"""
数据流结构分析报告生成器（v3）
读取 analyze_dataflow.py 采集的数据，生成带锚点链接、上下游映射列、
业务描述、多示例值、字段差异报告的 Markdown 报告。
增强内容（v3）：
- 总览表增加 API JSON 字段数列
- 覆盖率表增加业务描述列
- 逐表详情增加业务描述列（来自 BD_manual 文档）
- 说明+示例值合并，多示例展示，枚举值解释
- 总览章节增加 API↔ODS↔DWD 字段对比差异报告
用法:
python scripts/ops/gen_dataflow_report.py
python scripts/ops/gen_dataflow_report.py --output-dir export/dataflow_analysis
"""
from __future__ import annotations
import argparse
import json
import os
from datetime import datetime
from pathlib import Path
from dotenv import load_dotenv
def load_json(path: Path) -> dict | list | None:
if not path.exists():
return None
with open(path, "r", encoding="utf-8") as f:
return json.load(f)
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(description="生成数据流结构分析 Markdown 报告")
parser.add_argument("--output-dir", type=str, default=None,
help="输出目录（默认读取 SYSTEM_ANALYZE_ROOT 或 export/dataflow_analysis）")
return parser
def resolve_data_dir(override: str | None = None) -> Path:
if override:
return Path(override)
env_root = os.environ.get("SYSTEM_ANALYZE_ROOT")
if env_root:
return Path(env_root)
return Path("export/dataflow_analysis")
def _esc(s: str) -> str:
"""转义 Markdown 表格中的管道符"""
return str(s).replace("|", "\\|").replace("\n", " ") if s else ""
# ── 字段用途推测规则 ──
# 基于字段名模式 + 表名上下文推断字段可能的业务含义
# 置信度：高(≥80%) / 中(50-79%) / 低(<50%)
import re as _re
_FIELD_GUESS_RULES: list[tuple[str, str, str]] = [
# (字段名模式正则, 推测用途, 置信度)
# ── SCD2 / ETL 元数据 ──
(r"^scd2_", "SCD2 缓慢变化维度元数据", "高"),
(r"^etl_", "ETL 流程元数据", "高"),
(r"^dw_insert", "数仓装载时间戳", "高"),
(r"^content_hash$", "数据变更检测哈希", "高"),
(r"^source_file$", "ETL 来源文件标识", "高"),
(r"^source_endpoint$", "ETL 来源接口标识", "高"),
(r"^fetched_at$", "ETL 抓取时间", "高"),
(r"^payload$", "原始 JSON 全量存储", "高"),
# ── 主键 / 外键 ──
(r"^id$", "主键标识", "高"),
# ── 门店 / 组织（放在通用 _id$ 之前） ──
(r"^(site_id|shop_id|store_id)$", "门店标识", "高"),
(r"^(tenant_id|org_id)$", "租户/组织标识", "高"),
(r"(shop_name|site_name|store_name)", "门店名称", "高"),
# ── 时间类 ──
(r"(^|_)(create|created)(_at|_time|_date)$", "记录创建时间", "高"),
(r"(^|_)(update|updated|modify)(_at|_time|_date)$", "记录更新时间", "高"),
(r"(^|_)(delete|deleted)(_at|_time|_date)$", "逻辑删除时间", "高"),
(r"(^|_)(start|begin)(_at|_time|_date)$", "起始时间", "中"),
(r"(^|_)(end|expire)(_at|_time|_date)$", "结束/过期时间", "中"),
(r"(^|_)entry_time$", "入职/入场时间", "中"),
(r"(^|_)resign_time$", "离职时间", "中"),
(r"_time$", "时间戳字段", "中"),
(r"_date$", "日期字段", "中"),
# ── 通用派生（放在标志位之前，确保 derived_flag 等优先匹配派生） ──
(r"^derived_", "ETL 派生计算列", "高"),
(r"^calc_", "计算字段", "中"),
# ── 状态 / 标志 ──
(r"(^|_)is_delete$", "逻辑删除标志", "高"),
(r"^is_", "布尔标志位", "中"),
(r"(^|_)status$", "状态码", "中"),
(r"_status$", "状态字段", "中"),
(r"_enabled$", "启用/禁用开关", "中"),
(r"_flag$", "标志位", "中"),
# ── 金额 / 价格 ──
(r"(price|amount|fee|cost|money|balance|total)", "金额/价格相关", "中"),
(r"(discount|coupon|refund)", "优惠/退款相关", "中"),
# ── 人员 ──
(r"(real_name|nickname|^name$)", "姓名/昵称", "中"),
(r"(mobile|phone|tel)", "联系电话", "中"),
(r"(avatar|photo|image)", "头像/图片 URL", "中"),
(r"(gender|sex)", "性别", "高"),
