feat(release): 发布前快照硬门禁 + 云内快照 + ES/OSS 保护

按生产级备份组合落地:

- internal/releasesnapshot:新增门禁包,要求 root 下存在覆盖目标库、未过期
  (默认 24h)、含 wiki_pages.jsonl 与 sha256 的快照;本地文件再校验哈希。
  7 条单测(过期/换库/篡改/缺页/未接线/半成品)。
- 发布器:写目标库之前(Publish/Delete/Rollback 三条路)先过门禁;映射目标
  (源库→目标库,共享真实库)或 policy=always 时必须有快照;门禁未接线
  同样拒绝(fail-closed)。新错误码 release_target_snapshot_missing(不可重试)
  + 页面中文提示。6 条发布器单测。
- worker 配置新增 target_snapshot{root,max_age_hours,policy},校验 policy/
  max_age/绝对路径;compose.release-sim.yaml 只读挂载快照目录。
- scripts/target_kb_snapshot.sh:默认云内执行(与 DB 同 VPC,凭据取自 WeKnora
  容器)+ 默认不导 chunks(44s 出 8 张表);修复 \copy 不能跨行/不能带 WITH、
  ES 计数改用 knowledge_base_id.keyword。
- ES:注册 kb_backup 仓库(path.repo),首个快照 36s/4.7GB/0 分片失败,
  SLM kb-hourly(3d)+kb-daily(7d) 已建并执行成功。
- OSS:目标桶开启版本控制(此前未开启)。
- 文档 P6 增补硬门禁与云内执行;真实快照目录上门禁实测:两个绑定库通过、
  未覆盖库拒绝、未接线拒绝。
This commit is contained in:
姚章浩
2026-10-09 10:59:55 +08:00
parent f4e1ecc5f1
commit 1fa35e9870
14 changed files with 836 additions and 70 deletions
+80
View File
@@ -0,0 +1,80 @@
#!/usr/bin/env python3
"""目标 OSS 桶的保护设置:读取状态,或开启版本控制 / 跨区域复制检查。
只处理目标环境的桶(secrets/release-sim 下的 key 文件),凭据不打印。
用法:
python3 scripts/oss_bucket_settings.py status # 版本控制 / 复制 / 生命周期
python3 scripts/oss_bucket_settings.py enable-versioning # 写操作:开启版本控制
"""
import base64
import hashlib
import hmac
import pathlib
import sys
import urllib.error
import urllib.request
from email.utils import formatdate
ROOT = pathlib.Path(__file__).resolve().parents[1]
SECRETS = ROOT / "secrets" / "release-sim"
BUCKET = "dobot-weknora-test"
ENDPOINT = "oss-cn-shenzhen.aliyuncs.com"
def read_secret(name):
return (SECRETS / name).read_text().strip()
def request(method, subresource, body=""):
access_key = read_secret("target_object_access_key")
secret_key = read_secret("target_object_secret_key")
date = formatdate(usegmt=True)
content_type = "application/xml" if body else ""
canonical = f"/{BUCKET}/?{subresource}"
string_to_sign = f"{method}\n\n{content_type}\n{date}\n{canonical}"
signature = base64.b64encode(
hmac.new(secret_key.encode(), string_to_sign.encode(), hashlib.sha1).digest()
).decode()
headers = {"Date": date, "Authorization": f"OSS {access_key}:{signature}"}
if content_type:
headers["Content-Type"] = content_type
http_request = urllib.request.Request(
f"https://{BUCKET}.{ENDPOINT}/?{subresource}",
data=body.encode() if body else None, headers=headers, method=method)
try:
with urllib.request.urlopen(http_request, timeout=20) as response:
return response.status, response.read().decode("utf-8", "replace")
except urllib.error.HTTPError as error:
return error.code, error.read().decode("utf-8", "replace")
def versioning_state():
status, body = request("GET", "versioning")
if status != 200:
return f"HTTP {status}"
return "Enabled" if "<Status>Enabled</Status>" in body else "Disabled(未开启)"
def main():
action = sys.argv[1] if len(sys.argv) > 1 else "status"
