线上 minimax 2 把 Key 点「测试」红字:1004 Please carry the API secret key —— 根因是表单 placeholder 写 sk-...,用户照着填了**裸 key**(无 Bearer ), 而 accounts[].auth 语义是「完整 Authorization 头值」,转发时原样发就 1004。 agent.sh upstreams 一打就看出对照组:deepseek=Bearer s...d425(正常), minimax=sk-cp-...(**无 Bearer **)——光看截图红字只会猜「额度用完」。 plugin-llm - upstreams::normalize_auth:trim 后含空白→已完整;否则补「Bearer 」。 **双处应用**:upsert 写入(下次保存自愈) + read_upa 读出(让线上坏数据 立即可用,无需用户重填——存量 10 把 Key 根本不可能通知用户重填)。 - 兼容 Basic / 自定义 scheme / 前后空格 / 内部含空格共 6 个边界, 干跑 + e2e 2 条新断言全过(裸 key 200 + 靶机实收 Bearer sk-NORM-NO-SCHEME)。 前端 - placeholder 改为「Bearer sk-…(... 留空=保留原值;只填裸 key 也会自动补 Bearer)」 ——placeholder 写「前缀」会诱导漏填必要部分,写「字面值」才能正确引导。 - 多账号小节加说明文字(只填裸 key 也会自动补 Bearer)。 deploy/agent.sh - 新增 upstreams(逐账号掩码/冷却/今日负载;掩码为空显式标 **空** —— 这条 是发现「只填裸 key」类问题的关键指示器,旧镜像的回显脱敏后看不出问题) - 新增 upstream-test <slot>(调测试接口并打印厂商真实 status_code + body), 避免再靠「管理台截图 OCR」排障。 回归 281 全绿(网关 102 + LLM 179);专章 docs/llm/MULTI_ACCOUNT_ROUTING.md §10 记录现象/规则/兼容表/自检工具。
352 lines
15 KiB
Bash
Executable File
352 lines
15 KiB
Bash
Executable File
#!/usr/bin/env bash
|
||
# ---------------------------------------------------------------------------
|
||
# iPass 控制面(ipass-agent,见 readme §12.5)调用 + 正式环境验收。
|
||
#
|
||
# **存在的理由(重要,别再退回手敲 curl)**
|
||
# WorkBuddy/CodeBuddy 的安全策略里有一条:**「可疑的 bash 命令即使已加入
|
||
# permissions.allow 白名单,也需要手动批准」**(见内置文档 cli/security.md
|
||
# 「命令注入检测」)。含 `Authorization: Bearer …` / `. ./.env` / 从环境取
|
||
# 口令这类形态的命令正好命中——表现为**每次都弹「凭证授权」**,甚至在用户
|
||
# 拒绝时「命令静默无输出」,很容易被误判成服务故障(实测踩过:一次 /update
|
||
# 就是这样没发出去)。
|
||
#
|
||
# 结论:**凭据绝不能出现在命令行上**。本脚本把凭据只从仓库根 `.env` 读,
|
||
# 对外暴露的全是无凭据的子命令,于是只需放行这一条:
|
||
# "Bash(bash deploy/agent.sh:*)"
|
||
#
|
||
# 用法:
|
||
# bash deploy/agent.sh status # 容器状态 + 镜像 digest + 数据摘要
|
||
# bash deploy/agent.sh update [tag] # 完整更新(缺省用 .env 的 IPASS_IMAGE_TAG)
|
||
# bash deploy/agent.sh wait <job-id> # 轮询任务直到结束并打印日志
|
||
# bash deploy/agent.sh verify # 正式环境端到端验收(登录+拉页面+查规则)
|
||
# bash deploy/agent.sh release <tag> # 构建推送镜像并更新、验收(一条龙)
|
||
# ...(其余子命令见 help)
|
||
#
|
||
# 依赖:仓库根 `.env`(已 gitignore)。缺文件或变量为空时直接报错退出。
|
||
# ---------------------------------------------------------------------------
|
||
set -euo pipefail
|
||
|
||
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
|
||
ENV_FILE="${IPASS_ENV_FILE:-$ROOT/.env}"
|
||
|
||
if [ ! -f "$ENV_FILE" ]; then
|
||