(r"(birth|birthday)", "出生日期", "高"),
(r"(height|weight)", "身高/体重", "高"),
# ── 嵌套对象常见前缀 ──
(r"^siteProfile\.", "门店档案嵌套属性", "高"),
(r"^memberInfo\.", "会员信息嵌套属性", "中"),
(r"^assistantInfo\.", "助教信息嵌套属性", "中"),
(r"^tableInfo\.", "台桌信息嵌套属性", "中"),
(r"^orderInfo\.", "订单信息嵌套属性", "中"),
(r"^payInfo\.", "支付信息嵌套属性", "中"),
# ── 排序 / 显示 ──
(r"(sort|order|rank|seq)", "排序/序号", "低"),
(r"(remark|memo|note|comment|introduce)", "备注/说明文本", "中"),
(r"(url|link|qrcode|qr_code)", "链接/二维码", "中"),
# ── 通用 ID 后缀（放在具体 ID 规则之后） ──
(r"_id$", "关联实体 ID（外键）", "中"),
]
def _guess_field_purpose(field_name: str, table_name: str, layer: str) -> tuple[str, str]:
"""根据字段名和表上下文推测用途，返回 (推测用途, 置信度)。"""
fn_lower = field_name.lower()
for pattern, purpose, confidence in _FIELD_GUESS_RULES:
if _re.search(pattern, fn_lower):
return purpose, confidence
return f"待分析（{layer}层字段）", "低"
def _format_samples(samples: list[str], max_show: int = 5) -> str:
"""格式化多示例值，截断过长的值"""
if not samples:
return ""
shown = []
for s in samples[:max_show]:
s = _esc(s)
if len(s) > 30:
s = s[:27] + "..."
shown.append(f"`{s}`")
result = ", ".join(shown)
if len(samples) > max_show:
result += f" …共{len(samples)}种"
return result
def _is_enum_like(samples: list[str], total_records: int) -> bool:
"""判断字段是否像枚举（不同值少且记录数足够多）"""
if total_records < 5:
return False
return 1 < len(samples) <= 8
def generate_report(data_dir: Path) -> str:
"""生成完整的 Markdown 报告"""
manifest = load_json(data_dir / "collection_manifest.json")
if not manifest:
raise FileNotFoundError(f"找不到 collection_manifest.json: {data_dir}")
tables = manifest["tables"]
now = datetime.now()
lines: list[str] = []
def w(s: str = ""):
lines.append(s)
# ── 报告头 ──
w("# 飞球连接器 — 数据流结构分析报告")
w()
w(f"> 生成时间：{now.strftime('%Y-%m-%d %H:%M:%S')} CST")
w(f"> 分析范围：飞球（feiqiu）连接器，共 {len(tables)} 张 ODS 表")
w("> 数据来源：API JSON 采样 + PostgreSQL ODS/DWD 表结构 + 三层字段映射 + BD_manual 业务文档")
w()
# ── 1. 总览表（增加 API JSON 字段数列） ──
w("## 1. 总览")
w()
w("| # | ODS 表名 | 业务描述 | 采样记录数 | API JSON 字段数 | ODS 列数 | DWD 目标表 | DWD 总列数 |")
w("|---|---------|---------|-----------|---------------|---------|-----------|-----------|")
total_records = 0
total_ods_cols = 0
total_dwd_cols = 0
total_json_fields = 0
for i, t in enumerate(tables, 1):
dwd_names = ", ".join(t["dwd_tables"]) if t["dwd_tables"] else "—"
json_fc = t.get("json_field_count", 0)
w(f"| {i} | `{t['table']}` | {t['description']} | {t['record_count']} | {json_fc} | {t['ods_column_count']} | {dwd_names} | {t['dwd_column_count']} |")
total_records += t["record_count"]
total_ods_cols += t["ods_column_count"]
total_dwd_cols += t["dwd_column_count"]
total_json_fields += json_fc
w(f"| | **合计** | | **{total_records}** | **{total_json_fields}** | **{total_ods_cols}** | | **{total_dwd_cols}** |")
w()
# ── 1.1 字段对比差异报告 ──
_write_field_diff_report(w, data_dir, tables)
# ── 2. 全局统计 ──
w("## 2. 全局统计")
w()
# 2.1 JSON→ODS 映射覆盖
total_json = 0
total_mapped = 0
per_table_stats: list[dict] = []
for t in tables:
fm = load_json(data_dir / "field_mappings" / f"{t['table']}.json")
if not fm or "json_to_ods" not in fm:
per_table_stats.append({
"table": t["table"], "description": t["description"],
"json_count": 0, "mapped": 0, "unmapped": 0, "pct": "—",
})
continue
j2o = fm["json_to_ods"]
json_count = len(j2o)
mapped = sum(1 for m in j2o if m.get("ods_col") is not None)
unmapped = json_count - mapped
pct = f"{mapped / json_count * 100:.1f}%" if json_count > 0 else "—"
per_table_stats.append({
"table": t["table"], "description": t["description"],
"json_count": json_count, "mapped": mapped, "unmapped": unmapped, "pct": pct,
})
total_json += json_count
total_mapped += mapped
total_unmapped = total_json - total_mapped
w("### 2.1 JSON→ODS 映射覆盖")
w()
w(f"- JSON 字段总数：{total_json}")
if total_json > 0:
w(f"- 已映射到 ODS 列：{total_mapped}（{total_mapped / total_json * 100:.1f}%）")
w(f"- 仅存于 payload：{total_unmapped}（{total_unmapped / total_json * 100:.1f}%）")
else:
w("- 已映射到 ODS 列：0")
w("- 仅存于 payload：0")
w()
# 2.2 ODS→DWD 映射覆盖
w("### 2.2 ODS→DWD 映射覆盖")
w()
w(f"- DWD 列总数：{total_dwd_cols}")
w()
# 2.3 各表覆盖率（增加业务描述列）
w("### 2.3 各表 JSON→ODS 映射覆盖率")
w()
w("| ODS 表名 | 业务描述 | JSON 字段数 | 已映射 | 仅 payload | 覆盖率 |")
w("|---------|---------|-----------|-------|-----------|-------|")
sorted_stats = sorted(per_table_stats, key=lambda x: (0 if x["pct"] == "—" else -float(x["pct"].rstrip("%"))))
for s in sorted_stats:
w(f"| `{s['table']}` | {s['description']} | {s['json_count']} | {s['mapped']} | {s['unmapped']} | {s['pct']} |")
w()
# ── 3. 逐表详情 ──
w("## 3. 逐表详情")
w()
for idx, t in enumerate(tables, 1):
table_name = t["table"]
fm = load_json(data_dir / "field_mappings" / f"{table_name}.json")
jt = load_json(data_dir / "json_trees" / f"{table_name}.json")
ods_schema = load_json(data_dir / "db_schemas" / f"ods_{table_name}.json")
bd = load_json(data_dir / "bd_descriptions" / f"{table_name}.json")
# 锚点 ID
anchors = fm.get("anchors", {}) if fm else {}
api_anchor = anchors.get("api", f"api-{table_name}")
ods_anchor = anchors.get("ods", f"ods-{table_name}")
dwd_anchors = anchors.get("dwd", {})
dwd_tables_list = t.get("dwd_tables", [])
json_fc = t.get("json_field_count", 0)
w(f"### 3.{idx} {table_name}（{t['description']}）")
w()
w(f"- 任务代码：`{t['task_code']}`")
w(f"- 采样记录数：{t['record_count']}")
w(f"- API JSON 字段数：{json_fc}")
w(f"- ODS 列数：{t['ods_column_count']}")
if dwd_tables_list:
w(f"- DWD 目标表：{', '.join(dwd_tables_list)}")
else:
w("- DWD 目标表：—（仅 ODS 落地）")
w()
# ── API 源字段区块 ──
_write_api_section(w, fm, jt, bd, table_name, api_anchor, ods_anchor)
# ── ODS 表结构区块 ──
_write_ods_section(w, fm, ods_schema, bd, table_name, ods_anchor, api_anchor, dwd_anchors)
# ── DWD 表结构区块 ──
for dwd_name in dwd_tables_list:
dwd_anchor = dwd_anchors.get(dwd_name, f"dwd-{dwd_name}")
dwd_schema = load_json(data_dir / "db_schemas" / f"dwd_{dwd_name}.json")
_write_dwd_section(w, fm, dwd_schema, bd, dwd_name, dwd_anchor, ods_anchor, table_name)
return "\n".join(lines)
def _write_field_diff_report(w, data_dir: Path, tables: list[dict]):
