admin
42a130e4f2
- graphctl.py 并发修复
· 临时名原来固定成 graph.json.tmp ⇒ 两进程同时写直接撞 PermissionError（桌面实测第 14 条）
· 新增 _update()：读最新状态 → 重放本次变更 → 原子替换 → 回读核对 _rev，
被覆盖则重读重做（最多 3 次），仍失败就报错退出
· 修掉更隐蔽的一层：后写的整份覆盖前写会让节点状态静默消失
· start / report 改为走 _update，业务校验放在 mutate 内
- test_graphctl.py
· 每例只新建自己的目录、跑完不删，整轮收尾清一次 ⇒ 不再触发沙箱「批量删除」守卫
（此前 49 条里 20 条报 ERROR，全部是本机策略拦清理，用例本身零失败）
· 新增 T9_Concurrency 三个用例：并发开工、并发回报、版本号每次写都前进
· 49 → 52 条，实跑 Ran 52 tests OK（13 s）
- README：补「并发写是怎么保证的」三层说明，测试数与用时更新
460 lines
17 KiB
Python
460 lines
17 KiB
Python
#!/usr/bin/env python3
|
||
# -*- coding: utf-8 -*-
|
||
"""graphctl.py —— 「图模式」的独立实现（2026-10-18）。
|
||
|
||
## 它是什么
|
||
|
||
把目标拆成节点、按依赖推进的**独立**运行模式。**⛔ 不修改任何现有机制**：
|
||
不读也不写 `goal.json` / `tasks.json` / `collabd-state.json`，不参与六道闸，
|
||
不接管常驻循环，不改变现有会话的任何行为。
|
||
|
||
## 数据落点（独立目录，与现有状态完全分开）
|
||
|
||
<ws>/.workbuddy/graph/<gid>/graph.json 图的唯一真源
|
||
<ws>/.workbuddy/graph/<gid>/events.jsonl 每次状态变更的流水
|
||
|
||
## 图的结构
|
||
|
||
{
|
||
"gid": "20261018-2230-ui改造",
|
||
"title": "给三块功能各加一份文档",
|
||
"created_at": "2026-10-18T22:30:00",
|
||
"meta": {"max_parallel": 3},
|
||
"nodes": {
|
||
"N1": {"title": "...", "deps": [], "domain": ["docs/ui/"],
|
||
"accepts": ["产物在位"], "status": "pending",
|
||
"evidence": "", "attempts": 0}
|
||
}
|
||
}
|
||
|
||
`status` 四态：`pending` → `running` → `done` ／ `blocked`。
|
||
|
||
## 命令
|
||
|
||
init --ws <工作区> --title <目标> (--kpi "判据A;判据B" | --nodes-json <文件|->)
|
||
check --ws <ws> --gid <gid> 环／悬空依赖／空域／同批域冲突
|
||
ready --ws <ws> --gid <gid> 就绪集（含为什么没被选）
|
||
start --ws <ws> --gid <gid> --node N1
|
||
report --ws <ws> --gid <gid> --node N1 --state done|blocked [--evidence "..."]
|
||
show --ws <ws> --gid <gid>
|
||
list --ws <ws>
|
||
|
||
## 就绪的判据（四条全中）
|
||
|
||
① 依赖全部 done ② 自己还是 pending
|
||
③ 域不与任何 running 节点重叠，也不与本批已选节点重叠
|
||
④ 不超 max_parallel（running 占名额）
|
||
|
||
## 状态的走法
|
||
|
||
pending --start--> running --report done--> done
|
||
|--report blocked--> blocked --start--> running（重试，attempts 已累加）
|
||
|
||
`start` 只拒 `running`（已在跑）与 `done`（已完成）；`blocked` 可重开。
|
||
|
||
## 域的用法
|
||
|
||
`domain` 是一组路径前缀（相对工作区根）。两节点域重叠 ⇒ 不能同批跑。
|
||
判定：任一前缀互为前缀即视为重叠（`docs/` 与 `docs/ui/` 算重叠）。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import argparse
|
||
import io
|
||
import json
|
||
import os
|
||
import random
|
||
import sys
|
||
import time
|
||
from pathlib import Path
|
||
|
||
MARK = ".workbuddy/graph"
|
||
STATUSES = ("pending", "running", "done", "blocked")
|
||
|
||
|
||
# ── 基础设施 ────────────────────────────────────────────────────────────
|
||
def _now() -> str:
|
||
return time.strftime("%Y-%m-%dT%H:%M:%S")
|
||
|
||
|
||
def _root(ws: str) -> Path:
|
||
return Path(ws).resolve() / MARK
|
||
|
||
|
||
def _gdir(ws: str, gid: str) -> Path:
|
||
return _root(ws) / gid
|
||
|
||
|
||
def _load(ws: str, gid: str) -> dict:
|
||
p = _gdir(ws, gid) / "graph.json"
|
||
if not p.is_file():
|
||
raise SystemExit("找不到图：%s" % p)
|
||
return json.loads(p.read_text(encoding="utf-8"))
|
||
|
||
|
||
def _save(ws: str, gid: str, g: dict) -> None:
|
||
"""整份写回（新建图时用）。临时名带 pid ＋ 随机数 ⇒ 两进程不会撞同一个名字。"""
|
||
p = _gdir(ws, gid) / "graph.json"
|
||
tmp = p.with_suffix(".%d.%d.json.tmp" % (os.getpid(), random.randint(0, 1 << 30)))
|
||
tmp.write_text(json.dumps(g, ensure_ascii=False, indent=2) + "\n",
|
||
encoding="utf-8", newline="\n")
|
||
os.replace(tmp, p) # 原子替换，避免半截文件
|
||
|
||
|
||
def _update(ws: str, gid: str, mutate, retries: int = 3) -> dict:
|
||
"""读-改-写，带版本号乐观重试（2026-10-19 修两处实测缺陷）。
|
||
|
||
① 临时名原来固定成 `graph.json.tmp` ⇒ 两进程同时写直接撞 `PermissionError`
|
||
（桌面测试报告第 14 条实测：同时对 C、D 执行 start 就复现）。
|
||
② 更隐蔽的一层：就算临时名不撞，两人同时读-改-写会**丢更新** ——
|
||
后写的整份覆盖前写的，节点状态静默消失，连报错都没有。
|
||
③ 对策：图里带 `_rev`，每次在**最新磁盘状态**上重放变更，写完回读核对；
|
||
被别的进程覆盖就重读重做，最多 `retries` 次，仍失败则**报错退出**（⛔ 不静默）。
|
||
⚠️ 本机不用 OS 文件锁（实测会卡死），靠「唯一临时名 ＋ 原子替换 ＋ 回读核对」。
|
||
"""
|
||
p = _gdir(ws, gid) / "graph.json"
|
||
if not p.is_file():
|
||
raise SystemExit("找不到图：%s" % p)
|
||
for _ in range(retries):
|
||
g = json.loads(p.read_text(encoding="utf-8"))
|
||
rev = int(g.get("_rev") or 0)
|
||
mutate(g)
|
||
g["_rev"] = rev + 1
|
||
tmp = p.with_suffix(".%d.%d.json.tmp" % (os.getpid(), random.randint(0, 1 << 30)))
|
||
tmp.write_text(json.dumps(g, ensure_ascii=False, indent=2) + "\n",
|
||
encoding="utf-8", newline="\n")
|
||
os.replace(tmp, p)
|
||
try:
|
||
back = json.loads(p.read_text(encoding="utf-8"))
|
||
except Exception: # noqa: BLE001
|
||
back = {}
|
||
if int(back.get("_rev") or 0) == rev + 1:
|
||
return back
|
||
raise SystemExit("并发冲突：%s 连续 %d 次被别的进程覆盖，稍后重试" % (p, retries))
|
||
|
||
|
||
def _event(ws: str, gid: str, obj: dict) -> None:
|
||
p = _gdir(ws, gid) / "events.jsonl"
|
||
obj = dict(obj)
|
||
obj.setdefault("at", _now())
|
||
with io.open(p, "a", encoding="utf-8", newline="\n") as f:
|
||
f.write(json.dumps(obj, ensure_ascii=False) + "\n")
|
||
|
||
|
||
def _out(obj) -> None:
|
||
data = json.dumps(obj, ensure_ascii=False, indent=2) + "\n"
|
||
sys.stdout.buffer.write(data.encode("utf-8"))
|
||
|
||
|
||
# ── 域与依赖 ────────────────────────────────────────────────────────────
|
||
def _norm_dom(d) -> list:
|
||
if isinstance(d, str):
|
||
d = [d]
|
||
out = []
|
||
for x in (d or []):
|
||
s = str(x).strip().replace("\\", "/").lstrip("./")
|
||
if s:
|
||
out.append(s if s.endswith("/") else s + "/")
|
||
return out
|
||
|
||
|
||
def _dom_overlap(a, b) -> bool:
|
||
"""任一前缀互为前缀 ⇒ 重叠（docs/ 与 docs/ui/ 算重叠）。"""
|
||
for x in a:
|
||
for y in b:
|
||
if x.startswith(y) or y.startswith(x):
|
||
return True
|
||
return False
|
||
|
||
|
||
def _deps(n: dict) -> list:
|
||
v = n.get("deps") or []
|
||
return [str(x) for x in v] if isinstance(v, list) else []
|
||
|
||
|
||
def _is_done(nodes: dict, nid: str) -> bool:
|
||
return str((nodes.get(nid) or {}).get("status")) == "done"
|
||
|
||
|
||
# ── 校验 ────────────────────────────────────────────────────────────────
|
||
def _find_cycle(nodes: dict) -> list:
|
||
color, stack, hit = {}, [], []
|
||
|
||
def dfs(u):
|
||
color[u] = 1
|
||
stack.append(u)
|
||
for v in _deps(nodes.get(u) or {}):
|
||
