunked_rejected(self): + with self.assertRaises(RuntimeError): + api_client._parse_content_length( + "HTTP/1.1 200 OK\r\nTransfer-Encoding: chunked") + + def test_no_length_rejected(self): + with self.assertRaises(RuntimeError): + api_client._parse_content_length("HTTP/1.1 200 OK") + + def test_illegal_length_rejected(self): + with self.assertRaises(RuntimeError): + api_client._parse_content_length( + "HTTP/1.1 200 OK\r\nContent-Length: xx") + + def test_body_timeout_truncation(self): + with self.assertRaises(RuntimeError): + api_client._read_body(_FakeSocket([], timeout_at=1), + b"ab", 10) + + def test_body_early_close_truncation(self): + with self.assertRaises(RuntimeError): + api_client._read_body(_FakeSocket([]), b"ab", 10) + + def test_parse_non_http_status_zero(self): + status, _body = api_client._parse_response("garbage", b"{}") + self.assertEqual(status, 0) + + def test_parse_non_json_body_is_text(self): + _status, body = api_client._parse_response( + "HTTP/1.1 200 OK", b"not-json") + self.assertEqual(body, "not-json") + + def test_parse_empty_body_is_empty_dict(self): + _status, body = api_client._parse_response( + "HTTP/1.1 200 OK", b"") + self.assertEqual(body, {}) + + def test_unix_api_missing_socket_raises(self): + with self.assertRaises(RuntimeError): + api_client._unix_socket_api("/tmp/no-such-chv.sock", + "GET", "/vm.info") + + +class HostkeysTest(unittest.TestCase): + def test_keygen_reuse_existing_pair(self): + with tempfile.TemporaryDirectory() as tmp: + paths = hostkeys.host_key_paths(tmp, "vm1") + os.makedirs(os.path.dirname(paths["private"]), + exist_ok=True) + for key in ("private", "public"): + with open(paths[key], "w") as fh: + fh.write("x") + out = hostkeys.generate_host_keypair( + tmp, "vm1", runner=lambda cmd: (_ for _ in ()).throw( + AssertionError("reuse 不应调 keygen"))) + self.assertEqual(out, paths) + + def test_keygen_nonzero_rc_raises(self): + class _Res: + returncode = 1 + with tempfile.TemporaryDirectory() as tmp: + with self.assertRaises(RuntimeError): + hostkeys.generate_host_keypair( + tmp, "vm2", runner=lambda cmd: _Res()) + + def test_keygen_no_files_after_runner_raises(self): + class _Res: + returncode = 0 + with tempfile.TemporaryDirectory() as tmp: + with self.assertRaises(RuntimeError): + hostkeys.generate_host_keypair( + tmp, "vm3", runner=lambda cmd: _Res()) + + def test_seed_disk_nonzero_rc_raises(self): + class _Res: + returncode = 1 + with tempfile.TemporaryDirectory() as tmp: + with self.assertRaises(RuntimeError): + hostkeys.build_seed_disk( + tmp, os.path.join(tmp, "seed.iso"), + runner=lambda cmd: _Res()) + + def test_seed_disk_missing_output_raises(self): + class _Res: + returncode = 0 + with tempfile.TemporaryDirectory() as tmp: + with self.assertRaises(RuntimeError): + hostkeys.build_seed_disk( + tmp, os.path.join(tmp, "seed.iso"), + runner=lambda cmd: _Res()) + + +class OrphansAndKeyscanTest(unittest.TestCase): + def test_quarantine_failure_only_warns(self): + with self.assertLogs("cloud.cloud_hypervisor", + level="WARNING") as logs: + orphans._quarantine_registry("/tmp/no-such-dir-x/f.json") + self.assertTrue(any("quarantine_failed" in