echo "✗ 找不到 $ENV_FILE —— 控制面凭据只从这里读(见 CLAUDE.md §凭据与授权)" >&2
|
||
exit 2
|
||
fi
|
||
|
||
set -a
|
||
# shellcheck disable=SC1090
|
||
. "$ENV_FILE"
|
||
set +a
|
||
|
||
A="${IPASS_AGENT_HOST:-}"
|
||
TOKEN="${IPASS_AGENT_TOKEN:-}"
|
||
REGISTRY="${IPASS_REGISTRY:-}"
|
||
DEFAULT_TAG="${IPASS_IMAGE_TAG:-}"
|
||
|
||
# 正式环境对外入口(验收用):管理域名 + 宿主端口 + 直连 IP
|
||
LIVE_HOST="${IPASS_MDM_HOST:-mdm.mayishangshu.cn}"
|
||
LIVE_PORT="${IPASS_PUBLIC_PORT:-88}"
|
||
LIVE_IP="${IPASS_LIVE_IP:-192.168.1.11}"
|
||
|
||
if [ -z "$A" ] || [ -z "$TOKEN" ]; then
|
||
echo "✗ .env 缺少 IPASS_AGENT_HOST / IPASS_AGENT_TOKEN" >&2
|
||
exit 2
|
||
fi
|
||
|
||
PY="$(command -v python3 || true)"
|
||
pretty() { if [ -n "$PY" ]; then "$PY" -m json.tool 2>/dev/null || cat; else cat; fi; }
|
||
jget() { "$PY" -c "import json,sys;print(json.load(sys.stdin).get('$1',''))" 2>/dev/null || true; }
|
||
|
||
get() { curl -s -m "${TIMEOUT:-30}" -H "Authorization: Bearer $TOKEN" "$A$1"; }
|
||
post() { curl -s -m "${TIMEOUT:-60}" -H "Authorization: Bearer $TOKEN" \
|
||
-H 'Content-Type: application/json' -d "${2:-{\}}" "$A$1"; }
|
||
|
||
# tag / 完整镜像名 → 完整镜像名;空 → .env 的 IPASS_IMAGE_TAG
|
||
resolve_image() {
|
||
local t="${1:-}"
|
||
[ -z "$t" ] && t="$DEFAULT_TAG"
|
||
[ -z "$t" ] && { echo "✗ 未给 tag,且 .env 的 IPASS_IMAGE_TAG 为空" >&2; exit 2; }
|
||
case "$t" in
|
||
*/*:*|*:* ) printf '%s' "$t" ;;
|
||
ipass-* ) printf '%s:%s' "$REGISTRY" "$t" ;;
|
||
* ) printf '%s:ipass-%s' "$REGISTRY" "$t" ;;
|
||
esac
|
||
}
|
||
|
||
# ---- 正式环境访问(域名解析直连 .11,绕开 DNS) -----------------------------
|
||
# 全部在这里做,命令行里不出现任何凭据。
|
||
live_curl() { curl -sk -m "${TIMEOUT:-20}" --resolve "$LIVE_HOST:$LIVE_PORT:$LIVE_IP" "$@"; }
|
||
|
||
# 登录拿会话 cookie,写入 $1
|
||
live_login() {
|
||
local jar="$1" body code
|
||
body="$("$PY" - "$IPASS_ADMIN_USER" "$IPASS_ADMIN_PASSWORD" <<'PY'
|
||
import json, sys
|
||
print(json.dumps({"username": sys.argv[1], "password": sys.argv[2]}))
|
||
PY
|
||
)"
|
||
code="$(live_curl -o /dev/null -w '%{http_code}' -c "$jar" \
|
||
-H 'Content-Type: application/json' -d "$body" \
|
||
"https://$LIVE_HOST:$LIVE_PORT/api/admin/login")"
|
||
[ "$code" = "200" ]
|
||
}
|
||
|
||
cmd_wait() {
|
||
local id="$1" i j state=""
|
||
for i in $(seq 1 120); do
|
||
j="$(get "/jobs/$id?offset=0")"
|
||
state="$(printf '%s' "$j" | jget state)"
|
||
case "$state" in done|failed|error|cancelled) break ;; esac
|
||
sleep 1
|
||
done
|
||
if [ -n "$PY" ] && [ -n "$j" ]; then
|
||
printf '%s' "$j" | "$PY" -c '
|
||
import json, sys
|
||
d = json.load(sys.stdin)
|
||
print("state:", d.get("state"))
|
||
lg = d.get("log") or d.get("output") or []
|
||
if isinstance(lg, str):
|
||
lg = lg.splitlines()
|
||
for line in lg[-35:]:
|
||
print(line)
|
||
'
|
||
else
|
||
printf '%s\n' "$j"
|
||
fi
|
||
[ "$state" = "done" ]
|
||
}
|
||
|
||
# 正式环境验收:镜像对不对 + 新代码真的在线上跑(页面标记 + 接口下发)
|
||
cmd_verify() {
|
||
local jar rc=0 img dig
|
||
jar="$(mktemp -t ipass-live.XXXXXX)"
|
||
|
||
img="$(get /status | jget image)"
|
||
dig="$(get /status | jget digest)"
|
||
echo "· 镜像: $img"
|
||
echo "· digest: ${dig##*@}"
|
||
|
||
# 0) 默认 tag 漂移告警:.env 的 IPASS_IMAGE_TAG 落后于线上时,
|
||
# 裸 `update`(不带 tag)会**静默降级**回旧镜像 —— 只提示,不判失败。
|
||
# 发版一条龙结尾必跑 verify,所以这里是一次固定的自检点。
|
||
if [ -z "$DEFAULT_TAG" ]; then
|
||
echo " ! .env 未设 IPASS_IMAGE_TAG —— 裸 update 无缺省 tag(须显式传)"
|
||
elif [ "$img" = "${REGISTRY}:${DEFAULT_TAG}" ] || [ "$img" = "$DEFAULT_TAG" ]; then
|
||
echo " ✓ .env 默认 tag 与线上一致($DEFAULT_TAG)"
|
||
else
|
||
echo " ! .env 的 IPASS_IMAGE_TAG=$DEFAULT_TAG 落后于线上(${img##*:})"
|
||
echo " 裸 \`bash deploy/agent.sh update\` 会退回 $DEFAULT_TAG —— 请把 .env 同步成 ${img##*:}"
|
||
fi
|
||
|
||
echo "· 登录 ${LIVE_HOST}:${LIVE_PORT}(-> ${LIVE_IP})"
|
||
if ! live_login "$jar"; then
|
||
echo " ✗ 登录失败(HTTP 非 200)"; rm -f "$jar"; return 1
|
||
fi
|
||
echo " ✓ 登录 200"
|
||
|
||
# 1) 管理台页面:多页拆分 + 图表(这两条都是本轮改动)
|
||
local html
|
||
html="$(live_curl -b "$jar" "https://$LIVE_HOST:$LIVE_PORT/admin")"
|
||
local m
|
||
for m in 'assets/admin.css' 'assets/chart.js' 'statTokToday'; do
|
||
if printf '%s' "$html" | grep -qF -- "$m"; then
|
||
echo " ✓ 页面含「${m}」"
|
||
else
|
||
echo " ✗ 页面缺「${m}」—— 镜像可能没真正换新"; rc=1
|
||
fi
|
||
done
|
||
|
||
# 2) 接口下发子域名规则(后端常量真的生效)
|
||
local rule
|
||
rule="$(live_curl -b "$jar" "https://$LIVE_HOST:$LIVE_PORT/api/admin/domains" \
|
||
| "$PY" -c 'import json,sys;d=json.load(sys.stdin);print(json.dumps((d.get("data") or {}).get("subdomain_rule"),ensure_ascii=False))' 2>/dev/null || echo '')"
|
||
echo " · subdomain_rule = $rule"
|
||
case "$rule" in
|
||
*'"suffix": "dm"'*|*'"suffix":"dm"'*) echo " ✓ 规则后缀 = dm" ;;
|
||
*) echo " ✗ 规则未下发或后缀不对"; rc=1 ;;
|
||
esac
|
||
|
||
# 3) 趋势接口(日桶):新代码在跑的直接证据——旧镜像没有这个路由
|
||
local labels