if v not in nodes:
|
||
continue
|
||
if color.get(v) == 1:
|
||
i = stack.index(v)
|
||
hit.extend(stack[i:] + [v])
|
||
return True
|
||
if color.get(v, 0) == 0 and dfs(v):
|
||
return True
|
||
stack.pop()
|
||
color[u] = 2
|
||
return False
|
||
|
||
for k in nodes:
|
||
if color.get(k, 0) == 0 and dfs(k):
|
||
return hit
|
||
return []
|
||
|
||
|
||
def cmd_check(a) -> int:
|
||
g = _load(a.ws, a.gid)
|
||
nodes = g.get("nodes") or {}
|
||
errs, warns = [], []
|
||
|
||
cyc = _find_cycle(nodes)
|
||
if cyc:
|
||
errs.append("存在环：%s" % " → ".join(cyc))
|
||
|
||
for nid, n in nodes.items():
|
||
for d in _deps(n):
|
||
if d not in nodes:
|
||
errs.append("%s 的依赖 %s 不存在" % (nid, d))
|
||
if not _norm_dom(n.get("domain")):
|
||
warns.append("%s 没有声明域 ⇒ 域比较视为不冲突，可与任意节点同批；"
|
||
"保守做法是让它单独跑" % nid)
|
||
if str(n.get("status")) not in STATUSES:
|
||
errs.append("%s 的状态非法：%r" % (nid, n.get("status")))
|
||
if not (n.get("accepts") or []):
|
||
warns.append("%s 没有验收判据 ⇒ 判不出完成" % nid)
|
||
|
||
ids = list(nodes)
|
||
for i in range(len(ids)):
|
||
for j in range(i + 1, len(ids)):
|
||
a1, a2 = ids[i], ids[j]
|
||
if _norm_dom(nodes[a1].get("domain")) and \
|
||
_dom_overlap(_norm_dom(nodes[a1].get("domain")),
|
||
_norm_dom(nodes[a2].get("domain"))):
|
||
warns.append("%s 与 %s 的域重叠 ⇒ 永不同批" % (a1, a2))
|
||
|
||
_out({"ok": not errs, "errors": errs, "warnings": warns,
|
||
"nodes": len(nodes)})
|
||
return 0 if not errs else 1
|
||
|
||
|
||
# ── 就绪集 ──────────────────────────────────────────────────────────────
|
||
def _ready(g: dict) -> dict:
|
||
nodes = g.get("nodes") or {}
|
||
cap = int((g.get("meta") or {}).get("max_parallel") or 3)
|
||
running = [k for k, n in nodes.items() if str(n.get("status")) == "running"]
|
||
ready, waiting = [], []
|
||
|
||
for nid, n in nodes.items():
|
||
st = str(n.get("status"))
|
||
if st != "pending":
|
||
continue
|
||
nd = _deps(n)
|
||
lack = [d for d in nd if not _is_done(nodes, d)]
|
||
if lack:
|
||
waiting.append({"node": nid, "why": "依赖未完成：%s" % ", ".join(lack)})
|
||
continue
|
||
dom = _norm_dom(n.get("domain"))
|
||
clash = [r for r in running
|
||
if dom and _norm_dom((nodes.get(r) or {}).get("domain"))
|
||
and _dom_overlap(dom, _norm_dom((nodes.get(r) or {}).get("domain")))]
|
||
if clash:
|
||
waiting.append({"node": nid, "why": "域与在跑的冲突：%s" % ", ".join(clash)})
|
||
continue
|
||
ready.append(nid)
|
||
|
||
slots = max(0, cap - len(running))
|
||
picked, backlog = [], []
|
||
for nid in ready:
|
||
if len(picked) >= slots:
|
||
backlog.append(nid)
|
||
waiting.append({"node": nid, "why": "并发上限 %d 已占满" % cap})
|
||
continue
|
||
dom = _norm_dom((nodes.get(nid) or {}).get("domain"))
|
||
clash = [p for p in picked
|
||
if dom and _norm_dom((nodes.get(p) or {}).get("domain"))
|
||
and _dom_overlap(dom, _norm_dom((nodes.get(p) or {}).get("domain")))]
|
||
if clash:
|
||
# 同批也不能撞域：check 已警告「永不同批」，ready 必须兑现
|
||
waiting.append({"node": nid, "why": "域与本批已选节点重叠：%s" % ", ".join(clash)})
|
||
continue
|
||
picked.append(nid)
|
||
|
||