if action == "status":
print("versioning:", versioning_state())
for name in ("replication", "lifecycle"):
status, body = request("GET", name)
summary = "none" if status == 404 else " ".join(body.split())[:120]
print(f"{name}: HTTP {status} {summary}")
return
if action == "enable-versioning":
print("before:", versioning_state())
body = "<VersioningConfiguration><Status>Enabled</Status></VersioningConfiguration>"
status, response = request("PUT", "versioning", body)
print("PUT versioning:", f"HTTP {status}", " ".join(response.split())[:160])
print("after:", versioning_state())
return
print("unknown action:", action, file=sys.stderr)
sys.exit(2)
if __name__ == "__main__":
main()
+126 -60
View File
@@ -5,32 +5,47 @@
# 只做 SELECT + \copy,不写目标库任何一行。
#
# 用法:
# bash scripts/target_kb_snapshot.sh # 用 production_kb_bindings 里已绑定的目标库
# bash scripts/target_kb_snapshot.sh <kb_id> [<kb_id>] # 指定目标库
# bash scripts/target_kb_snapshot.sh # 已绑定的目标库;云内执行(快)
# EXECUTION=tunnel bash scripts/target_kb_snapshot.sh # 走本机 SSH 隧道(慢,兜底)
# bash scripts/target_kb_snapshot.sh <kb_id> [<kb_id>] # 指定目标库
# INCLUDE_CHUNKS=1 bash scripts/target_kb_snapshot.sh # 连 chunks 一起导(大,按需)
# PULL_CHUNKS=1 bash scripts/target_kb_snapshot.sh # 云内执行时把 chunks 结果也拉回本机
#
# 环境变量:
# EXECUTION cloud(默认,与 DB 同 VPC 的云主机上跑)| tunnel(本机经 SSH 隧道)
# SNAPSHOT_ROOT 输出根目录(默认 backups/target-kb-snapshots)
# TUNNEL_HOST 目标 PG 隧道地址(默认 172.17.0.1)
# TUNNEL_PORT 目标 PG 隧道端口(默认 25432)
# INCLUDE_CHUNKS 1=连 chunks 一起导出(较大,默认 1)
# INCLUDE_CHUNKS 1=连 chunks 一起导(默认 0:秒级完成;向量在 ES 侧,行可从源端重发)
# PULL_CHUNKS 1=云内执行时把 chunks 结果也拉回本机(默认 0,留在云主机)
# TUNNEL_HOST/TUNNEL_PORT 隧道模式的目标 PG 地址(默认 172.17.0.1:25432)
# SSH_CONFIG SSH 配置路径(默认 ~/.ssh/config.weknora-cloud-test)
# SSH_HOST 云主机别名(默认 weknora-test)
# APP_CONTAINER 云上提供 DB 凭据的容器(默认 weknora-test-v071-app-1)
# REMOTE_ROOT 云上快照目录(默认 /opt/weknora-test/kb-snapshots)
# WITH_ES 1=同时记录目标 ES 的文档数(默认 1)
# NOTE 附加到清单的备注(例如发布单号)
#
# 覆盖范围(PG 行级):knowledge_bases / knowledges / knowledge_tags /
# knowledge_tag_relations / chunks / resources / resource_bindings /
# wiki_pages / wiki_folders(含软删行,不含表结构)。
# knowledge_tag_relations / wiki_pages / wiki_folders / resources /
# resource_bindings(+ chunks,可选)。含软删行,不含表结构。
#
# 不覆盖:Elasticsearch 文档内容与向量、OSS 对象本体(只在 resources.jsonl 里
# 记路径/大小/哈希做清单)、表结构与索引。整库级恢复请用阿里云 RDS 快照/PITR。
# 不覆盖:Elasticsearch 文档与向量(见 ES 仓库 kb_backup 与 SLM 策略 kb-hourly/kb-daily)、
# OSS 对象本体(只在 resources.jsonl 里记路径/大小/哈希做清单)、表结构与索引。
# 整库级恢复请用阿里云 RDS 快照/PITR。
set -euo pipefail
WORKDIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
cd "$WORKDIR"
EXECUTION="${EXECUTION:-cloud}"
SNAPSHOT_ROOT="${SNAPSHOT_ROOT:-$WORKDIR/backups/target-kb-snapshots}"
TUNNEL_HOST="${TUNNEL_HOST:-172.17.0.1}"
TUNNEL_PORT="${TUNNEL_PORT:-25432}"
INCLUDE_CHUNKS="${INCLUDE_CHUNKS:-1}"
INCLUDE_CHUNKS="${INCLUDE_CHUNKS:-0}"
