#!/usr/bin/env python3 # 织忆写时四操作·影子日志分布报告(t_3e98b65a AC3) # # 影子观察 48h 后,把 /var/lib/memoryweave/write-ops-shadow.log(JSONL,每行一个 # WriteVerdict)聚合成四操作分布 + 误判抽样。转 apply 前跑这个,人工判抽样样本, # 确认 UPDATE/CONFLICT 误判 ≤10% 才转 apply(止损线,见卡 body)。 # # 用法: # zhiyi-writeops-report.py 全部统计 # zhiyi-writeops-report.py --log 指定影子日志路径 # zhiyi-writeops-report.py --sample 5 误判抽样条数(默认 5) # zhiyi-writeops-report.py --since 48h 只看近 48h(时间过滤,默认全部) # # 环境变量: # ZHIYI_WRITE_OPS_LOG 影子日志路径(默认 /var/lib/memoryweave/write-ops-shadow.log) import argparse import json import os import sys def load_log(path): if not os.path.exists(path): sys.exit(f"影子日志不存在: {path}") rows = [] with open(path) as f: for ln, line in enumerate(f, 1): line = line.strip() if not line: continue try: rows.append(json.loads(line)) except json.JSONDecodeError: print(f"[warn] 第 {ln} 行非 JSON,跳过: {line[:60]}") return rows def parse_duration(s): """'48h' → 48*3600 秒;'7d' → 7*86400;None → 全部。""" if not s: return None s = s.strip().lower() mult = 1 if s.endswith("h"): mult = 3600 s = s[:-1] elif s.endswith("d"): mult = 86400 s = s[:-1] elif s.endswith("m"): mult = 60 s = s[:-1] return float(s) * mult def main(): ap = argparse.ArgumentParser(description="织忆写时四操作影子日志分布报告") ap.add_argument("--log", default=os.getenv("ZHIYI_WRITE_OPS_LOG", "/var/lib/memoryweave/write-ops-shadow.log")) ap.add_argument("--sample", type=int, default=5, help="误判抽样条数(默认 5)") ap.add_argument("--since", default=None, help="时间过滤,如 48h/7d(默认全部)") args = ap.parse_args() rows = load_log(args.log) if not rows: print("影子日志为空 —— 写时四操作尚无判定记录。") return 1 since = parse_duration(args.since) if since is not None: import time cutoff = time.time() - since rows = [r for r in rows if r.get("timestamp", "") >= time.strftime("%Y-%m-%dT%H:%M:%S", time.localtime(cutoff))[:19]] print(f"[filter] 只统计近 {args.since} 的判定") total = len(rows) dist = {"ADD": 0, "UPDATE": 0, "NOOP": 0, "CONFLICT-MARK": 0, "未知": 0} for r in rows: op = r.get("op", "未知") dist[op] = dist.get(op, 0) + 1 print("=" * 52) print(f"写时四操作影子分布报告(共 {total} 条判定)") print("=" * 52) for op in ["ADD", "UPDATE", "NOOP", "CONFLICT-MARK", "未知"]: n = dist.get(op, 0) pct = 100.0 * n / total if total else 0 bar = "#" * int(pct / 2) print(f" {op:<14} {n:>5} {pct:5.1f}% {bar}") update_conflict = dist.get("UPDATE", 0) + dist.get("CONFLICT-MARK", 0) if total: print("-" * 52) print(f"UPDATE+CONFLICT 占比 = {100.0 * update_conflict / total:.1f}% " f"(止损线:误判 >10% 则保持影子不转 apply)") # 误判抽样:优先抽 UPDATE / CONFLICT(高风险操作),按判定时间新→旧 risky = [r for r in rows if r.get("op") in ("UPDATE", "CONFLICT-MARK")] risky.sort(key=lambda r: r.get("timestamp", ""), reverse=True) print("=" * 52) print(f"误判抽样(前 {min(args.sample, len(risky))} 条高风险判定,人工复核)") print("=" * 52) for r in risky[:args.sample]: print(f" [{r.get('op')}] sim={r.get('similarity', 0):.3f} " f"target={r.get('target_id', '')}") print(f" 候选: {r.get('candidate', '')[:70]}") print(f" 理由: {r.get('reason', '')}") print(f" 时间: {r.get('timestamp', '')}") return 0 if __name__ == "__main__": sys.exit(main())