refactor(remote_env): 依赖安装改远程分离式——nohup逐包安装+进度文件,轮询无状态

This commit is contained in:
ymq 2026-08-28 21:46:27 +08:00
parent 023d4a67fa
commit ffe19ab95b

View File

@ -239,126 +239,102 @@ def collect_skill_deps(skills_base: str, org_id: str, heavy_skip: str) -> dict:
return {"specs": sorted(specs.values()), "heavy_skipped": sorted(heavy_skipped)}
async def install_remote_deps(env: dict, specs: list, pip_index: str,
timeout: int = 1200) -> dict:
"""在远程 ~/<remote_dir>/.venv-skills 建 venv 并安装依赖。幂等。
async def launch_remote_install(env: dict, specs: list, pip_index: str, rel: str) -> bool:
"""在远程以 nohup 分离进程启动逐包安装(脱离 SSH 会话,不怕断连/重启)。
specs 经 base64 传输(含 >=[] 等特殊字符,避开 shell 展开)。
返回 {ok, installed:[...], failed:[...], error}
进度写 ~/<rel>/.venv-skills-install.progress(OK:/FAILED: 行),
完成写 ~/<rel>/.venv-skills-install.done(汇总)。
本调用秒级返回,是否成功启动由返回值给出。
"""
if not specs:
return {"ok": True, "installed": [], "failed": [], "skipped_reason": "无可安装依赖"}
err = _validate_remote_dir(env.get("remote_dir", ""))
if err:
return {"ok": False, "error": err}
rel = _rel_remote_dir(env.get("remote_dir", ""))
payload = base64.b64encode(json.dumps(specs).encode()).decode()
# 不用 set -e / 管道:管道会吞掉 pip 的退出码($? 变成 tail 的状态)。
# set -f 关 glob:pip 规格里的 * / [...] 不能被远程 shell 展开。
# 完整输出由 SSH 捕获,报告时截取尾部。
script = (
'RD="$HOME/' + rel + '"; VENV="$RD/.venv-skills"; '
'command -v python3 >/dev/null || { echo "NO_PYTHON3"; exit 3; }; '
'if [ ! -x "$VENV/bin/python" ]; then python3 -m venv "$VENV" || { echo "VENV_FAIL"; exit 4; }; fi; '
'set -f; RC=0; "$VENV/bin/pip" install --no-input --disable-pip-version-check '
+ (('-i ' + shlex.quote(pip_index) + ' ') if pip_index else '')
+ '$(echo "' + payload + '" | base64 -d | python3 -c '
'"import json,sys;print(chr(32).join(json.load(sys.stdin)))") '
'|| RC=$?; '
'echo "INSTALL_RC=$RC"'
)
ssh_cmd = ["ssh"] + _ssh_common_args(env) + [_ssh_target(env), script]
try:
proc = await asyncio.create_subprocess_exec(
*ssh_cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
stdout, stderr = await asyncio.wait_for(proc.communicate(), timeout=timeout + 30)
except asyncio.TimeoutError:
return {"ok": False, "error": f"依赖安装超时({timeout}s),venv 可能装了一半,可重跑"}
except Exception as e:
return {"ok": False, "error": "SSH 执行失败: " + str(e)[:300]}
out = stdout.decode("utf-8", "replace")
if "NO_PYTHON3" in out:
return {"ok": False, "error": "远程主机缺少 python3"}
if proc.returncode not in (0, 1):
return {"ok": False, "error": "venv 创建失败: "
+ (out + stderr.decode('utf-8', 'replace'))[-600:]}
# 解析 pip 退出码(由脚本末尾 echo INSTALL_RC=$RC 输出)
rc = -1
for line in out.splitlines():
if line.startswith("INSTALL_RC="):
try:
rc = int(line.split("=", 1)[1])
except Exception:
pass
# pip 逐包结果解析:新版 pip 解析器失败时要么全装要么全不装,
# 从 ERROR 行提取包名。常见格式:
# ERROR: Could not find a version that satisfies the requirement <name> ...
# ERROR: No matching distribution found for <name>
failed = []
for line in out.splitlines():
if "ERROR:" not in line:
continue
m = (re.search(r"the requirement ([A-Za-z0-9._\[\]-]+)", line)
or re.search(r"No matching distribution found for ([A-Za-z0-9._\[\]-]+)", line))
failed.append(m.group(1) if m else line.strip()[:120])
if rc != 0 and not failed:
failed = ["部分包安装失败,详见日志末尾"]
if rc == 0:
installed = list(specs)
failed = []
else:
# 批量失败 → 逐包降级:一次 SSH 会话内逐包安装(pip 幂等,已装的秒过),
# 部分成功优于全败。
return await _install_per_package(env, specs, pip_index, rel, timeout)
return {"ok": rc == 0, "installed": installed, "failed": failed,
"venv": "~/" + rel + "/.venv-skills",
"tail": out[-800:]}
async def _install_per_package(env: dict, specs: list, pip_index: str,
rel: str, timeout: int) -> dict:
"""逐包安装(批量失败后的降级路径)。返回格式与 install_remote_deps 一致。"""
payload = base64.b64encode(json.dumps(specs).encode()).decode()
index_arg = ('-i ' + shlex.quote(pip_index) + ' ') if pip_index else ''
script = (
# 分离脚本:建 venv → 逐包安装 → 写进度/完成标记。
# 用 set -f 防 pip 规格里 * / [...] 被远程 shell 展开。
inner = (
'RD="$HOME/' + rel + '"; VENV="$RD/.venv-skills"; '
'[ -x "$VENV/bin/pip" ] || { echo "VENV_MISSING"; exit 4; }; '
'set -f; '
'PROG="$RD/.venv-skills-install.progress"; DONE="$RD/.venv-skills-install.done"; '
'if [ ! -x "$VENV/bin/python" ]; then '
'python3 -m venv "$VENV" || { echo "VENV_FAIL" > "$DONE"; exit 4; }; fi; '
'set -f; : > "$PROG"; '
'echo "' + payload + '" | base64 -d | python3 -c '
'"import json,sys;print(chr(10).join(json.load(sys.stdin)))" | '
'while IFS= read -r pkg; do '
' [ -n "$pkg" ] || continue; '
' if "$VENV/bin/pip" install --no-input --disable-pip-version-check '
+ index_arg + '"$pkg" >/dev/null 2>&1; then echo "OK:$pkg"; '
' else echo "FAILED:$pkg"; fi; '
'done'
'[ -n "$pkg" ] || continue; '
'if "$VENV/bin/pip" install --no-input --disable-pip-version-check '
+ index_arg + '"$pkg" >/dev/null 2>&1; then echo "OK:$pkg" >> "$PROG"; '
'else echo "FAILED:$pkg" >> "$PROG"; fi; done; '
'OKC=$(grep -c "^OK:" "$PROG" 2>/dev/null || echo 0); '
'FC=$(grep -c "^FAILED:" "$PROG" 2>/dev/null || echo 0); '
'echo "OK=$OKC FAILED=$FC" > "$DONE"'
)
script = (
'RD="$HOME/' + rel + '"; '
'command -v python3 >/dev/null || { echo "NO_PYTHON3"; exit 3; }; '
'mkdir -p "$RD"; '
'nohup bash -c ' + shlex.quote(inner) +
' </dev/null >"$RD/.venv-skills-install.log" 2>&1 & echo LAUNCHED'
)
ssh_cmd = ["ssh"] + _ssh_common_args(env) + [_ssh_target(env), script]
try:
proc = await asyncio.create_subprocess_exec(
*ssh_cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
stdout, _stderr = await asyncio.wait_for(proc.communicate(), timeout=timeout + 30)
except asyncio.TimeoutError:
return {"ok": False, "error": f"逐包安装超时({timeout}s),可重跑"}
stdout, stderr = await asyncio.wait_for(proc.communicate(), timeout=60)
except Exception as e:
return {"ok": False, "error": "SSH 执行失败: " + str(e)[:300]}
logger.warning("launch_remote_install 启动失败: %s", e)
return False
out = stdout.decode("utf-8", "replace")
if "NO_PYTHON3" in out:
return False
return "LAUNCHED" in out
async def poll_remote_install(env: dict, rel: str) -> dict:
"""轮询远程安装进度。短 SSH 调用。
返回 {done:bool, ok:bool, installed:[], failed:[], venv, ...}
"""
script = (
'RD="$HOME/' + rel + '"; DONE="$RD/.venv-skills-install.done"; '
'PROG="$RD/.venv-skills-install.progress"; '
'if [ -f "$DONE" ]; then echo "STATE=done"; cat "$DONE"; echo "===PROG==="; '
'cat "$PROG" 2>/dev/null; '
'else echo "STATE=running"; cat "$PROG" 2>/dev/null; fi'
)
ssh_cmd = ["ssh"] + _ssh_common_args(env) + [_ssh_target(env), script]
try:
proc = await asyncio.create_subprocess_exec(
*ssh_cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