return {"gid": g.get("gid"), "running": running, "ready": picked,
|
||
"backlog": backlog, "waiting": waiting,
|
||
"cap": cap, "slots": slots}
|
||
|
||
|
||
def cmd_ready(a) -> int:
|
||
g = _load(a.ws, a.gid)
|
||
_out(_ready(g))
|
||
return 0
|
||
|
||
|
||
# ── 状态流转 ────────────────────────────────────────────────────────────
|
||
def cmd_start(a) -> int:
|
||
def _mut(g):
|
||
n = (g.get("nodes") or {}).get(a.node)
|
||
if not n:
|
||
raise SystemExit("没有这个节点：%s" % a.node)
|
||
if str(n.get("status")) == "running":
|
||
raise SystemExit("节点 %s 已在跑，不能重复开工" % a.node)
|
||
if str(n.get("status")) == "done":
|
||
raise SystemExit("节点 %s 已完成，不能开工" % a.node)
|
||
# pending 与 blocked 都可开工：blocked 是「这轮没过」，重试时 attempts 保留
|
||
n["status"] = "running"
|
||
_update(a.ws, a.gid, _mut)
|
||
_event(a.ws, a.gid, {"node": a.node, "to": "running", "by": a.by or ""})
|
||
_out({"ok": True, "node": a.node, "status": "running"})
|
||
return 0
|
||
|
||
|
||
def cmd_report(a) -> int:
|
||
if a.state not in ("done", "blocked"):
|
||
raise SystemExit("--state 只认 done / blocked")
|
||
|
||
def _mut(g):
|
||
n = (g.get("nodes") or {}).get(a.node)
|
||
if not n:
|
||
raise SystemExit("没有这个节点：%s" % a.node)
|
||
if a.evidence:
|
||
n["evidence"] = a.evidence
|
||
n["status"] = a.state
|
||
if a.state == "blocked":
|
||
n["attempts"] = int(n.get("attempts") or 0) + 1
|
||
|
||
g = _update(a.ws, a.gid, _mut)
|
||
_event(a.ws, a.gid, {"node": a.node, "to": a.state,
|
||
"evidence": a.evidence or "", "by": a.by or ""})
|
||
r = _ready(g)
|
||
_out({"ok": True, "node": a.node, "status": a.state,
|
||
"next_ready": r["ready"]})
|
||
return 0
|
||
|
||
|
||
# ── 建图 ────────────────────────────────────────────────────────────────
|
||
def _gid_of(title: str) -> str:
|
||
safe = "".join(c for c in title if c not in '\\/:*?"<>|\r\n\t')[:24].strip()
|
||
return "%s-%s" % (time.strftime("%Y%m%d-%H%M"), safe or "graph")
|
||
|
||
|
||
def cmd_init(a) -> int:
|
||
if a.nodes_json:
|
||
raw = sys.stdin.read() if a.nodes_json == "-" else \
|
||
Path(a.nodes_json).read_text(encoding="utf-8")
|
||
spec = json.loads(raw)
|
||
elif a.kpi:
|
||
spec = {"nodes": [
|
||
{"id": "N%d" % (i + 1), "title": k.strip(), "deps": [],
|
||
"domain": [], "accepts": [k.strip()]}
|
||
for i, k in enumerate([x for x in a.kpi.replace("；", ";").split(";") if x.strip()])
|
||
]}
|
||
else:
|
||
raise SystemExit("要给 --kpi 或 --nodes-json")
|
||
|
||
nodes = {}
|
||
for i, n in enumerate(spec.get("nodes") or []):
|
||
raw = n.get("id")
|
||
nid = str(raw).strip() if raw not in (None, "") else ""
|
||
if not nid:
|
||
k = i + 1
|
||
while ("N%d" % k) in nodes: # 自动编号要避开已占用的 id
|
||
k += 1
|
||
nid = "N%d" % k
|
||
if nid in nodes: # 显式重复 ⇒ 硬错，⛔ 不静默覆盖前一个
|
||
raise SystemExit("节点 id 重复：%s" % nid)
|
||
nodes[nid] = {
|
||
"title": str(n.get("title") or nid),
|
||
"deps": [str(x) for x in (n.get("deps") or [])],
|
||
"domain": _norm_dom(n.get("domain")),
|
||
"accepts": [str(x) for x in (n.get("accepts") or [])],
|
||
"status": "pending", "evidence": "", "attempts": 0,
|
||
}