PULL_CHUNKS="${PULL_CHUNKS:-0}"
SSH_CONFIG="${SSH_CONFIG:-$HOME/.ssh/config.weknora-cloud-test}"
SSH_HOST="${SSH_HOST:-weknora-test}"
APP_CONTAINER="${APP_CONTAINER:-weknora-test-v071-app-1}"
REMOTE_ROOT="${REMOTE_ROOT:-/opt/weknora-test/kb-snapshots}"
WITH_ES="${WITH_ES:-1}"
NOTE="${NOTE:-}"
@@ -53,34 +68,15 @@ done
STAMP="$(date -u +%Y%m%dT%H%M%SZ)"
PREFIX="$(printf '%s' "${KB_IDS[0]}" | cut -c1-8)"
OUT="$SNAPSHOT_ROOT/${STAMP}-${PREFIX}${PREFIX:+_}"
OUT="${OUT%_}"
OUT="$SNAPSHOT_ROOT/${STAMP}-${PREFIX}"
mkdir -p "$OUT"
# ---------- 连接串(不打印) ----------
SECRET_FILE="secrets/release-sim/target_database_url"
[ -f "$SECRET_FILE" ] || { echo "missing $SECRET_FILE"; exit 1; }
PGURL="$(python3 - "$SECRET_FILE" "$TUNNEL_HOST" "$TUNNEL_PORT" <<'PY'
import re, sys, pathlib
url = pathlib.Path(sys.argv[1]).read_text().strip()
url = re.sub(r'@[^/@]+:\d+/', '@%s:%s/' % (sys.argv[2], sys.argv[3]), url)
print(url)
PY
)"
ENDPOINT="$TUNNEL_HOST:$TUNNEL_PORT ($(python3 - "$SECRET_FILE" <<'PY'
import sys, pathlib, re, urllib.parse
url = pathlib.Path(sys.argv[1]).read_text().strip()
parsed = urllib.parse.urlparse(url)
print("host=%s db=%s" % (parsed.hostname or "?", (parsed.path or "/?").lstrip("/") or "?"))
PY
))"
# ---------- 生成导出脚本(本身也是留档的一部分) ----------
KB_SQL="'$(printf '%s' "${KB_IDS[*]}" | sed "s/ /','/g")'"
# ---------- 生成导出脚本(本身也是留档的一部分) ----------
cat >"$OUT/export.sql" <<SQL
-- 目标知识库只读行级快照:$STAMP
-- 生成者:scripts/target_kb_snapshot.sh
-- 生成者:scripts/target_kb_snapshot.sh(EXECUTION=$EXECUTION)
-- 目标库:${KB_IDS[*]}
-- 说明:全部为 SELECT + \copy,不修改目标库。
\pset pager off
@@ -92,8 +88,10 @@ cat >"$OUT/export.sql" <<SQL
\copy (SELECT to_jsonb(r) FROM knowledge_tag_relations r WHERE r.knowledge_id IN (SELECT id FROM knowledges WHERE knowledge_base_id IN ($KB_SQL))) TO 'out/knowledge_tag_relations.jsonl'
\copy (SELECT to_jsonb(t) FROM wiki_pages t WHERE t.knowledge_base_id IN ($KB_SQL)) TO 'out/wiki_pages.jsonl'
\copy (SELECT to_jsonb(t) FROM wiki_folders t WHERE t.knowledge_base_id IN ($KB_SQL)) TO 'out/wiki_folders.jsonl'
\copy (SELECT to_jsonb(b) FROM resource_bindings b WHERE b.owner_id IN (SELECT id FROM knowledges WHERE knowledge_base_id IN ($KB_SQL)) OR b.owner_id IN (SELECT id FROM chunks WHERE knowledge_base_id IN ($KB_SQL))) TO 'out/resource_bindings.jsonl'
\copy (SELECT to_jsonb(r) FROM resources r WHERE r.id IN (SELECT resource_id FROM resource_bindings WHERE owner_id IN (SELECT id FROM knowledges WHERE knowledge_base_id IN ($KB_SQL)) OR owner_id IN (SELECT id FROM chunks WHERE knowledge_base_id IN ($KB_SQL)))) TO 'out/resources.jsonl'
-- 归属集合用 UNION 子查询(云上实测 0.6~0.8s;OR+IN 会退化成慢查询)。
-- 注意:psql 的 \copy 是单行元命令,且不接受 WITH 开头,整条语句必须写成一行。
\copy (SELECT to_jsonb(b) FROM resource_bindings b WHERE b.owner_id IN (SELECT id FROM knowledges WHERE knowledge_base_id IN ($KB_SQL) UNION SELECT id FROM chunks WHERE knowledge_base_id IN ($KB_SQL))) TO 'out/resource_bindings.jsonl'
