156 lines
6.0 KiB
Plaintext
156 lines
6.0 KiB
Plaintext
server {
|
||
# 文件名前缀是 000-，会比 ai.sggai.site.conf 更早加载。
|
||
# 这样可以在不修改原配置文件的情况下，让这里的 mock 逻辑优先生效。
|
||
listen 80;
|
||
listen 443 ssl;
|
||
server_name ai.sggai.site;
|
||
|
||
ssl_certificate /etc/letsencrypt/live/ai.sggai.site/fullchain.pem;
|
||
ssl_certificate_key /etc/letsencrypt/live/ai.sggai.site/privkey.pem;
|
||
|
||
location ^~ /.well-known/acme-challenge/ {
|
||
root /var/www;
|
||
default_type text/plain;
|
||
try_files $uri =404;
|
||
}
|
||
|
||
client_max_body_size 200m;
|
||
|
||
gzip off;
|
||
gunzip off;
|
||
|
||
# 公共反代配置放在 server 级别，主后端 proxy_pass location 会继承这些配置。
|
||
proxy_http_version 1.1;
|
||
proxy_set_header Host $host;
|
||
proxy_set_header Authorization $http_authorization;
|
||
proxy_set_header Content-Type $http_content_type;
|
||
proxy_set_header Accept $http_accept;
|
||
proxy_set_header User-Agent $http_user_agent;
|
||
proxy_set_header Originator $http_originator;
|
||
proxy_set_header Session_id $http_session_id;
|
||
proxy_set_header X-Codex-Beta-Features $http_x_codex_beta_features;
|
||
proxy_set_header X-Codex-Turn-Metadata $http_x_codex_turn_metadata;
|
||
proxy_set_header X-Stainless-Arch $http_x_stainless_arch;
|
||
proxy_set_header X-Stainless-Lang $http_x_stainless_lang;
|
||
proxy_set_header X-Stainless-Os $http_x_stainless_os;
|
||
proxy_set_header X-Stainless-Package-Version $http_x_stainless_package_version;
|
||
proxy_set_header X-Stainless-Retry-Count $http_x_stainless_retry_count;
|
||
proxy_set_header X-Stainless-Runtime $http_x_stainless_runtime;
|
||
proxy_set_header X-Stainless-Runtime-Version $http_x_stainless_runtime_version;
|
||
proxy_set_header X-Stainless-Timeout $http_x_stainless_timeout;
|
||
proxy_set_header X-App $http_x_app;
|
||
proxy_set_header Anthropic-Beta $http_anthropic_beta;
|
||
proxy_set_header Anthropic-Dangerous-Direct-Browser-Access $http_anthropic_dangerous_direct_browser_access;
|
||
proxy_set_header Anthropic-Version $http_anthropic_version;
|
||
proxy_set_header Accept-Encoding "";
|
||
proxy_set_header X-Real-IP "";
|
||
proxy_set_header X-Forwarded-For "";
|
||
proxy_set_header X-Forwarded-Proto "";
|
||
proxy_set_header X-Forwarded-Host "";
|
||
proxy_set_header X-Forwarded-Port "";
|
||
proxy_set_header Connection "";
|
||
proxy_buffering off;
|
||
proxy_request_buffering off;
|
||
proxy_cache off;
|
||
proxy_cache_bypass 1;
|
||
proxy_connect_timeout 600s;
|
||
proxy_read_timeout 3600s;
|
||
proxy_send_timeout 3600s;
|
||
|
||
# 只拦截很小的非流式 Chat Completions 请求。
|
||
# 不满足条件的请求会内部跳转到正常后端，不直接返回 mock。
|
||
location = /v1/chat/completions {
|
||
client_body_buffer_size 16k;
|
||
|
||
content_by_lua_block {
|
||
-- 不满足 mock 条件时，统一内部跳转到这个 named location。
|
||
local backend = "@ai_sggai_site_backend"
|
||
|
||
-- 请求体总大小上限：只处理非常小的探测类请求。
|
||
local max_body_bytes = 1024
|
||
|
||
-- 只处理 POST；其它方法按普通请求转发到真实后端。
|
||
if ngx.req.get_method() ~= "POST" then
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- 先看 Content-Length，超过阈值就不读取 body，直接放行。
|
||
local content_length = tonumber(ngx.var.http_content_length)
|
||
if not content_length then
|
||
-- 没有 Content-Length 的请求可能是 chunked，不为了判断 mock 去读取未知大小的 body。
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- 请求体太大，说明不是目标小请求，直接走真实后端。
|
||
if content_length > max_body_bytes then
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- 到这里说明 body 足够小，可以安全读取并解析 JSON。
|
||
ngx.req.read_body()
|
||
|
||
local body = ngx.req.get_body_data()
|
||
if not body or #body > max_body_bytes then
|
||
-- body 不在内存里或实际大小仍然超限时，不 mock，转发到真实后端。
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- JSON 解析失败，或 stream 不是 false，都不 mock。
|
||
local cjson = require "cjson.safe"
|
||
local payload = cjson.decode(body)
|
||
if type(payload) ~= "table" or payload.stream ~= false then
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- 必须有 messages 数组；没有就不 mock。
|
||
if type(payload.messages) ~= "table" then
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
local has_hi_content = false
|
||
|
||
-- 遍历所有 message，只匹配精确的 "content": "hi"。
|
||
for _, message in ipairs(payload.messages) do
|
||
if type(message) == "table" and message.content == "hi" then
|
||
has_hi_content = true
|
||
break
|
||
end
|
||
end
|
||
|
||
-- 没有精确命中 "content": "hi"，交给真实后端处理。
|
||
if not has_hi_content then
|
||
return ngx.exec(backend)
|
||
end
|
||
|
||
-- 命中小请求 + stream=false + "content": "hi" 条件，直接返回 mock 响应。
|
||
ngx.status = ngx.HTTP_OK
|
||
ngx.header["Content-Type"] = "application/json; charset=utf-8"
|
||
ngx.say([[{
|
||
"id": "chatcmpl-mock",
|
||
"object": "chat.completion",
|
||
"created": 1716030000,
|
||
"model": "xxx",
|
||
"choices": [
|
||
{
|
||
"index": 0,
|
||
"message": { "role": "assistant", "content": "ok" },
|
||
"finish_reason": "stop"
|
||
}
|
||
],
|
||
"usage": { "prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2 }
|
||
}]])
|
||
return ngx.exit(ngx.HTTP_OK)
|
||
}
|
||
}
|
||
|
||
# 仅供上面的 Lua 逻辑通过 ngx.exec 内部跳转使用，外部 URL 不能直接访问这个 location。
|
||
location @ai_sggai_site_backend {
|
||
proxy_pass http://10.1.0.1:3001;
|
||
}
|
||
|
||
# 普通流量保持原来的上游转发行为。
|
||
location / {
|
||
proxy_pass http://10.1.0.1:3001;
|
||
}
|
||
}
|