|
||
gid = a.gid or _gid_of(a.title)
|
||
g = {"gid": gid, "title": a.title, "created_at": _now(),
|
||
"meta": {"max_parallel": int(spec.get("max_parallel") or a.max_parallel or 3)},
|
||
"nodes": nodes}
|
||
d = _gdir(a.ws, gid)
|
||
d.mkdir(parents=True, exist_ok=True)
|
||
_save(a.ws, gid, g)
|
||
_event(a.ws, gid, {"event": "init", "nodes": list(nodes)})
|
||
_out({"ok": True, "gid": gid, "dir": str(d), "nodes": len(nodes),
|
||
"note": "独立图：⛔ 不参与现有机制（不写 goal.json / tasks.json）"})
|
||
return 0
|
||
|
||
|
||
def cmd_show(a) -> int:
|
||
_out(_load(a.ws, a.gid))
|
||
return 0
|
||
|
||
|
||
def cmd_list(a) -> int:
|
||
root = _root(a.ws)
|
||
rows = []
|
||
if root.is_dir():
|
||
for d in sorted(root.iterdir()):
|
||
p = d / "graph.json"
|
||
if not p.is_file():
|
||
continue
|
||
try:
|
||
g = json.loads(p.read_text(encoding="utf-8"))
|
||
except Exception:
|
||
continue
|
||
nodes = g.get("nodes") or {}
|
||
cnt = {}
|
||
for n in nodes.values():
|
||
cnt[str(n.get("status"))] = cnt.get(str(n.get("status")), 0) + 1
|
||
rows.append({"gid": g.get("gid"), "title": g.get("title"),
|
||
"nodes": len(nodes), "by_status": cnt,
|
||
"created_at": g.get("created_at")})
|
||
_out({"ws": str(Path(a.ws).resolve()), "graphs": rows})
|
||
return 0
|
||
|
||
|
||
# ── 入口 ────────────────────────────────────────────────────────────────
|
||
def main(argv=None) -> int:
|
||
ap = argparse.ArgumentParser(prog="graphctl", description="图模式独立实现（不碰现有机制）")
|
||
sub = ap.add_subparsers(dest="cmd", required=True)
|
||
|
||
p = sub.add_parser("init", help="按判据或节点定义建图")
|
||
p.add_argument("--ws", required=True)
|
||
p.add_argument("--title", required=True)
|
||
p.add_argument("--gid", default="")
|
||
p.add_argument("--kpi", default="", help="判据，分号分隔；每条生成一个节点")
|
||
p.add_argument("--nodes-json", default="", help="节点定义文件（- 表示标准输入）")
|
||
p.add_argument("--max-parallel", type=int, default=3)
|
||
p.set_defaults(fn=cmd_init)
|
||
|
||
for name, fn, extra in (("check", cmd_check, None), ("ready", cmd_ready, None),
|
||
("show", cmd_show, None)):
|
||
p = sub.add_parser(name)
|
||
p.add_argument("--ws", required=True)
|
||
p.add_argument("--gid", required=True)
|
||
p.set_defaults(fn=fn)
|
||
|
||
p = sub.add_parser("start", help="节点开工")
|
||
p.add_argument("--ws", required=True)
|
||
p.add_argument("--gid", required=True)
|
||
p.add_argument("--node", required=True)
|
||
p.add_argument("--by", default="")
|
||
p.set_defaults(fn=cmd_start)
|
||
|
||
p = sub.add_parser("report", help="节点回报")
|
||
p.add_argument("--ws", required=True)
|
||
p.add_argument("--gid", required=True)
|
||
p.add_argument("--node", required=True)
|
||
p.add_argument("--state", required=True)
|
||
p.add_argument("--evidence", default="")
|
||
p.add_argument("--by", default="")
|
||
p.set_defaults(fn=cmd_report)
|
||
|
||
p = sub.add_parser("list")
|
||
p.add_argument("--ws", required=True)
|
||
p.set_defaults(fn=cmd_list)
|
||
|
||
a = ap.parse_args(argv)
|
||
return a.fn(a)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
sys.exit(main())
|