"""生成 API↔ODS↔DWD 字段对比差异报告（汇总表 + 逐表分表）"""
w("### 1.1 API↔ODS↔DWD 字段对比差异")
w()
w("以下汇总各表在三层之间的字段差异（点击数字跳转至分表详情）：")
w()
w("| ODS 表名 | API→ODS 未映射 | ODS 无 JSON 源 | ODS→DWD 未映射 | DWD 无 ODS 源 | 主要差异原因 |")
w("|---------|--------------|--------------|--------------|-------------|------------|")
# 收集每表差异数据，用于汇总表和分表
etl_meta_cols = {"source_file", "source_endpoint", "fetched_at", "payload", "content_hash"}
diff_rows: list[dict] = []
for t in tables:
table_name = t["table"]
fm = load_json(data_dir / "field_mappings" / f"{table_name}.json")
if not fm:
w(f"| `{table_name}` | — | — | — | — | 无映射数据 |")
diff_rows.append(None)
continue
anchors = fm.get("anchors", {})
api_anchor = anchors.get("api", f"api-{table_name.replace('_', '-')}")
ods_anchor = anchors.get("ods", f"ods-{table_name.replace('_', '-')}")
dwd_anchors = anchors.get("dwd", {})
diff_anchor = f"diff-{table_name.replace('_', '-')}"
j2o = fm.get("json_to_ods", [])
o2d = fm.get("ods_to_dwd", {})
d2o = fm.get("dwd_to_ods", {})
# ── API→ODS 未映射字段 ──
api_unmapped_flat: list[str] = []
api_unmapped_nested: list[str] = []
for m in j2o:
if m.get("ods_col") is None:
jp = m.get("json_path", "")
if "." in jp:
api_unmapped_nested.append(jp)
else:
api_unmapped_flat.append(jp)
api_unmapped_total = len(api_unmapped_flat) + len(api_unmapped_nested)
# ── ODS 无 JSON 源 ──
ods_schema = load_json(data_dir / "db_schemas" / f"ods_{table_name}.json")
ods_mapped_cols = {m["ods_col"] for m in j2o if m.get("ods_col")}
ods_no_json_fields: list[str] = []
if ods_schema and "columns" in ods_schema:
for col in ods_schema["columns"]:
if col["name"] not in ods_mapped_cols and col["name"] not in etl_meta_cols:
ods_no_json_fields.append(col["name"])
# ── ODS→DWD 未映射 ──
ods_cols_with_dwd = set(o2d.keys())
ods_no_dwd_fields: list[str] = []
if ods_schema and "columns" in ods_schema:
for col in ods_schema["columns"]:
if col["name"] not in ods_cols_with_dwd and col["name"] not in etl_meta_cols:
ods_no_dwd_fields.append(col["name"])
# ── DWD 无 ODS 源 ──
dwd_no_ods_fields: list[tuple[str, str]] = [] # (dwd_table, dwd_col)
for dwd_name, entries in d2o.items():
for entry in entries:
if entry.get("ods_source") == "—":
dwd_no_ods_fields.append((dwd_name, entry["dwd_col"]))
# 差异原因
reasons: list[str] = []
if api_unmapped_nested:
reasons.append(f"嵌套对象 {len(api_unmapped_nested)} 个")
if api_unmapped_flat:
reasons.append(f"平层未映射 {len(api_unmapped_flat)} 个")
if dwd_no_ods_fields:
reasons.append(f"SCD2/派生列 {len(dwd_no_ods_fields)} 个")
reason_str = "；".join(reasons) if reasons else "—"
# 汇总表单元格：数量 + 跳转链接
def _cell(count: int) -> str:
if count == 0:
return "0"
return f"[{count}](#{diff_anchor})"
w(f"| `{table_name}` | {_cell(api_unmapped_total)} | {_cell(len(ods_no_json_fields))} | {_cell(len(ods_no_dwd_fields))} | {_cell(len(dwd_no_ods_fields))} | {reason_str} |")
diff_rows.append({
"table_name": table_name,
"diff_anchor": diff_anchor,
"api_anchor": api_anchor,
"ods_anchor": ods_anchor,
"dwd_anchors": dwd_anchors,