\copy (SELECT to_jsonb(r) FROM resources r WHERE r.id IN (SELECT b.resource_id FROM resource_bindings b WHERE b.owner_id IN (SELECT id FROM knowledges WHERE knowledge_base_id IN ($KB_SQL) UNION SELECT id FROM chunks WHERE knowledge_base_id IN ($KB_SQL)))) TO 'out/resources.jsonl'
SQL
if [ "$INCLUDE_CHUNKS" = "1" ]; then
cat >>"$OUT/export.sql" <<SQL
@@ -101,61 +99,130 @@ if [ "$INCLUDE_CHUNKS" = "1" ]; then
SQL
fi
# ---------- 在容器里执行(只读导出到挂载目录) ----------
REMOTE_DIR=""
if [ "$EXECUTION" = "cloud" ]; then
# ---------- 云内执行:与 DB 同 VPC,内网直连,不受家用带宽限制 ----------
REMOTE_DIR="$REMOTE_ROOT/$STAMP-${PREFIX}"
ssh -F "$SSH_CONFIG" "$SSH_HOST" "mkdir -p '$REMOTE_DIR'"
scp -q -F "$SSH_CONFIG" "$OUT/export.sql" "$SSH_HOST:$REMOTE_DIR/export.sql"
ssh -F "$SSH_CONFIG" "$SSH_HOST" \
APP_CONTAINER="$APP_CONTAINER" REMOTE_DIR="$REMOTE_DIR" INCLUDE_CHUNKS="$INCLUDE_CHUNKS" \
'bash -s' <<'REMOTE'
set -euo pipefail
env_value() { docker inspect "$APP_CONTAINER" --format '{{range .Config.Env}}{{println .}}{{end}}' | grep -E "^$1=" | head -1 | cut -d= -f2-; }
DB_HOST_VALUE=$(env_value DB_HOST); DB_PORT_VALUE=$(env_value DB_PORT)
DB_NAME_VALUE=$(env_value DB_NAME); DB_USER_VALUE=$(env_value DB_USER)
DB_PASS_VALUE=$(env_value DB_PASSWORD)
PGURL=$(python3 - "$DB_USER_VALUE" "$DB_PASS_VALUE" "$DB_HOST_VALUE" "$DB_PORT_VALUE" "$DB_NAME_VALUE" <<'PY'
import sys, urllib.parse as u
user, password, host, port, name = sys.argv[1:6]
print(f"postgres://{u.quote(user)}:{u.quote(password)}@{host}:{port}/{name}")
PY
)
docker run --rm --network host -e PGURL="$PGURL" \
-v "$OUT:/out" -v "$OUT/export.sql:/scripts/export.sql:ro" postgres:16 \
-v "$REMOTE_DIR:/out" -v "$REMOTE_DIR/export.sql:/scripts/export.sql:ro" postgres:16-alpine \
sh -c 'psql "$PGURL" -q -f /scripts/export.sql' >/dev/null
rm -f "$REMOTE_DIR/export.sql"
if [ "$INCLUDE_CHUNKS" = "1" ] && [ -f "$REMOTE_DIR/chunks.jsonl" ]; then
gzip -f "$REMOTE_DIR/chunks.jsonl"
fi
python3 - "$REMOTE_DIR" <<'PY'
import hashlib, json, pathlib, sys
out = pathlib.Path(sys.argv[1]); files = []
for path in sorted(out.glob("*.jsonl*")):
digest = hashlib.sha256(); size = 0
with path.open("rb") as handle:
for block in iter(lambda: handle.read(1 << 20), b""):
digest.update(block); size += len(block)
files.append({"file": path.name, "bytes": size, "sha256": digest.hexdigest()})
(out / "REMOTE_MANIFEST.json").write_text(json.dumps({"host_files": files}, indent=2) + "\n")
print("remote files:", " ".join(f"{f['file']}={f['bytes']}" for f in files))
PY
exit 0
REMOTE
# 小表拉回本机;chunks 默认留在云主机(PULL_CHUNKS=1 才拉)
scp -q -F "$SSH_CONFIG" "$SSH_HOST:$REMOTE_DIR/*.jsonl" "$OUT/" 2>/dev/null || true
scp -q -F "$SSH_CONFIG" "$SSH_HOST:$REMOTE_DIR/REMOTE_MANIFEST.json" "$OUT/" 2>/dev/null || true
if [ "$PULL_CHUNKS" = "1" ]; then