stdout, _stderr = await asyncio.wait_for(proc.communicate(), timeout=30)
except Exception as e:
return {"done": False, "error": "轮询失败: " + str(e)[:200]}
out = stdout.decode("utf-8", "replace")
done = "STATE=done" in out
body = out.split("STATE=", 1)[1] if "STATE=" in out else out
if "===PROG===" in body:
summary, prog = body.split("===PROG===", 1)
else:
summary, prog = body, ""
installed, failed = [], []
for line in out.splitlines():
for line in prog.splitlines():
line = line.strip()
if line.startswith("OK:"):
installed.append(line[3:])
elif line.startswith("FAILED:"):
failed.append(line[7:])
return {"ok": len(failed) == 0 and len(installed) > 0,
"installed": installed, "failed": failed,
"venv": "~/" + rel + "/.venv-skills",
"mode": "per-package", "tail": out[-500:]}
result = {
"done": done,
"ok": done and len(failed) == 0 and len(installed) > 0,
"installed": installed,
"failed": failed,
"venv": "~/" + rel + "/.venv-skills",
"mode": "per-package-detached",
}
if done:
result["summary"] = summary.strip()
return result
# ── 总编排 ────────────────────────────────────────────────────
@ -399,55 +375,33 @@ async def init_remote_env(env: dict, org_id: str = "", sor=None,
report["error"] = "技能同步失败: " + str(sk.get("error", ""))
return report
# 4. 依赖解析(快)+ 安装(慢,后台执行)
# 4. 依赖解析(快)+ 启动远程分离式安装(nohup,脱离 SSH 会话,
# 进度落远程文件,轮询无状态——服务重启/断连都不影响)
cfg = await _load_cfg(sor)
deps = collect_skill_deps(skills_base or _default_skills_base(), org_id,
cfg["heavy_skip"])
report["steps"]["deps_found"] = deps
task_key = _start_deps_install(env, deps["specs"], cfg)
report["deps_task"] = task_key
report["steps"]["deps_install"] = {"status": "running", "task": task_key}
launched = await launch_remote_install(env, deps["specs"], cfg["pip_index"], rel)
report["steps"]["deps_launch"] = {"ok": launched, "count": len(deps["specs"])}
if not launched:
report["warning"] = "依赖安装进程启动失败(远程可能缺 python3),请重新运行初始化"
return report
# ── 依赖安装后台任务与状态 ────────────────────────────────────
# 依赖安装可能耗时数分钟(超过 nginx 600s 代理超时),不能在 API 请求内同步等。
# 编排只启动后台任务并返回任务号,前端用 init_remote_env_status 轮询结果。
# 状态存进程内存(单进程部署;重启丢状态则重跑,幂等无副作用)。
# ── 依赖安装状态查询(无状态,轮询远程进度文件)────────────────
# 安装进程由远程 nohup 托管,进度写 ~/<remote_dir>/.venv-skills-install.*。
# 查询只需 SSH 读文件,不依赖本进程内存,服务重启后仍可查。
_INSTALL_TASKS = {} # {task_key: {"status": ..., "result": ..., "started": ts}}
def _start_deps_install(env: dict, specs: list, cfg: dict) -> str:
from appPublic.uniqueID import getID
task_key = "deps_" + getID()
_INSTALL_TASKS[task_key] = {"status": "running", "result": None,
"specs": list(specs)}
async def _run():
try:
r = await install_remote_deps(env, specs, cfg["pip_index"],
cfg["install_timeout"])
_INSTALL_TASKS[task_key]["result"] = r
_INSTALL_TASKS[task_key]["status"] = "done"
except Exception as e:
_INSTALL_TASKS[task_key]["result"] = {"ok": False, "error": str(e)[:500]}
_INSTALL_TASKS[task_key]["status"] = "done"
asyncio.create_task(_run())
return task_key
def get_deps_install_status(task_key: str) -> dict:
t = _INSTALL_TASKS.get(task_key or "")
if not t:
return {"ok": False, "error": "任务不存在(可能服务重启过,请重新初始化)"}
if t["status"] == "running":
return {"ok": True, "status": "running",
"message": f"依赖安装中({len(t.get('specs', []))} 个包),请稍候再查询"}
r = dict(t.get("result") or {})
r["status"] = "done"
return r
async def get_deps_install_status(sor, owner_type: str, owner_id: str) -> dict:
"""查询远程依赖安装进度。从 sd_work_envs 取远程配置后轮询进度文件。"""
from .work_env import get_work_env
env = await get_work_env(
sor, owner_id if owner_type == "user" else "",
owner_id if owner_type == "org" else "")
if env.get("mode") != "remote" or not env.get("remote_dir"):
return {"ok": False, "error": "当前非远程模式,无安装任务"}
rel = _rel_remote_dir(env.get("remote_dir", ""))
return await poll_remote_install(env, rel)
def _default_skills_base() -> str: