feat(deploy): 新增 .11 一键更新脚本 + 上游多模型/模型发现/用量统计

功能(本轮 5 项需求的落地):
- 上游可挂多模型:llm-upstreams.json 每上游新增 models 清单(仅用于模型发现,不参与路由)
- 模型发现:GET /v1/models(OpenAI {object:list} 形状,网关本地应答)+ /api/v1/models + /api/v1/upstreams
- 用量统计:/api/admin/usage(管理台「用量统计」Tab)+ /api/v1/usage + /me 的 llm_usage 摘要
  口径:只统计真正转发出去的请求,401/403/404/429 不计;与配额计数 st: 刻意分开
- 账号展示名 display_name(纯展示,不进 ukey:)
- 修复 /v1 版本段重复拼接(llm::join_target)→ 上游 base 自带 /v1 时不再 404 空体
- 修复删账号未清用量键(stats::forget)

部署脚本 deploy/update-ipass.sh:
- 拉镜像 → tar 备份数据 → 删旧容器 → 按旧容器复刻端口/挂载/网络起新容器 → 自检
- 参数从 docker inspect 静态配置读取(docker port 对停止态容器返回空,会让端口静默回落)
- 端口按 host:container 去重(inspect 同一映射含 IPv4/IPv6 两项,重复 -p 会 address already in use)
- --bg 用 setsid/start_new_session 真正脱离会话(仅 nohup & 会被父进程组连带杀掉)
- 旧镜像自动打 rollback-<时间> 标签,收尾打印回滚命令

回归:bash tests/run.sh all → 网关 78 + LLM 75 = 153 条断言全绿。
文档:readme §12.4 / §15 / §16、CLAUDE.md 常用操作 + 排障速查已同步。
This commit is contained in:
admin committed 2026-09-12 10:09:56 +08:00
1 parent 94a48bef6e
commit e38237dc4a
22 files changed
+1719 -93

No files matched your search

+15 -6
View File
@@ -4,8 +4,8 @@
```bash
bash tests/run.sh # 全部(域名网关 + LLM 代理)
bash tests/run.sh gateway # 只跑域名网关(55 条断言)
bash tests/run.sh llm # 只跑 LLM 代理
bash tests/run.sh gateway # 只跑域名网关(78 条断言)
bash tests/run.sh llm # 只跑 LLM 代理(75 条断言)
```
脚本会自动 `cargo build`,起 nginx,跑断言,收尾停进程。**期望输出「失败 0」。**
@@ -18,10 +18,10 @@ bash tests/run.sh llm # 只跑 LLM 代理
| `gateway/nginx.conf.in` | 域名网关测试配置**模板**(占位符 `@MODULE@` / `@REPO@` / `@RUN@`) |
| `gateway/e2e.sh` | **55 条断言**:鉴权会话 / 账号 / 申请审核 / 免审 / 越权隔离 / Key 重置 / 清理 |
| `gateway/backend/index.html` | 转发靶机内容(断言里检查 `"from":"backend"`) |
| `llm/nginx.conf.in` | LLM 代理测试配置模板(18090 网关) |
| `llm/upstream.py` | 假上游(18999):**记录它实际收到的 path / auth / host / body** |
| `llm/nginx.conf.in` | LLM 代理测试配置模板(18090 网关 + **18091 管理端口**,供账号态 `/api/v1/*` 断言) |
| `llm/upstream.py` | 假上游(18999):**记录它实际收到的 path / auth / host / body**;`/<slot>/v1/models` 原样透传 |
| `llm/sse_test.py` | 流式探针:输出 `CHUNKS` / `FIRST` / `TOTAL`,判断 SSE 有没有被缓冲 |
| `llm/e2e.sh` | LLM 侧断言:鉴权 / 白名单 / 停用摘除 / **Key 替换** / 路径拼接 / SSE |
| `llm/e2e.sh` | LLM 侧断言:鉴权 / 白名单 / 停用摘除 / **Key 替换** / 路径拼接(含版本段去重)/ SSE / **模型发现** / **账号态用量** / **管理员用量总览与删账号清用量** |
| `llm/fixtures/*.json` | 固定账号与上游表(**合成数据**,见下) |
| `.run/` | 运行目录(生成的 `nginx.conf`、`data/`、`logs/`、`cookie.jar`)—— 已 gitignore |
@@ -47,10 +47,19 @@ sse_test.py → 首块 < 0.5s 到达(证明 proxy_buffering off 生效)
## fixtures 说明
`llm/fixtures/users.json` 与 `llm-upstreams.json` 是**合成夹具**:
`pwd_hash` 是占位、`api_key` 是 `aaaa…` / `bbbb…`、上游 `auth` 是
`pwd_hash` 多为占位、`api_key` 是 `aaaa…` / `bbbb…`、上游 `auth` 是
`sk-UPSTREAM-REAL-SECRET` 这类明显假值。它们**不是**任何环境的真实凭据,
刻意入库,好让 LLM 侧回归在干净机器上一把跑通。
两个例外值得知道:
- `admin` 账号的 `pwd_hash` 是**按项目算法真算出来的**(口令 `123456`)——
【10】要真登录管理台才能查用量、删账号。算法见 `rust/src/util.rs::hash_password`:
`x = sha256(salt + ":" + pwd)`,再迭代 1000 轮 `x = sha256(x + salt)`。
改口令就照它重算,别写成占位(写错的表现是「登录 401」)。
- `deepseeklike` 上游的 `base` **自带 `/v1`**,专测**版本段去重**
(少这一步会拼成 `/v1/v1/...` → 上游 404 且响应体为空)。
真实凭据只在仓库根的 `data/*.json`(已 gitignore,绝不入库)。
## 写这类脚本的引号陷阱(曾因此假绿,浪费一轮排查)
+53 -2
View File
@@ -82,9 +82,11 @@ ck "口令长度不足登录(非管理员路径)→ 401" 401 "$(code -X POST
echo
echo "【2】账号管理"
BODY=$(jbody username alice password alice123 role user)
BODY=$(jbody username alice password alice123 role user display_name 爱丽丝)
CREATE=$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/users/create -H 'Content-Type: application/json' -d "$BODY")
ck "创建 alice → 200" 200 "$(echo "$CREATE" | jqv code)"
ck "★展示名(中文姓名)已保存" "爱丽丝" "$(echo "$CREATE" | jqv data.display_name)"
ck " name 回落到展示名" "爱丽丝" "$(echo "$CREATE" | jqv data.name)"
ALICE_KEY=$(echo "$CREATE" | jqv data.api_key)
if [ -n "${ALICE_KEY:-}" ] && [ "${ALICE_KEY#sk_}" != "$ALICE_KEY" ]; then KEYED=yes; else KEYED=no; fi
ck "返回 api_key(sk_ 前缀)" yes "$KEYED"
@@ -186,7 +188,56 @@ BODY=$(jbody username bob)
ck "删除 bob → 200" 200 "$(code -b "$JAR" -X POST $ADMIN/api/admin/users/delete -H 'Content-Type: application/json' -d "$BODY")"
echo
echo "【9】管理端与清理"
echo "【9】LLM 上游多模型 / 模型清单 / 用量"
BODY=$(jbody slot deepseek base https://api.deepseek.com/v1 auth 'Bearer sk-test-not-real' models 'deepseek-chat, deepseek-reasoner' note 测试上游)
UP=$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/upstreams/save -H 'Content-Type: application/json' -d "$BODY")
ck "保存上游(含模型清单)→ 200" 200 "$(echo "$UP" | jqv code)"
ck "★模型清单入库(2 个)" 2 "$(echo "$UP" | jqv data.model_count)"
ck " 上游列表回显模型名" yes \
"$(curl -s -b "$JAR" $ADMIN/api/admin/upstreams | grep -q deepseek-reasoner && echo yes || echo no)"
ck " base_url 由域名常量拼出(不是写死占位)" "https://lm.mayishangshu.cn/deepseek/v1" "$(echo "$UP" | jqv data.base_url)"
# 与 auth/note 同款语义:省略 models = 保持原清单
BODY=$(jbody slot deepseek base https://api.deepseek.com/v1)
ck "★省略 models 保存 → 原清单不被清掉" 2 \
"$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/upstreams/save -H 'Content-Type: application/json' -d "$BODY" | jqv data.model_count)"
# 显式传空串 = 清空(契约的另一半,必须与"省略"区分开)
BODY=$(jbody slot deepseek base https://api.deepseek.com/v1 models '')
ck "★显式传空 models → 清空" 0 \