"api_unmapped_flat": api_unmapped_flat,
"api_unmapped_nested": api_unmapped_nested,
"ods_no_json_fields": ods_no_json_fields,
"ods_no_dwd_fields": ods_no_dwd_fields,
"dwd_no_ods_fields": dwd_no_ods_fields,
})
w()
# ── 逐表差异分表 ──
sub_idx = 0
for row in diff_rows:
if row is None:
continue
has_any = (row["api_unmapped_flat"] or row["api_unmapped_nested"]
or row["ods_no_json_fields"] or row["ods_no_dwd_fields"]
or row["dwd_no_ods_fields"])
if not has_any:
continue
sub_idx += 1
table_name = row["table_name"]
w(f'')
w()
w(f"#### 1.1.{sub_idx} {table_name} 字段差异明细")
w()
api_anchor = row["api_anchor"]
ods_anchor = row["ods_anchor"]
dwd_anchors = row["dwd_anchors"]
# 加载辅助数据：json_trees（示例值）、bd_descriptions（业务说明）
jt = load_json(data_dir / "json_trees" / f"{table_name}.json")
bd = load_json(data_dir / "bd_descriptions" / f"{table_name}.json")
jt_lookup: dict[str, dict] = {}
if jt and "fields" in jt:
for fld in jt["fields"]:
jt_lookup[fld["path"]] = fld
ods_descs = bd.get("ods_fields", {}) if bd else {}
dwd_descs_all = bd.get("dwd_fields", {}) if bd else {}
def _sample_str(field_name: str, layer: str, dwd_tbl: str = "") -> str:
"""从 json_trees 或 bd_descriptions 获取示例值字符串"""
if layer == "API":
entry = jt_lookup.get(field_name, {})
samples = entry.get("samples", [])
total_recs = entry.get("total_records", 0)
if not samples:
single = entry.get("sample", "")
if single:
samples = [str(single)]
if _is_enum_like(samples, total_recs):
return ", ".join(f"`{_esc(s)}`" for s in samples[:5])
if samples:
return _format_samples(samples, max_show=3)
return ""
def _desc_str(field_name: str, layer: str, dwd_tbl: str = "") -> str:
"""从 bd_descriptions 获取业务说明"""
key = field_name.split(".")[-1].replace("[]", "").lower()
if layer in ("ODS", "API"):
desc = ods_descs.get(key, "")
elif layer == "DWD" and dwd_tbl:
desc = dwd_descs_all.get(dwd_tbl, {}).get(key, "")
else:
desc = ""
if desc and len(desc) > 40:
desc = desc[:37] + "..."
return _esc(desc)
# ── API→ODS 未映射（平层） ──
if row["api_unmapped_flat"]:
w(f"**API→ODS 未映射（平层）** — {len(row['api_unmapped_flat'])} 个")
w()
w("| # | JSON 字段 | 推测用途 | 置信度 | 示例值 | 说明 | 状态 |")
w("|---|----------|---------|-------|-------|------|------|")
for i, f in enumerate(row["api_unmapped_flat"], 1):
purpose, conf = _guess_field_purpose(f, table_name, "API")
sample = _sample_str(f, "API")
desc = _desc_str(f, "API")
w(f"| {i} | **[`{_esc(f)}`](#{api_anchor})** | {_esc(purpose)} | {conf} | {sample} | {desc} | **⚠️ 未映射** |")
w()
# ── API→ODS 未映射（嵌套对象） ──
if row["api_unmapped_nested"]:
w(f"API→ODS 未映射（嵌套对象）— {len(row['api_unmapped_nested'])} 个
")
w()
w("| # | JSON 字段 | 推测用途 | 置信度 | 示例值 | 说明 | 状态 |")
w("|---|----------|---------|-------|-------|------|------|")
for i, f in enumerate(row["api_unmapped_nested"], 1):
purpose, conf = _guess_field_purpose(f, table_name, "API")
sample = _sample_str(f, "API")
desc = _desc_str(f, "API")
w(f"| {i} | [`{_esc(f)}`](#{api_anchor}) | {_esc(purpose)} | {conf} | {sample} | {desc} | 📦 嵌套 |")
w()
w("