line + for line in logs.output)) + + def test_write_registry_failure_only_warns(self): + with tempfile.TemporaryDirectory() as tmp: + blocker = os.path.join(tmp, "blocker") + with open(blocker, "w") as fh: + fh.write("x") + with self.assertLogs("cloud.cloud_hypervisor", + level="WARNING") as logs: + orphans._write_registry( + os.path.join(blocker, "registry.json"), {}) + self.assertTrue(any("save_failed" in line + for line in logs.output)) + + def test_keyscan_filters_short_and_non_ssh_lines(self): + text = "\n".join([ + "# comment", + "", + "only-two", + "127.0.0.1 ecdsa-sha2-nistp256 AAAA", + "[127.0.0.1]:2222 ssh-ed25519 AAAA-good", + "other-host ssh-ed25519 AAAA-other", + ]) + kept = ssh_transport.parse_keyscan_output(text) + self.assertEqual(kept, ["[127.0.0.1]:2222 ssh-ed25519 AAAA-good"]) + + +class CompatAndConfigTest(unittest.TestCase): + def test_tmpfs_unknown_executor_fail_closed(self): + isolated, _reason = compat215.tmpfs_isolated("unknown-kind", + "/tmp") + self.assertFalse(isolated) + + def test_resolve_inside_empty_rel_is_base(self): + self.assertEqual(compat215.resolve_inside("/a/b", ""), "/a/b") + + def test_resolve_inside_absolute_is_none(self): + self.assertIsNone(compat215.resolve_inside("/a/b", "/etc/passwd")) + + def test_to_int_bool_and_float_and_bad(self): + with self.assertRaises(ValueError): + chv_config._to_int(True) + self.assertEqual(chv_config._to_int(3.9), 3) + self.assertEqual(chv_config._to_int(" 7 "), 7) + with self.assertRaises(ValueError): + chv_config._to_int(object()) + + def test_to_float_bool_and_bad(self): + with self.assertRaises(ValueError): + chv_config._to_float(False) + self.assertEqual(chv_config._to_float("2.5"), 2.5) + with self.assertRaises(ValueError): + chv_config._to_float(object()) + + def test_assemble_conflict_raises(self): + with self.assertRaises(ValueError): + chv_config._assemble_chv_config( + [{"cpus": "2"}, {"cpus": "4"}]) + + +if __name__ == "__main__": + unittest.main() + + +class WorkerAdapterTest(unittest.TestCase): + """to10：worker_adapter 的直接行为测试（worker 薄包装 + 由旧测试 test_chv_fix/test_chv_gaps 覆盖，此处测本体）。""" + + def test_call_breakpoint_prefers_signature_kwargs(self): + from cloud.chv import worker_adapter + seen = {} + + class _Ex: + def goal_resume_breakpoint_task(self, task, + model_provider=None, + extra_env=None): + seen.update(task=task, model_provider=model_provider, + extra_env=extra_env) + return "ok" + out = worker_adapter.call_goal_resume_breakpoint( + _Ex(), {"id": "t"}, model_provider="m", + extra_env={"A": "1"}) + self.assertEqual(out, "ok") + self.assertEqual(seen["model_provider"], "m") + self.assertEqual(seen["extra_env"], {"A": "1"}) + + def test_call_breakpoint_single_arg_executor(self): + from cloud.chv import worker_adapter + + class _Ex: + def goal_resume_breakpoint_task(self, task): + return "single" + self.assertEqual( + worker_adapter.call_goal_resume_breakpoint(_Ex(), {"id": "t"}), + "single") + + def test_audit_callback_extracts_project_and_swallows(self): + from cloud.chv import worker_adapter + calls = [] + cb = worker_adapter.make_chv_audit_callback(calls.append) + # append 