|
||
labels="$(live_curl -b "$jar" "https://$LIVE_HOST:$LIVE_PORT/api/admin/trend?days=7" \
|
||
| "$PY" -c 'import json,sys
|
||
d=(json.load(sys.stdin).get("data") or {})
|
||
print(len(d.get("labels") or []))' 2>/dev/null || echo '')"
|
||
if [ "$labels" = "7" ]; then
|
||
echo " ✓ 趋势接口返回 7 天序列"
|
||
else
|
||
echo " ✗ 趋势接口异常(labels=${labels:-空})—— 镜像可能没真正换新"; rc=1
|
||
fi
|
||
|
||
# 4) 上游账号级负载(1.0.14 新增):投影里出现 accounts[].load = 新代码在跑。
|
||
# 1.0.14 起**单 auth 的旧记录也会被合成为一条 account**,所以只要有任一上游记录,
|
||
# 新镜像必定返回该字段;旧镜像返回的 items 里根本没有 accounts 键 —— 新旧可分。
|
||
# 只验「字段在不在」不验数值(线上未必有流量)。三态避免「空库」时假失败。
|
||
local acct
|
||
acct="$(live_curl -b "$jar" "https://$LIVE_HOST:$LIVE_PORT/api/admin/upstreams" \
|
||
| "$PY" -c 'import json,sys
|
||
d = json.load(sys.stdin).get("data") or {}
|
||
items = d.get("items") or [] if isinstance(d, dict) else (d if isinstance(d, list) else [])
|
||
if not items:
|
||
print("noupstream"); sys.exit(0)
|
||
for u in items:
|
||
for a in (u.get("accounts") or []):
|
||
if isinstance(a.get("load"), dict) and "requests" in a["load"]:
|
||
print("ok"); sys.exit(0)
|
||
print("old")' 2>/dev/null || echo '')"
|
||
case "$acct" in
|
||
ok) echo " ✓ 上游带账号级负载字段(accounts[].load)" ;;
|
||
noupstream) echo " · 线上没有任何上游,跳过账号负载证据(不判失败)" ;;
|
||
*) echo " ✗ 上游接口缺 accounts[].load —— 镜像可能没真正换新"; rc=1 ;;
|
||
esac
|
||
|
||
rm -f "$jar"
|
||
if [ "$rc" = "0" ]; then echo "✔ 正式环境验收通过"; else echo "✗ 正式环境验收未通过"; fi
|
||
return $rc
|
||
}
|
||
|
||
# 上游诊断(排障多账号路由 / 额度切换):登录管理台拉完整投影,
|
||
# 逐账号打印 **Authorization 掩码** / 冷却 / 重置周期 / 今日负载。
|
||
# 掩码为空会显式标 `**空**`——「配了 Key 却 401」最常见的原因就是它。
|
||
# ⚠️ 只回掩码(模块的 view() 本来也不回真值),真 Key 永不出容器。
|
||
cmd_upstreams() {
|
||
local jar json
|
||
jar="$(mktemp -t ipass-up.XXXXXX)"
|
||
if ! live_login "$jar"; then
|
||
echo "✗ 登录失败(HTTP 非 200)"; rm -f "$jar"; return 1
|
||
fi
|
||
json="$(live_curl -b "$jar" "https://$LIVE_HOST:$LIVE_PORT/api/admin/upstreams")"
|
||
rm -f "$jar"
|
||
if [ -z "$PY" ]; then printf '%s\n' "$json"; return 0; fi
|
||
printf '%s' "$json" | "$PY" -c '
|
||
import json, sys
|
||
d = json.load(sys.stdin).get("data") or {}
|
||
items = d.get("items") or [] if isinstance(d, dict) else (d if isinstance(d, list) else [])
|
||
if not items:
|
||
print("(线上没有任何上游)"); raise SystemExit(0)
|
||
for u in items:
|
||
accts = u.get("accounts") or []
|
||
print("%s base=%s enabled=%s 账号=%d" % (
|
||
u.get("slot"), u.get("base"), u.get("enabled"), len(accts)))
|
||
if not accts:
|
||
print(" (无账号 —— 转发时不会带 Authorization 头)")
|
||
for a in accts:
|
||
ld = a.get("load") or {}
|
||
m = a.get("auth_mask")
|
||
if not m or m == "-":
|
||
m = "**空** (转发不带 Authorization 头 → 上游报鉴权错)"
|
||
print(" #%s Authorization=%s%s resets=%s 今日 %s 请求 / 入 %s / 出 %s%s" % (
|
||
a.get("idx"), m,
|
||
" [冷却中]" if a.get("cooled") else "",
|
||
a.get("resets_text") or "-",
|
||
ld.get("requests", 0), ld.get("tokens_in", 0), ld.get("tokens_out", 0),
|
||
(" 耗尽%s" % ld["exhausts"]) if ld.get("exhausts") else ""))
|
||
'
|
||
}
|
||
|
||
# 上游连通性测试(不靠管理台浏览器):登录后调 /api/admin/upstreams/test,
|
||
# 把上游真实响应的 code / 关键错误体打出来。比点面板按钮的「测试」多两个价值:
|
||
# ① agent 自动化可调;② 出错时直接看厂商原话(不用肉眼对截图 OCR)。
|
||
# 用法:bash deploy/agent.sh upstream-test minimax
|
||
cmd_upstream_test() {
|
||
local slot="$1" jar body
|
||
jar="$(mktemp -t ipass-uptest.XXXXXX)"
|
||
if ! live_login "$jar"; then
|
||
echo "✗ 登录失败"; rm -f "$jar"; return 1
|
||
fi
|
||
body="$(live_curl -b "$jar" -H 'Content-Type: application/json' \
|
||
-d "{\"slot\":\"$slot\"}" \
|
||
"https://$LIVE_HOST:$LIVE_PORT/api/admin/upstreams/test")"
|
||
rm -f "$jar"
|
||
if [ -z "$PY" ]; then printf '%s\n' "$body"; return 0; fi
|
||
printf '%s' "$body" | "$PY" -c '
|
||
import json, sys
|
||
d = json.load(sys.stdin)
|
||
data = d.get("data") or {}
|
||
code = data.get("status_code") or d.get("code")
|
||
status = data.get("status") or d.get("msg", "")
|
||
body = data.get("body") or ""
|
||
print(f" 上游 HTTP {code} {status}")
|
||
if body:
|
||
short = body if len(body) <= 300 else body[:300] + "…"
|
||
print(" body:", short)
|
||
'
|
||
}
|
||
|
||
CMD="${1:-}"; shift || true
|
||
|
||
case "$CMD" in
|
||
status|check|health) get "/$CMD" | pretty ;;
|
||
containers) get "/containers" | pretty ;;
|
||
logs) get "/logs?tail=${1:-200}" ;;
|
||
jobs) get "/jobs?limit=${1:-5}" | pretty ;;
|
||
job) [ -n "${1:-}" ] || { echo "用法: bash deploy/agent.sh job <job-id> [offset]" >&2; exit 2; }
|
||
get "/jobs/$1?offset=${2:-0}" | pretty ;;
|
||
data) [ -n "${1:-}" ] || { echo "用法: bash deploy/agent.sh data users|domains|upstreams" >&2; exit 2; }
|
||
get "/data?file=$1" | pretty ;;
|
||
upstreams) cmd_upstreams ;;
|
||
upstream-test) [ -n "${1:-}" ] || { echo "用法: bash deploy/agent.sh upstream-test <slot>" >&2; exit 2; }
|
||
cmd_upstream_test "$1" ;;
|
||
pull) IMG="$(resolve_image "${1:-}")"; echo "· 拉取 $IMG"
|
||
post "/pull" "{\"image\":\"$IMG\"}" | pretty ;;
|
||
update) IMG="$(resolve_image "${1:-}")"; echo "· 更新到 $IMG"
|
||
post "/update" "{\"image\":\"$IMG\"}" | pretty ;;
|
||
wait) [ -n "${1:-}" ] || { echo "用法: bash deploy/agent.sh wait <job-id>" >&2; exit 2; }
|
||
cmd_wait "$1" ;;
|
||
verify) cmd_verify ;;
|
||
rollback) if [ -n "${1:-}" ]; then post "/rollback" "{\"tag\":\"$1\"}" | pretty
|
||
else post "/rollback" '{}' | pretty; fi ;;
|
||
restart) post "/restart" '{}' | pretty ;;
|
||
raw) [ -n "${1:-}" ] || { echo "用法: bash deploy/agent.sh raw GET /check" >&2; exit 2; }
|
||
get "$2" | pretty ;;
|
||
release)
|
||
TAG="${1:-}"
|
||
[ -n "$TAG" ] || { echo "用法: bash deploy/agent.sh release 1.0.7" >&2; exit 2; }
|
||
IMG="$(resolve_image "$TAG")"
|
||
echo "== 1/3 构建并推送 $IMG =="
|
||
bash "$ROOT/deploy/release.sh" "${TAG#ipass-}"
|
||
echo
|
||
echo "== 2/3 更新正式环境 =="
|
||
JOB="$(post "/update" "{\"image\":\"$IMG\"}" | jget job)"
|
||
[ -n "$JOB" ] || { echo "✗ 未拿到 job id" >&2; exit 1; }
|
||
echo "· job = $JOB"
|
||
cmd_wait "$JOB" || { echo "✗ 更新任务未成功" >&2; exit 1; }
|
||
echo
|
||
echo "== 3/3 验收 =="
|
||
cmd_verify ;;
|
||
*)
|
||
cat <<'USAGE'
|
||
iPass 控制面包装(凭据只从 .env 读,命令行里不出现任何 key)
|
||
|
||
状态与信息
|
||
bash deploy/agent.sh status # 容器状态 + 镜像 digest + 数据摘要
|
||
bash deploy/agent.sh containers # 所有容器(标出受管容器)
|
||
bash deploy/agent.sh logs [tail] # 受管容器日志
|
||
bash deploy/agent.sh data users|domains|upstreams
|
||
bash deploy/agent.sh upstreams # 上游排障:逐账号掩码/冷却/今日负载
|
||
bash deploy/agent.sh upstream-test <slot> # 上游连通性:调测试接口并打印厂商真实响应
|
||
|
||
变更
|
||
bash deploy/agent.sh update [tag] # 拉镜像 + 重建容器(缺省用 .env 的 IPASS_IMAGE_TAG)
|
||
bash deploy/agent.sh wait <job-id> # 轮询任务直到结束并打印日志
|
||
bash deploy/agent.sh rollback [tag] # 回滚
|
||
bash deploy/agent.sh restart # 只重启
|
||
bash deploy/agent.sh pull [tag] # 只拉镜像
|
||
|
||
发版一条龙(构建推送 → 更新 → 验收)
|
||
bash deploy/agent.sh release 1.0.7
|
||
|
||
正式环境验收
|
||
bash deploy/agent.sh verify # 登录 + 拉页面标记 + 查 subdomain_rule
|
||
|
||
任务
|
||
bash deploy/agent.sh jobs [limit]
|
||
bash deploy/agent.sh job <id> [offset]
|
||
bash deploy/agent.sh raw GET /check
|
||
|
||
传 1.0.6 / ipass-1.0.6 / 完整镜像名 均可。
|
||
USAGE
|
||
[ -n "$CMD" ] && exit 2 || exit 0 ;;
|
||
esac
|