scp -q -F "$SSH_CONFIG" "$SSH_HOST:$REMOTE_DIR/chunks.jsonl.gz" "$OUT/" 2>/dev/null || true
fi
else
# ---------- 隧道模式(兜底):本机经 SSH 隧道直连 ----------
SECRET_FILE="secrets/release-sim/target_database_url"
[ -f "$SECRET_FILE" ] || { echo "missing $SECRET_FILE"; exit 1; }
PGURL="$(python3 - "$SECRET_FILE" "$TUNNEL_HOST" "$TUNNEL_PORT" <<'PY'
import re, sys, pathlib
url = pathlib.Path(sys.argv[1]).read_text().strip()
print(re.sub(r'@[^/@]+:\d+/', '@%s:%s/' % (sys.argv[2], sys.argv[3]), url))
PY
)"
docker run --rm --network host -e PGURL="$PGURL" \
-v "$OUT:/out" -v "$OUT/export.sql:/scripts/export.sql:ro" postgres:16 \
sh -c 'psql "$PGURL" -q -f /scripts/export.sql' >/dev/null
fi
# ---------- ES 文档数(只读,失败不阻塞) ----------
ES_SUMMARY=""
if [ "$WITH_ES" = "1" ]; then
ES_URL="http://${TUNNEL_HOST}:29200"
for id in "${KB_IDS[@]}"; do
count="$(curl -s --max-time 15 -H 'Content-Type: application/json' \
-X POST "$ES_URL/weknora_test_chunks/_count" \
-d "{\"query\":{\"term\":{\"knowledge_base_id\":\"$id\"}}}" \
-X POST "http://${TUNNEL_HOST}:29200/weknora_test_chunks/_count" \
-d "{\"query\":{\"term\":{\"knowledge_base_id.keyword\":\"$id\"}}}" \
| python3 -c 'import json,sys; print(json.load(sys.stdin).get("count","?"))' 2>/dev/null || echo "?")"
ES_SUMMARY="${ES_SUMMARY}${id}=${count} "
done
fi
# ---------- 清单 + 说明 ----------
python3 - "$OUT" "$STAMP" "$ENDPOINT" "$INCLUDE_CHUNKS" "$ES_SUMMARY" "$NOTE" "${KB_IDS[@]}" <<'PY'
python3 - "$OUT" "$STAMP" "$EXECUTION" "$INCLUDE_CHUNKS" "$ES_SUMMARY" "$NOTE" "$REMOTE_DIR" "${KB_IDS[@]}" <<'PY'
import hashlib, json, pathlib, sys
out = pathlib.Path(sys.argv[1]); stamp = sys.argv[2]; endpoint = sys.argv[3]
out = pathlib.Path(sys.argv[1]); stamp, execution = sys.argv[2], sys.argv[3]
include_chunks = sys.argv[4] == "1"; es_summary = sys.argv[5].strip(); note = sys.argv[6]
kb_ids = sys.argv[7:]
remote_dir = sys.argv[7]; kb_ids = sys.argv[8:]
files = []
for path in sorted(out.glob("*.jsonl")):
digest = hashlib.sha256()
with path.open("rb") as handle:
for block in iter(lambda: handle.read(1 << 20), b""):
digest.update(block)
rows = sum(1 for _ in path.open("rb"))
files.append({"file": path.name, "rows": rows, "bytes": path.stat().st_size,
"sha256": digest.hexdigest()})
files.append({"file": path.name, "rows": sum(1 for _ in path.open("rb")),
"bytes": path.stat().st_size, "sha256": digest.hexdigest(), "location": "local"})
for path in sorted(out.glob("chunks.jsonl.gz")):
digest = hashlib.sha256()
with path.open("rb") as handle:
for block in iter(lambda: handle.read(1 << 20), b""):
digest.update(block)
files.append({"file": path.name, "bytes": path.stat().st_size,
"sha256": digest.hexdigest(), "location": "local"})
remote_manifest = out / "REMOTE_MANIFEST.json"
if remote_manifest.exists():
for item in json.loads(remote_manifest.read_text()).get("host_files", []):
if not any(existing["file"] == item["file"] for existing in files):