只接受 1 参，回调内部 except 应吞掉，不炸 + cb("booted", {"project_id": "p1"}, "detail") + self.assertEqual(calls, []) + got = [] + cb2 = worker_adapter.make_chv_audit_callback( + lambda pid, ev, detail: got.append((pid, ev, detail))) + cb2("booted", {"project_id": "p1"}, "detail") + cb2("booted", "not-a-dict", "detail") + self.assertEqual(got[0], ("p1", "chv_vm_booted", "detail")) + self.assertEqual(got[1], (None, "chv_vm_booted", "detail")) + + def test_reconcile_non_chv_returns_empty(self): + from cloud.chv import worker_adapter + self.assertEqual( + worker_adapter.reconcile_warm_pool(object(), lambda *a: None, + "startup"), + []) + + def test_watchdog_vm_name_error_returns_none(self): + from cloud.chv import worker_adapter + + class _Bad: + def vm_name(self, task): + raise RuntimeError("boom") + self.assertIsNone(worker_adapter.watchdog_vm_name(_Bad(), {})) + + def test_cancel_watchdog_loop_kills_on_flag(self): + import threading + from cloud.chv import worker_adapter + killed = [] + stop = threading.Event() + worker_adapter.cancel_watchdog_loop( + "t1", "chv-t1", stop, {"t1"}, + lambda tid: None, + lambda vm: killed.append(vm), + None) + # stop 未置位时循环每 1 秒轮询：flag 已命中，第一轮即 + # 关 VM 并返回；这里不依赖日志（logger 传 None 时 + # log_event 内部不炸——若炸则本项失败）。 + self.assertEqual(killed, ["chv-t1"]) + + def test_quota_watchdog_loop_marks_and_kills(self): + import threading + from cloud.chv import worker_adapter + killed = [] + marked = [] + stop = threading.Event() + worker_adapter.quota_watchdog_loop( + "t2", "chv-t2", stop, lambda: True, + marked.append, + lambda vm: killed.append(vm), + None, interval=0.01) + self.assertEqual(marked, ["t2"]) + self.assertEqual(killed, ["chv-t2"]) + + +class AuditSpillTest(unittest.TestCase): + """to10：_emit 回调失败时落盘（ADR-14），不中断、不抛。""" + + def test_spill_on_callback_failure(self): + import json as _json + from cloud.chv.executor import CloudHypervisorExecutor + with tempfile.TemporaryDirectory() as tmp: + ex = CloudHypervisorExecutor( + binary="/bin/true", kernel="/tmp/k", rootfs="/tmp/r", + state_dir=tmp, ssh_key="/tmp/key") + calls = [] + + def _bad(event, task, detail, kind=None): + raise RuntimeError("db down") + ex.audit_callback = _bad + ex._emit("booted", {"id": "t-spill"}, "vm=chv-t-spill") + self.assertEqual(ex.audit_failure_count, 1) + spill = os.path.join(tmp, "chv-audit-spill.jsonl") + self.assertTrue(os.path.exists(spill)) + row = _json.loads(open(spill).readline()) + self.assertEqual(row["event"], "booted") + self.assertEqual(row["task_id"], "t-spill") + self.assertEqual(row["detail"], "vm=chv-t-spill") + + def test_spill_rotation_keeps_old(self): + from cloud.chv.executor import CloudHypervisorExecutor + with tempfile.TemporaryDirectory() as tmp: + ex = CloudHypervisorExecutor( + binary="/bin/true", kernel="/tmp/k", rootfs="/tmp/r", + state_dir=tmp, ssh_key="/tmp/key") + spill = os.path.join(tmp, "chv-audit-spill.jsonl") + with open(spill, "w") as fh: + fh.write("x" * (1024 * 1024 + 10)) + ex._spill_audit("booted", {"id": "t2"}, "d", None) + self.assertTrue(os.path.exists(spill + ".old")) + self.assertTrue(os.path.getsize(spill) < 