fix(mcp+adapter): 流式 tool_calls 解析修复 + MCP transport 超时保护

openai adapter transform_stream_chunk:
- OpenAI 流式分片是嵌套格式 function.{name,arguments},原样透传后
  json.Unmarshal 到扁平 ToolCall{name,arguments} 时 name 恒为空,
  accumulateStream flushToolCall 因 acc.name=="" 静默丢弃整个工具调用
  (流式路径自上线起 tool_calls 全部丢失的根因)
- 现在正确解包 function.name → name,function.arguments → raw_arguments
  (保留原始 JSON 字符串分片,由 accumulateStream 按 index 拼接)
- 不按 name 过滤分片:OpenAI 流式续传块 name 为空但携带 arguments 分片

mcp stdio/sse transport:
- stdio Send() 无超时:server 进程卡死时插件加载永久阻塞
- sse http.Client 无超时:远程 server 网络抖动/无响应时永久阻塞,
  导致 webui 等后续插件全部无法启动(生产实例偶发启动卡死根因)
- stdio 加 60s 请求超时;sse client 加 30s 整体 + 10s 拨号超时
This commit is contained in:
JianFeeeee
2026-08-25 18:58:19 +08:00
parent 8c5b35a6b1
commit 51eb98e0ae
4 changed files with 49 additions and 4 deletions

View File

@ -99,7 +99,28 @@ function adapter.transform_stream_chunk(raw_chunk)
unified.reasoning_content = delta.reasoning_content
end
if delta.tool_calls then
unified.tool_calls = delta.tool_calls
local tcs = {}
for _, tc in ipairs(delta.tool_calls) do
-- OpenAI 流式格式: {function:{name,arguments}, id, type, index}
-- homed StreamChunk.ToolCalls 期望扁平格式: {id, type, name, raw_arguments}
local fn = tc["function"]
local name = (type(fn) == "table" and fn.name) or tc.name or ""
local raw_args = ""
if type(fn) == "table" and type(fn.arguments) == "string" then
raw_args = fn.arguments
elseif type(tc.arguments) == "string" then
raw_args = tc.arguments
end
-- 不能按 name 过滤OpenAI 流式分片中后续块 name 为空但携带 arguments
-- accumulateStream 按 index 累积并在 flushToolCall 时校验 name
table.insert(tcs, {
id = tc.id or "",
type = tc.type or "function",
name = name,
raw_arguments = raw_args
})
end
unified.tool_calls = tcs
end
return json.encode(unified)
end