item["location"] = f"cloud:{remote_dir}"
files.append(item)
manifest = {
"kind": "weknora-target-kb-snapshot",
"created_at": stamp,
"target_endpoint": endpoint,
"execution": execution,
"knowledge_base_ids": kb_ids,
"profile": "kb-rows" + ("" if include_chunks else " (chunks excluded)"),
"profile": "kb-rows" + ("" if include_chunks else " (chunks excluded: 向量在 ES 侧,行可从源端重发)"),
"covers": ["knowledge_bases", "knowledges", "knowledge_tags", "knowledge_tag_relations",
"wiki_pages", "wiki_folders", "resource_bindings", "resources"]
+ (["chunks"] if include_chunks else []),
"does_not_cover": ["Elasticsearch documents and vectors", "OSS object bodies (listed by reference in resources.jsonl)",
"table structure / indexes", "rows outside the listed knowledge bases"],
"does_not_cover": ["Elasticsearch documents and vectors (see ES repository kb_backup, SLM kb-hourly/kb-daily)",
"OSS object bodies (listed by reference in resources.jsonl)",
"table structure / indexes", "rows outside the listed knowledge bases",
"whole-database recovery (use Aliyun RDS snapshot / PITR)"],
"es_document_counts": es_summary or None,
"remote_directory": remote_dir or None,
"note": note or None,
"files": files,
}
(out / "MANIFEST.json").write_text(json.dumps(manifest, ensure_ascii=False, indent=2) + "\n")
summary = " ".join(f"{item['file']}={item['rows']}" for item in files)
print("snapshot:", out)
print("files:", summary)
print("bytes:", sum(item["bytes"] for item in files))
if es_summary:
print("es docs:", es_summary)
print("files:", " ".join(f"{item['file']}={item.get('rows', '-')}" for item in files))
print("local bytes:", sum(item["bytes"] for item in files if item.get("location") == "local"))
print("es docs:", es_summary or "-")
PY
cat >"$OUT/README.md" <<'MD'
@@ -163,12 +230,12 @@ cat >"$OUT/README.md" <<'MD'
## 它是什么
发布 / 撤权 / 回滚**之前**的目标库现场:KB 维度的 PG 行(含软删行),每张表一个
JSONL(`to_jsonb(row)` 一行一条),附 `MANIFEST.json`(行数 / 字节 / sha256 / 覆盖范围)
与 `export.sql`(导出用的原始查询)。生成过程只有 SELECT + `\copy`,不修改目标库。
JSONL(`to_jsonb(row)` 一行一条),附 `MANIFEST.json`(行数 / 字节 / sha256 / 覆盖范围 /
位置)与 `export.sql`(导出用的原始查询)。生成过程只有 SELECT + `\copy`,不修改目标库。
## 它不是什么
- 不含 Elasticsearch 文档与向量、不含 OSS 对象本体(`resources.jsonl` 里只有
路径 / 大小 / 哈希清单)、不含表结构与索引。
- 不含 Elasticsearch 文档与向量(见 ES 仓库 `kb_backup` 与 SLM 策略 `kb-hourly`/`kb-daily`)、
不含 OSS 对象本体(`resources.jsonl` 里只有路径 / 大小 / 哈希清单)、不含表结构与索引。
- 不是整库级灾备。**整库级恢复请用阿里云 RDS 的快照 / 按时间点恢复(PITR)**:
恢复到临时实例后,再按 KB 导出需要的行,这样连索引和约束一起回到某个时间点。
@@ -176,12 +243,11 @@ JSONL(`to_jsonb(row)` 一行一条),附 `MANIFEST.json`(行数 / 字节
先确认目标库当前状态已被另一份快照覆盖(避免二次覆盖),然后按外键顺序灌回:
```bash
URL=<目标库连接串> # 例如经隧道:postgres://...@172.17.0.1:25432/<db>
URL=<目标库连接串>
OUT=<本快照目录>
docker run --rm --network host -e PGURL="$URL" -v "$OUT:/snap:ro" postgres:16 \
psql "$PGURL" -v ON_ERROR_STOP=1 <<'SQL'
BEGIN;
-- 每张表:把 JSONL 灌进临时表,再按类型还原成行
DO $$
DECLARE t text;
BEGIN