1024) + + +class LockfileScriptTest(unittest.TestCase): + """to10：generate_lockfile.py 的 generate/verify（合成 + BUILDINFO，不碰真实客机清单，不编版本号）。""" + + def _load_script(self): + import importlib.util + path = os.path.join( + os.path.dirname(os.path.dirname(__file__)), + "..", "chv-guest", "generate_lockfile.py") + path = os.path.normpath(path) + spec = importlib.util.spec_from_file_location( + "chv_generate_lockfile", path) + mod = importlib.util.module_from_spec(spec) + spec.loader.exec_module(mod) + return mod + + def test_generate_reads_real_dpkg_version(self): + import json as _json + mod = self._load_script() + with tempfile.TemporaryDirectory() as tmp: + bi = os.path.join(tmp, "bi.json") + out = os.path.join(tmp, "lock.json") + with open(bi, "w") as fh: + _json.dump({"tag": "t", "contents": ["coreutils"]}, + fh) + self.assertEqual(mod.generate(bi, out), 0) + lock = _json.load(open(out)) + self.assertEqual(lock["tag"], "t") + self.assertEqual(lock["unresolved"], []) + self.assertTrue(lock["versions"]["dpkg"][0].startswith( + "coreutils ")) + + def test_verify_fails_on_unresolved(self): + import json as _json + mod = self._load_script() + with tempfile.TemporaryDirectory() as tmp: + lock = os.path.join(tmp, "lock.json") + with open(lock, "w") as fh: + _json.dump({"tag": "t", + "versions": {"dpkg": ["a 1.0"], + "pip": ["b 2.0"]}, + "unresolved": ["zzz"]}, fh) + self.assertEqual(mod.verify(lock, None), 1) diff --git a/cloud/worker.py b/cloud/worker.py index be3ff87..9ea598d 100644 --- a/cloud/worker.py +++ b/cloud/worker.py @@ -133,7 +133,6 @@ interrupted -> interrupted，is_defect 恒 False）；同一派发内 import argparse import hashlib -import inspect import json import os from datetime import datetime, timezone @@ -250,6 +249,9 @@ from cloud.queue import budget_exceeded, ready_tasks from cloud.repos import RepoError, checkout_dir, resolve_repo from cloud.sandbox import CloudExecutor, CloudHypervisorExecutor, \ DockerExecutor, LocalExecutor +# to10：CHV 强相关接线抽到 cloud.chv.worker_adapter，见 ADR-13； +# worker 只留薄包装（方法名/签名/调用点不变）。 +from cloud.chv import worker_adapter as _chv_worker_adapter from cloud.uploads import upload_root as _upload_root from cloud.sandbox import _default_runner as _sandbox_default_runner from cloud.caps import effective_capacity, resolve_capacity @@ -428,27 +430,13 @@ def _merge_env(base, override): def _call_goal_resume_breakpoint(executor, task, model_provider=None, extra_env=None): - """通用断点调用：执行器提供 goal_resume_breakpoint_task 时调用它。 - - Local 仍走原路径（调用方直接传 model_provider，不经过这里， - 以免改生产语义）；这里只用于非 Local 执行器：签名接受 - model_provider 时传它（CHV/Docker 均接受），否则只传 task - （旧假执行器单参）。feat-chv-secfix（H2）：CHV 的 - gateway_env 不再经 SSH env 前缀注入（密钥值进 argv 的 - 旧路径已废除），改走 provisioned seed 盘客内文件投递； - CHV 没有 Local 那样的进程 env 通道，seed 文件是唯一通道， - keyscan 模式下密钥投递 fail-closed。""" - method = getattr(executor, "goal_resume_breakpoint_task") - try: - params = inspect.signature(method).parameters - except (TypeError, ValueError): - return method(task) - kwargs = {} - if "model_provider" in params: - kwargs["model_provider"] = model_provider - if "extra_env" in params and extra_env