"$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/upstreams/save -H 'Content-Type: application/json' -d "$BODY" | jqv data.model_count)"
# 换行/顿号分隔也能吃(粘贴厂商文档的常见形态)。
# 注意这里必须用 JSON 的 \n 转义:直接写**字面换行**会让 body 变成非法 JSON,
# 解析失败 → 参数全空 → 断言会因为"上游不存在/400"而假失败。
BODY=$(jbody slot deepseek base https://api.deepseek.com/v1 models 'deepseek-chat\ndeepseek-reasoner、deepseek-coder')
ck "换行 + 顿号分隔解析出 3 个" 3 \
"$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/upstreams/save -H 'Content-Type: application/json' -d "$BODY" | jqv data.model_count)"
ck "非法模型名(含空格)→ 400" 400 \
"$(code -b "$JAR" -X POST $ADMIN/api/admin/upstreams/save -H 'Content-Type: application/json' -d "$(jbody slot deepseek base https://api.deepseek.com/v1 models 'bad model name')")"
ck "无 Key 查可用模型 → 401" 401 "$(code $ADMIN/api/v1/models)"
ck "alice 查可用模型 → 200" 200 "$(code -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/models)"
ck " 含 deepseek-chat" yes \
"$(curl -s -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/models | grep -q deepseek-chat && echo yes || echo no)"
ck " 带 OpenAI 口径的 base_url" yes \
"$(curl -s -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/models | grep -q 'lm.mayishangshu.cn/deepseek/v1' && echo yes || echo no)"
ck "无 Key 查用量 → 401" 401 "$(code $ADMIN/api/v1/usage)"
ck "alice 查用量 → 200" 200 "$(code -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/usage)"
ck " 没调用过 → 累计 0" 0 "$(curl -s -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/usage | jqv data.total)"
ck " 回显限额" yes "$(curl -s -H "X-API-Key: $ALICE_KEY" $ADMIN/api/v1/usage | grep -q limits && echo yes || echo no)"
ck "管理员用量总览 → 200" 200 "$(code -b "$JAR" $ADMIN/api/admin/usage)"
ck " 含 users / models / total 三段" yes \
"$(curl -s -b "$JAR" $ADMIN/api/admin/usage | grep -q '"models"' && echo yes || echo no)"
BODY=$(jbody username alice display_name 爱丽丝改名)
D1=$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/users/display -H 'Content-Type: application/json' -d "$BODY")
ck "改展示名 → 200" 200 "$(echo "$D1" | jqv code)"
ck " 新展示名生效" "爱丽丝改名" "$(echo "$D1" | jqv data.display_name)"
BODY=$(jbody username alice display_name '')
ck "清空展示名 → name 回落账号名" alice \
"$(curl -s -b "$JAR" -X POST $ADMIN/api/admin/users/display -H 'Content-Type: application/json' -d "$BODY" | jqv data.name)"
echo
echo "【10】管理端与清理"
ck "管理台页面 → 200" 200 "$(code -H 'Host: proxy-api.yourdomain.com' $ADMIN/admin)"
ck "管理台页面返回 HTML" yes "$(curl -s -H 'Host: proxy-api.yourdomain.com' $ADMIN/admin | grep -qi '<!DOCTYPE html' && echo yes || echo no)"
ck "purge → 200" 200 "$(code -b "$JAR" -X POST $ADMIN/api/admin/purge)"
+113
View File
@@ -18,6 +18,8 @@ HERE=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
RUN=${RUN:-$(cd "$HERE/.." && pwd)/.run/llm}
GW=${GW:-http://127.0.0.1:18090}
# 管理端口(同一份 nginx / 同一块共享内存)——用量与模型清单从这边读
API=${API:-http://127.0.0.1:18091}
UP_LOG=${UPSTREAM_LOG:-$RUN/upstream.log}
PY=${PY:-python3}
@@ -52,6 +54,23 @@ print(json.loads(last).get(sys.argv[2], '') if last else '')
up_count() { if [ -f "$UP_LOG" ]; then awk 'NF{n++} END{print n+0}' "$UP_LOG"; else echo 0; fi }
post() { curl -s -H "Authorization: Bearer $1" -X POST "$GW$2" -H 'Content-Type: application/json' -d "$3"; }
# 从 stdin 的 JSON 里按点分路径取值
jqv() {
"$PY" -c "
import sys, json
try: d = json.load(sys.stdin)
except Exception: print(''); sys.exit()
for k in sys.argv[1].split('.'):
if d is None: break
d = d[int(k)] if k.isdigit() and isinstance(d, list) else (d.get(k) if isinstance(d, dict) else None)
print('' if d is None else d)
" "$1" 2>/dev/null
}
# 在 stdin 里找子串(命中输出 yes,否则 no)
has() { if grep -q "$1"; then echo yes; else echo no; fi; }
# 反向断言:**不**出现该子串才算过
nhas() { if grep -q "$1"; then echo no; else echo yes; fi; }
echo "【1】鉴权(OpenAI 兼容错误体)"
BODY=$(jbody model gpt-4o)
ck "无 Key → 401" 401 "$(code -X POST $GW/openai/v1/chat/completions)"
@@ -86,6 +105,14 @@ post "$BOB_KEY" /other/v1/chat/completions "$BODY" >/dev/null
ck "★base=/llm + path 正确拼接" "/llm/v1/chat/completions" "$(up_field path)"
ck "用的是该上游自己的 Key" "Bearer sk-OTHER-UPSTREAM" "$(up_field auth)"
# ★ 版本段去重(实测踩过的坑:厂商 base_url 自带 /v1,客户端也带 /v1 → /v1/v1 → 上游 404 空响应)
post "$BOB_KEY" /deepseeklike/v1/chat/completions "$BODY" >/dev/null
ck "★base 自带 /v1 + 客户端带 /v1 → 只保留一份" "/v1/chat/completions" "$(up_field path)"
post "$BOB_KEY" /deepseeklike/chat/completions "$BODY" >/dev/null
ck " 客户端不带 /v1 → 照常前缀上 base 的 /v1" "/v1/chat/completions" "$(up_field path)"
post "$BOB_KEY" /deepseeklike/v2/chat/completions "$BODY" >/dev/null
ck " 不同版本段不误判为重复" "/v1/v2/chat/completions" "$(up_field path)"
echo
echo "【5】★ 流式(SSE 不被缓冲)"
SSE=$(GW="$GW" KEY="$ALICE_KEY" "$PY" "$HERE/sse_test.py" 2>&1)
@@ -129,6 +156,92 @@ else
echo "【7】HTTPS:跳过(本机 nginx 无 http_ssl_module 或无 openssl)"
fi
echo
echo "【8】★ 模型发现(网关本地应答,不去上游)"
M0=$(up_count)
ck "无 Key → 401" 401 "$(code $GW/v1/models)"
ck "alice GET /v1/models → 200" 200 "$(code -H "Authorization: Bearer $ALICE_KEY" $GW/v1/models)"
ck " 形状是 OpenAI 的 object=list" yes \
"$(curl -s -H "Authorization: Bearer $ALICE_KEY" $GW/v1/models | has '"object":"list"')"
ck " 含白名单内上游的模型 gpt-4o" yes \
"$(curl -s -H "Authorization: Bearer $ALICE_KEY" $GW/v1/models | has 'gpt-4o')"
ck " 不含白名单外上游的模型 claude-3-5-sonnet" yes \
"$(curl -s -H "Authorization: Bearer $ALICE_KEY" $GW/v1/models | nhas 'claude-3-5-sonnet')"
ck " 不含已停用上游的模型 should-not-appear" yes \
"$(curl -s -H "Authorization: Bearer $ALICE_KEY" $GW/v1/models | nhas 'should-not-appear')"
ck "bob(白名单为空=不限)能看到 other 的模型" yes \
"$(curl -s -H "Authorization: Bearer $BOB_KEY" $GW/v1/models | has 'claude-3-5-sonnet')"
ck "GET /models 同样本地应答 → 200" 200 "$(code -H "Authorization: Bearer $ALICE_KEY" $GW/models)"
ck "★整个模型发现过程 0 次上游调用" 0 "$(( $(up_count) - M0 ))"
# 带 slot 的 /v1/models 是**另一条路**:照常透传给厂商,取"那家厂商的"模型清单
N0=$(up_count)
ck "★/<slot>/v1/models 仍透传上游 → 200" 200 \
"$(code -H "Authorization: Bearer $ALICE_KEY" $GW/openai/v1/models)"
ck " 上游确实收到了 1 次请求" 1 "$(( $(up_count) - N0 ))"
ck " 转发路径未剥错(/v1/models)" "/v1/models" "$(up_field path)"
ck " 响应体来自上游(不是网关本地那份)" yes \
"$(curl -s -H "Authorization: Bearer $ALICE_KEY" $GW/openai/v1/models | has 'upstream-model')"
echo
echo "【9】★ 账号态 API:可用模型 + 自己的用量"
ck "无 Key 查模型 → 401" 401 "$(code $API/api/v1/models)"
ck "无 Key 查用量 → 401" 401 "$(code $API/api/v1/usage)"
ck "alice 查模型 → 200" 200 "$(code -H "X-API-Key: $ALICE_KEY" $API/api/v1/models)"
ck " 模型清单里带 base_url(lm 域名 + slot)" yes \
"$(curl -s -H "X-API-Key: $ALICE_KEY" $API/api/v1/models | has 'lm.mayishangshu.cn/openai')"
# ★ base_url 必须**保留上游 base 的路径**:openai 的 base 无路径 → 无;other 的 base 带 /llm → 带 /llm。
# 搞错这一条会让"base 自带 /v1"的上游变成 /v1/v1/... → 上游 404 且响应体为空,极难排查。
ck "★base_url 不凭空拼 /v1(base 无路径就不带)" yes \
"$(curl -s -H "X-API-Key: $ALICE_KEY" $API/api/v1/models | nhas 'lm.mayishangshu.cn/openai/')"
ck "★base_url 保留上游 base 的路径 /llm" yes \
"$(curl -s -H "X-API-Key: $BOB_KEY" $API/api/v1/models | has 'lm.mayishangshu.cn/other/llm')"
ALICE_USAGE=$(curl -s -H "X-API-Key: $ALICE_KEY" $API/api/v1/usage)
ck "alice 查用量 → 200" 200 "$(code -H "X-API-Key: $ALICE_KEY" $API/api/v1/usage)"
ALICE_TOTAL=$(echo "$ALICE_USAGE" | jqv data.total)
ck "★alice 累计用量 > 0(前面真发过请求)" yes \
"$(awk -v t="${ALICE_TOTAL:-0}" 'BEGIN{print (t>0)?"yes":"no"}')"
ck " 今日 = 累计(都是刚发的)" "$ALICE_TOTAL" "$(echo "$ALICE_USAGE" | jqv data.today)"
ck " 分上游明细含 openai" yes "$(echo "$ALICE_USAGE" | has '"slot": *"openai"')"
ck "★被拒的越权/未知上游不进明细(无 other/downstream)" yes \
"$(echo "$ALICE_USAGE" | nhas '"slot": *"other"')"
ck " 回显限额供客户端自我节流" yes "$(echo "$ALICE_USAGE" | has '"limits"')"
# ★ 关键语义:统计的是"真的转发出去"的请求 —— 超限被 429 的那一发不计
CAROL_TOTAL=$(curl -s -H "X-API-Key: $CAROL_KEY" $API/api/v1/usage | jqv data.total)
ck "★carol 发了 3 次、第 3 次被限流 → 用量只记 2" 2 "${CAROL_TOTAL:-x}"
DAVE_TOTAL=$(curl -s -H "X-API-Key: $DAVE_KEY" $API/api/v1/usage | jqv data.total)
ck "★dave 同上(每日配额)→ 用量只记 2" 2 "${DAVE_TOTAL:-x}"
ck "bob 的明细含 other(他真调过)" yes \
"$(curl -s -H "X-API-Key: $BOB_KEY" $API/api/v1/usage | has '"slot": *"other"')"
echo
echo "【10】管理员用量总览 + 删账号清理用量键"
ck "无会话 → 401" 401 "$(code $API/api/admin/usage)"
ck "不是接口路径 → 404" 404 "$(code $API/api/v1/nope)"
JAR=$RUN/cookie.jar
LOGIN=$(curl -s -c "$JAR" -X POST $API/api/admin/login -H 'Content-Type: application/json' \
-d "$(jbody username admin password 123456)")
ck "管理员登录 → 200" 200 "$(echo "$LOGIN" | jqv code)"
USAGE=$(curl -s -b "$JAR" $API/api/admin/usage)
ck "用量总览含 alice" yes "$(echo "$USAGE" | has '"username": *"alice"')"
ck " 总览含按上游分组(openai)" yes "$(echo "$USAGE" | has '"slot": *"openai"')"
ck " 全局合计 > 0" yes "$(echo "$USAGE" | jqv data.total.total | awk '{print ($1>0)?"yes":"no"}')"
ck "★删除前用量榜里有 bob(他真调过 other)" yes "$(echo "$USAGE" | has '"username": *"bob"')"
# ★ 删除账号必须**连带清掉用量键**(`ust:` / `ums:`)——它们是按前缀扫出来出报表的,
# 不清就会让已删账号一直挂在用量榜上(与"删账号必须摘 ukey:"同类的正确性问题)。
curl -s -b "$JAR" -X POST $API/api/admin/users/delete -H 'Content-Type: application/json' \
-d "$(jbody username bob)" >/dev/null
ck "★bob 已被删除(他的 Key 立即失效 → 401)" 401 "$(code -H "X-API-Key: $BOB_KEY" $API/api/v1/usage)"
USAGE2=$(curl -s -b "$JAR" $API/api/admin/usage)
ck "★删除后用量榜不再出现 bob" yes "$(echo "$USAGE2" | nhas 'bob')"
ck " 但 alice 仍在榜单上" yes "$(echo "$USAGE2" | has '"username": *"alice"')"
echo
echo "============================================================"
printf " 通过 \033[32m%d\033[0m 失败 \033[31m%d\033[0m\n" "$pass" "$fail"
+13 -2
View File
@@ -1,6 +1,7 @@
{
"openai": {
"base": "http://127.0.0.1:18999",
"models": ["gpt-4o", "gpt-4o-mini"],
"auth": "Bearer sk-UPSTREAM-REAL-SECRET",
"enabled": true,
"note": "本地靶机;auth 是占位值,不是真密钥",
@@ -8,16 +9,26 @@
},
"other": {
"base": "http://127.0.0.1:18999/llm",
"models": ["claude-3-5-sonnet"],
"auth": "Bearer sk-OTHER-UPSTREAM",
"enabled": true,
"note": "不在 alice 白名单内;base 带路径,用于验证路径拼接",
"note": "不在 alice 白名单内;base 带路径,用于验证路径拼接与模型发现过滤",
"created_at": 1757500000
},
"downstream": {
"base": "http://127.0.0.1:18999",
"models": ["should-not-appear"],
"auth": "",
"enabled": false,
"note": "已停用,用于验证「停用即从共享内存摘除」→ 404",
"note": "已停用,用于验证「停用即从共享内存摘除」→ 404,且其模型不应出现在模型发现里",
"created_at": 1757500000
},
"deepseeklike": {
"base": "http://127.0.0.1:18999/v1",
"models": ["ds-chat"],
"auth": "Bearer sk-DS-REAL",
"enabled": true,
"note": "base 自带 /v1(厂商文档的常见写法),用于验证版本段不重复拼接",
"created_at": 1757500000
}
}
+11
View File
@@ -1,6 +1,17 @@
{
"admin": {
"username": "admin",
"role": "admin",
"salt": "5a1b2c3d4e5f60718293a4b5c6d7e8f9",
"pwd_hash": "062b45d8f3cbef3d30464859355401bb37193868dc42eb941de86855416ab5e8",
"api_key": "sk_admin_0000000000000000000000000000000000000000000000ab",
"allowed_models": [],
"temp_grant_until": 0,
"created_at": 1757500000
},
"alice": {
"username": "alice",
"display_name": "爱丽丝",
"role": "user",
"salt": "00112233445566778899aabbccddeeff",
"pwd_hash": "00",
+16
View File
@@ -62,4 +62,20 @@ http {
proxy_read_timeout 60s;
}
}
# ---- 管理 server(只为在同一次回归里验账号态用量/模型接口)----
# 端口 18091;与生产一样只暴露 /api/,账号态用 X-API-Key 鉴权。
# 放在这里而不是另起一次 nginx:这样两个接口读的就是**同一块共享内存**,
# 「LLM 请求真的被计入用量」才是被验证到的事实,而不是各自独立的假象。
server {
listen 18091;
server_name _;
client_max_body_size 2m;
client_body_buffer_size 1m;
location /api/ {
ipass_api on;
}
}
}
+24 -3
View File
@@ -21,11 +21,10 @@ PORT = int(os.environ.get("UPSTREAM_PORT", "18999"))
class H(BaseHTTPRequestHandler):
protocol_version = "HTTP/1.1"
def do_POST(self):
n = int(self.headers.get("Content-Length") or 0)
body = self.rfile.read(n)
def _record(self, body: bytes):
rec = {
"path": self.path,
"method": self.command,
"auth": self.headers.get("Authorization", "-"),
"host": self.headers.get("Host", "-"),
"body": body.decode("utf-8", "replace"),
@@ -33,6 +32,28 @@ class H(BaseHTTPRequestHandler):
with open(LOG, "a") as f:
f.write(json.dumps(rec, ensure_ascii=False) + "\n")
def do_GET(self):
"""`GET /<slot>/v1/models` 必须原样透传到上游——这里验的就是这条。
网关自己的 `GET /v1/models`(无 slot 段)在本地应答、不会打到这里,
两条路互不干扰。
"""
self._record(b"")
payload = json.dumps({
"object": "list",
"data": [{"id": "upstream-model", "object": "model"}],
}).encode()
self.send_response(200)
self.send_header("Content-Type", "application/json")
self.send_header("Content-Length", str(len(payload)))
self.end_headers()
self.wfile.write(payload)
def do_POST(self):
n = int(self.headers.get("Content-Length") or 0)
body = self.rfile.read(n)
self._record(body)
if b'"stream"' in body and b"true" in body:
self.send_response(200)
self.send_header("Content-Type", "text/event-stream")
+3 -1
View File
@@ -157,9 +157,11 @@ suite_llm() {
echo "LLM 端口未就绪" >&2
ngx_stop "$RUN"; kill "$up_pid" 2>/dev/null; return 1
fi
wait_tcp 127.0.0.1 18091 || { echo "管理端口未就绪" >&2; ngx_stop "$RUN"; kill "$up_pid" 2>/dev/null; return 1; }
echo
GW="http://127.0.0.1:18090" UPSTREAM_LOG="$RUN/upstream.log" RUN="$RUN" PY="$PY" HAVE_TLS="$HAVE_TLS" \
GW="http://127.0.0.1:18090" API="http://127.0.0.1:18091" \
UPSTREAM_LOG="$RUN/upstream.log" RUN="$RUN" PY="$PY" HAVE_TLS="$HAVE_TLS" \
bash "$HERE/llm/e2e.sh"
local rc=$?