is not None: - kwargs["extra_env"] = extra_env - return method(task, **kwargs) + """通用断点调用（to10：本体已移到 + cloud.chv.worker_adapter.call_goal_resume_breakpoint， + 此处只留薄包装——旧测试从 cloud.worker import 它，名 + 字与签名不变，行为逐字不变）。""" + return _chv_worker_adapter.call_goal_resume_breakpoint( + executor, task, model_provider=model_provider, + extra_env=extra_env) def _resolve_model(model=None, config_path=None): @@ -790,16 +778,12 @@ class Worker: # 读审计表即可，不另起一套 VM 事件表；成本/配额仍走 # worker 现有 quota 语义，本回调绝不记账。 if isinstance(self.executor, CloudHypervisorExecutor): - def _chv_audit(event, task, detail, kind=None): - try: - project_id = task.get("project_id") \ - if isinstance(task, dict) else None - self._audit(project_id, "chv_vm_" + event, - detail or "") - except Exception: - pass + # to10：回调本体已移到 worker_adapter.make_chv_audit_ + # callback（原 __init__ 闭包逐字搬运），此处只挂接。 try: - self.executor.audit_callback = _chv_audit + self.executor.audit_callback = \ + _chv_worker_adapter.make_chv_audit_callback( + self._audit) except Exception: pass # CHV 残留清理：仅孤儿清理接线（feat-chv-arch95 定案， @@ -1073,36 +1057,11 @@ class Worker: return self._cost_store def _reconcile_warm_pool(self, when): - """显式调生产残留清理 facade，返回已清理 VM 名。 - - 仅 CHV worker 有效（其它执行器返回 []）。实现是 - cloud.chv.orphans.reconcile_orphans_for_executor()： - 直接读 registry 文件做清理，不构造 WarmPool、不碰 - 实验接口（方法内 import，失败时按下面分支审计后 - 返回 []）。调用失败不炸主流程，但进审计（失败本身 - 也审计，不静默丢弃）。orphan 条目绝不发给任务， - reconcile 只做清理。""" - executor = getattr(self, "executor", None) - if not isinstance(executor, CloudHypervisorExecutor): - return [] - try: - from cloud.chv.orphans import \ - reconcile_orphans_for_executor as _reconcile - cleaned = _reconcile(executor) or [] - except Exception as exc: - try: - self._audit(None, "chv_warm_pool_reconcile_failed", - "when=%s error=%s" % (when, exc)) - except Exception: - pass - return [] - if cleaned: - try: - self._audit(None, "chv_warm_pool_reconcile", - "when=%s cleaned=%s" % (when, ",".join(cleaned))) - except Exception: - pass - return cleaned + """显式调生产残留清理 facade，返回已清理 VM 名 + （to10：本体已移到 worker_adapter.reconcile_warm_ + pool，此处只留薄包装，旧测试调用点不变）。""" + return _chv_worker_adapter.reconcile_warm_pool( + getattr(self, "executor", None), self._audit, when) def close(self): """关闭惰性建的成本存储连接（注入的 cost_store 由调用方管）。 @@ -3617,61 +3576,22 @@ class Worker: return cleaned def _chv_cancel_watchdog(self, task_id, vm_name, stop_event): - """CHV 同步路径的取消看门狗（feat-chv-fix，P1-2，Docker - 取消看门狗同款模式）：同步 SSH 阻塞时 worker 的 cancel() - 只能改库、清 _running 项，阻塞中的 run 本身不会醒； - 本线程每 1 秒查库/取消标志，命中时调 cancel_vm 真关 - VM，然后退出。查库失败时当没取消，不误杀。""" - while not stop_event.wait(1.0): - try: - if task_id in self._cancel_requested: - cancelled = True - else: - db_task = self.get_task(task_id) - cancelled = (db_task is not None and - db_task.get("status") == "interrupted") - except Exception: - cancelled = False - if not cancelled: - continue - try: - cancel = getattr(self.executor, "cancel_vm", None) - if cancel is not None and vm_name: - cancel(vm_name) - log_event(self.logger, "worker_chv_killed", - task_id=task_id, - message="cancel watchdog killed VM %s " - "(task interrupted/cancelled)" - % vm_name) - except Exception: - pass - return + """CHV 同步路径的取消看门狗（to10：循环本体已移到 + worker_adapter.cancel_watchdog_loop，此处只传回调）。""" + _chv_worker_adapter.cancel_watchdog_loop( + task_id, vm_name, stop_event, self._cancel_requested, + self.get_task, getattr(self.executor, "cancel_vm", None), + self.logger) def _chv_quota_watchdog(self, task_id, vm_name, stop_event, quota_check, interval=30.0): - """CHV 同步路径的配额看门狗（feat-chv-fix，P1-2，Docker - 配额看门狗同款）：命中时记 _quota_hit_tasks 并调 - cancel_vm 真关 VM；_run_attempt 在同步返回后把结果 - 改写成 quota_exceeded，走统一配额中断收尾。""" - while not stop_event.wait(interval): - try: - over = bool(quota_check()) if quota_check else False - except Exception: - over = False - if not over: - continue - self._mark_quota_hit(task_id) - try: - cancel = getattr(self.executor, "cancel_vm", None) - if cancel is not None and vm_name: - cancel(vm_name) - log_event(self.logger, "worker_chv_killed", - task_id=task_id, - message="quota watchdog killed VM %s " - "(quota exceeded)" % vm_name) - except Exception: - pass - return + """CHV 同步路径的配额看门狗（to10：循环本体已移到 + worker_adapter.quota_watchdog_loop，此处只传回调）。""" + _chv_worker_adapter.quota_watchdog_loop( + task_id, vm_name, stop_event, quota_check, + self._mark_quota_hit, + getattr(self.executor, "cancel_vm", None), + self.logger, interval=interval) def cleanup_stale_containers(self): """启动清理兜底：清掉上一轮残留的 codex- 容器。 @@ -5976,10 +5896,10 @@ class Worker: # process 已轮询 quota_check）。 if isinstance(self.executor, CloudHypervisorExecutor) \ and hasattr(self.executor, "cancel_vm"): - try: - _chv_vm_name = self.executor.vm_name(task) - except Exception: - _chv_vm_name = None + # to10：VM 名派生已移到 worker_adapter.watchdog_vm_ + # name（原 try/except 块逐字搬运）。 + _chv_vm_name = _chv_worker_adapter.watchdog_vm_name( + self.executor, task) if _chv_vm_name: _watchdog_stop = threading.Event() _watchdog_thread = threading.Thread( diff --git a/docs/chv-architecture.md b/docs/chv-architecture.md index 28e81c3..ed06307 100644 --- a/docs/chv-architecture.md +++ b/docs/chv-architecture.md @@ -232,3 +232,94 @@ live 状态——server 进程没有 VM 登记表，live 状态编不 出来；前端无该字段时不显示 VM 状态，不猜测。VM 控制 按钮仍不做（后端无控制端点且安全审计 No-Go）。 + +## ADR-13 CHV worker 适配模块（to10 轨） + +- 决策：新建 `cloud/chv/worker_adapter.py`，worker.py 只留 + 薄包装。抽入：断点调用（原顶层函数本体）、审计回调 + （原 `__init__` 闭包）、孤儿清理（原 `_reconcile_warm_pool` + 本体）、两条看门狗循环（原 `_chv_*_watchdog` 本体，self + 依赖全部改成回调参数）、看门狗 VM 名派生（原 `_run_attempt` + try/except 块）。worker 方法名、签名、调用点、错误文案 + 全部不变，旧测试继续走 worker 方法；适配器本身另有直接 + 行为测试。适配器只依赖标准库 + chv 叶子模块 + + `cloud.logging_config`，不 import `cloud.worker`，无循环； + 公共 API 全注解（#257 的类型测试扫 `cloud/chv/*.py`， + 新模块不注解会直接炸该测试——本轨已补齐）。 +- 明确不抽（硬抽无收益，只会撕通用流程）：cancel() 里的 + `cancel_vm` 分支（与 Docker `kill_container` 分支相邻的 + 通用取消流）、`cleanup_stale_containers` 的 CHV 分支 + （与 Docker 清理相邻的通用清理流）、网关守卫（Local / + Docker / CHV 三分支共用的派发流）、`_executor_name` 的 + CHV 分支（通用执行器名映射）、`_run_attempt` 的看门狗 + 线程启动（线程管理是 worker 通用逻辑，只抽了 VM 名派生）、 + VM 死亡写回（通用终局流）。以上六处留在 worker.py， + 不是遗漏，是评估后有理由的结论。 +- 附带清理：`inspect` 在 worker.py 的唯一用处随断点函数 + 移走，import 一并删除（留着会触发 F401）；worker.py + 基线已有的 I001（两处 import 排序）与 F821（playwright + 播种块的 `task`）与本轨无关，不顺手重排整个文件。 + +## ADR-14 审计失败落盘兜底（to10 轨） + +- 决策：选“本地审计落盘文件”作为兜底，不选“只记日志 + + 计数”（已做，仍不够：日志会轮转、计数只剩数字，事后 + 找不回哪次事件丢了）。实现：`executor._emit` 回调失败 + 时，除记宿主 error 日志 + `audit_failure_count` +1 外， + 再把事件 append 到 `state_dir/chv-audit-spill.jsonl` + （每行一 JSON：ts/event/kind/detail/task_id），文件超 + 1MB 时轮转成 `.old` 再写。 +- 取舍（写清，不当成完整修复）：① 落盘本身 best-effort， + 写失败只记日志，绝不中断任务；② 落盘不是 DB 替代， + 任务 timeline 查询看不到它，需宿主人工查阅；③ 轮转 + 只留最新两代，更早的丢失事件仍会丢；④ detail 沿用 + `_emit` 契约（只含 VM 名/路径/URL/kind），绝不含密钥值， + 落盘文件随 state_dir 的 0700 权限保护。审计完整性本身 + 仍无保证——保证需要 DB 事务级审计写入，那是 worker + 全局语义的改动，不在本轨硬做。 +- 不选的另一条路：“审计失败即中断任务”——ADR-9 已否决： + 审计是可观测性不是门禁，DB 短暂不可用时硬中断会把 + 一次审计失败放大成任务失败，且与 worker 全局 best-effort + 语义冲突。 + +## ADR-15 前端只读 VM 展示四态与可用性统一文案（to10 轨） + +- 决策：`chvVmDisplay(task)` 返回四态而非此前两态文案： + dead（字段 present 且 true）、not-reported（CHV 任务但 + 字段 absent，只表示未报告）、unknown（任务详情本身未 + 加载）、not-applicable（非 CHV，不显示行）。旧入口 + `chvVmStatusText` 保留为 display 的 text 包装（unknown / + not-applicable 回空串），已测调用点不变。详情页 VM 行 + 改用 `chvVmDisplay(t).text` 并加 `role="status"` 与 + `aria-label="VM 状态"`。 +- 决策：`chvAvailabilityText(availability)` 统一新建任务 / + Workers / 环境三处可用性文案。此前三处各写各的，Workers + 的未知分支恒写“仅全局管理员可见”（加载失败/未加载时 + 也是这句，是文案 bug），本轨统一后按 reason 分三句， + known 零在线沿用派发页的完整版（不拦提交、后端是最终 + 权威）。无障碍打磨：可用性区与提交结果区加 + `aria-live="polite"`（结果区另加 `role="status"`），附件 + 三个 input 加 `aria-label`（此前只有 title），禁用态本身 + 已由 `disabled` 传达，不另造禁用文案。控制按钮仍不做 + （后端无控制端点且安全审计 No-Go）；组件渲染仍不在 + 测试覆盖内（无 testing-library 依赖，不临时加依赖冒充）。 + +## ADR-16 to10 轨的类型与 lint 口径（诚实标注） + +- mypy：`--follow-imports=silent cloud/chv/`，只查 chv 本身 + （不拖入无关文件）。基线该口径 115 错 → 本轨 0 错； + 默认 follow-imports 口径基线 117 错 → 本轨 2 错，且 + 2 错全在作用域外的 `cloud/marker_tools.py:142`（与 chv + 无关，不顺手修它），chv 自身在默认口径下也是 0 错。 +- ruff：作用域 F/E9/W/I（未用变量、语法错、空白、import + 排序），基线 8 错 → 本轨 0 错；默认全规则基线 208 错 + （含本仓库未采纳的风格意见，如百分号格式化、 + best-effort 清理的 try/except-pass），本轨不声称全规则 + 通过。worker.py 不在 lint 作用域内：其基线 I001 两处与 + F821 一处与本轨无关，未重排；本轨只删了随抽取而变多余 + 的 `inspect` import（否则新增 F401）。 +- 覆盖率：`coverage --source=cloud/chv` 对 CHV 套件（不含 + worker/server 全量），基线 85% → 本轨 87%（1826 行， + 缺 268 → 242）；未覆盖大头仍是 executor 深层生命周期 + 分支（需假 VMM 全链路）与 disk 配额分支，不靠删测试 + 或改分母凑数。 diff --git a/docs/chv-production-readiness.md b/docs/chv-production-readiness.md index 122afa5..6985ea2 100644 --- a/docs/chv-production-readiness.md +++ b/docs/chv-production-readiness.md @@ -459,3 +459,116 @@ 本 PR 顺序合并 → KVM 宿主 + 防火墙清单 → 镜像构建发布 → §11.5 + 本节真机清单全过 → 默认关闭下单租户/受信项目 灰度 → 全量多租户（须另行决策，不在本轨内默认开启）。 + +## 13. to10 提升轨 feat-chv-to10（堆叠于 #257 之上） + +本节是准备工作，不是启用结论：代码侧的 to10 项已做， +生产就绪与动态安全仍未到 10 分（见 §13.4 外部关卡）。 + +### 13.1 合并前 diff 复核清单（以当前 main 头预填） + +- 当前 main 头（本轨核查时）：`aebc6238a613066316ec18796b83ea8fc2d8835e` + （Merge PR #201）。链上 10 个 PR 在核查时全部 + OPEN / MERGEABLE（#216 base main，#221→#257 依次堆叠， + 本 PR base 为 feat-chv-arch-maint-fe）。合并时必须重新 + 核查：main 头若已前进，整条链须逐个 rebase 并重跑测试， + 绝不直接合并旧基底。 +- 逐项复核点（合并执行者逐项打勾，任一项不符即停）： + 1. server 任务创建白名单含 `cloud-hypervisor`，环境创建 + 白名单为 local/docker/cloud-hypervisor（不含 cloud）， + 两处白名单差异仍在前端文案里写清。 + 2. worker 断点分支：Local 原路径不变，CHV 经 + `worker_adapter.call_goal_resume_breakpoint`（薄包装名 + `_call_goal_resume_breakpoint` 仍在 worker.py，旧测试 + 导入点不变）。 + 3. sandbox re-export：`cloud/chv` 不再 export WarmPool + （#257 已移实验模块），`cloud/sandbox.py` 的 CHV + re-export 与 `cloud/chv/__init__.py` 一致。 + 4. 前端产物：`cloud/console.html` 与 `dist/index.html` + 的 sha256 一致（本轨：`269820d1…f389`，353163 字节）， + 产物内含新文案（“无法判断 VM 状态”1 处、 + “CHV worker 在线数未知”3 处）；合并时重建后复核， + 不直接沿用本轨哈希。 + 5. `package.json` 的 vitest 测试脚本与 lockfile 增量只含 + vitest 新条目，无旧包版本变动。 + 6. 本轨新增文件：`cloud/chv/worker_adapter.py`、 + `cloud/tests/test_chv_to10.py`、 + `chv-guest/generate_lockfile.py`；本轨改动文件： + `cloud/worker.py`（薄包装 + 删多余 `inspect` import）、 + `cloud/chv/executor.py`（审计落盘 + 两函数抽 helper）、 + `cloud/chv/config.py`、`disk.py`、`hardening.py`、 + `hostkeys.py`、`ssh_transport.py`、`experimental_pool.py`、 + `compat215.py`（类型注解 + ruff 整理）、前端 6 文件。 + 7. CI：链上全部 PR 的 `verify` 仍是账单故障秒级失败模式 + （本轨核查时 #257 等均为 UNSTABLE，非代码失败）； + CI 变绿前不合并，绝不改 CI 配置绕过。 + 8. 旧链 #199→#205→#207 与 #209–#215 仍全部 OPEN： + 合并前复核它们与本链的交集（server 白名单、worker + 断点分支、sandbox re-export），有冲突先解决冲突方， + 不静默覆盖。 + +### 13.2 真机验证清单（用例与验收标准细化） + +以下用例全部待 KVM 宿主，验收标准是“通过才算”，不是 +“跑过就算”；任一用例未跑即记为未验证，不写进已验。 + +- V-1 冷启动：CHV worker 在 KVM 宿主上起 VM → guest + SSH 就绪（钉死 known_hosts，provisioned 模式）→ 跑通 + 一个真实任务（host.py 在客机内执行）→ 任务完成且 VM + 已销毁、任务盘已删。验收：任务状态 completed，宿主 + state_dir 无该任务残留盘与 seed。 +- V-2 seed 首启时序：provisioned 模式下客机 cloud-init + 在 sshd 启动前应用 seed（host key 不再生、客内密钥文件 + 0600）。验收：`sshd -T` 显示 PasswordAuthentication no + 等加固项，known_hosts 钉死校验通过，非 TOFU 回落。 +- V-3 快照恢复：任务中途 hibernate（pause + snapshot + + shutdown）→ restore 后显式 resume → 断点续跑完成。 + 验收：恢复后 known_hosts 不变（不重预置），快照文件 + 落盘非空且经 `snapshot_not_persisted` 校验。 +- V-4 网关真实任务：sub2api 网关任务在 CHV 上跑通， + 密钥只走 seed 客内文件，宿主 `ps` / `/proc/*/cmdline` + 里看不到密钥值。验收：任务完成 + 宿主进程表抽查无密钥。 +- V-5 取消与配额真杀：同步 SSH 阻塞中取消任务、配额超限 + 时，看门狗调 `cancel_vm` 真关 VM。验收：VM 进程消失， + 任务状态 interrupted 且原因写回正确。 +- V-6 审计落盘：DB 审计回调失败时，`chv-audit-spill.jsonl` + 里有该事件行、`audit_failure_count` +1、任务不中断。 + 验收：落盘行 event/task_id/detail 与丢失事件一致。 +- V-7 动态安全负向：客机 root unset 代理后直连（应被 + tap/防火墙挡住）、跨 VM 探活与 IP 冒用（SSH 钉死应拒）、 + 崩溃注入后残留扫描（私钥/seed/盘应被清理或脏标记拒起）。 + 验收：每一项都有“被挡住”的证据，不是有代码就过。 +- V-8 负载：4 VM 并发（`CHV_MAX_VMS` 默认值）下容量门 + 计数含重启后残留 socket 扫描，死 socket 不占名额、活 + VM 不被超发。验收：并发任务全部完成或明确排队拒绝， + 无静默降级到 Docker/Local。 +- V-9 供应链清单：`generate_lockfile.py` 在构建宿主上跑 + 出真实 lockfile，`--verify` 通过且 unresolved 为空； + 再用 `check_lockfile.py` 对 BUILDINFO.versions 复核。 + 验收：lockfile 由构建宿主产出，不是本轨合成数据。 + +### 13.3 灰度启用清单 + +- 阶段 0（合并后、默认关闭）：合并整条链，CHV 仍默认 + 关闭，不进生产任务路由；先跑 §13.2 的 V-1～V-9。 +- 阶段 1（单租户可信项目）：只对明确指定的单租户项目 + 开 CHV（`CHV_REUSE_SINGLE_TENANT_EXPERIMENT` 与 + `CHV_SINGLE_TENANT_PROJECT` 的声明口径不变，复用 + 实验仍需同时声明，两者缺一即 preflight 拒绝）。 +- 阶段 2（受限多租户）：放开到受邀项目，监控 VM 残留盘、 + 审计落盘、容量门与快照配额三项指标一周无异常。 +- 阶段 3（全量多租户）：仅在阶段 2 指标干净 + 动态安全 + 审计结论由 No-Go 改为有条件启用后，才可讨论；绝不 + 从阶段 0 直跳全量。 + +### 13.4 外部关卡（本轨代码做不到，逐项列出） + +1. CI 账单恢复变绿（硬门禁，合并前提）。 +2. 链上 10 个 PR 按顺序合并 + 本 PR 合并（需先过 §13.1）。 +3. 用户侧 provision KVM 宿主（裸金属或嵌套虚拟化实例， + 须见 `/dev/kvm`；生产宿主与本 VM 均无 KVM，无法替代）。 +4. 客机镜像真实构建与发布（`build.sh` 全量跑通 + V-9）。 +5. §13.2 真机验证全部通过（含 V-7 动态安全负向）。 +6. 负载测试（V-8）与安全审计结论更新（No-Go → 有条件启用）。 +7. 前端 live 点击验证（用户侧 live browser 检查清单见 + 本 PR 描述；回环地址限制下浏览器任务无法代劳）。