feat: OpenCode Go 会话 id 升级为对话级稳定(方案 A)
- models.py: ChatRequest 显式声明 conversation_id(内部字段,不透传上游) - openai_compat: chat/chat_stream/_req_headers 支持 session_id(有则用之,无则回退 UUID) - main.py: 非流式/流式主链路把 req.conversation_id 传为 session_id - Chat.vue: 对话级 convId(randomUUID),随对话持久化,清空对话时重建,注入 payload - 新增 2 个单测;157 passed / 0 failed;typecheck 通过;真实 key 冒烟稳定会话 OK
This commit is contained in:
1 parent
50f972c51b
commit
1b1d564f44
7 files changed
+90
-9
No files matched your search
+18
-1
@@ -4,7 +4,24 @@
|
|||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 2026-09-18 · 添加 OpenCode Go 供应商(订阅编程模型)
|
## 2026-09-18 · OpenCode Go 会话 id 升级为「对话级稳定」(方案 A)
|
||||||
|
|
||||||
|
**背景**:上一轮 `x-opencode-session` 是每请求随机 UUID —— 能过上游 400 强校验,但每次 id 都变,吃不到文档所述的提示词缓存。本轮按老板要求实现**稳定会话**:同一段对话共用同一个 session id。
|
||||||
|
|
||||||
|
### 改动
|
||||||
|
- **`app/models.py`**:`ChatRequest` 显式声明 `conversation_id: str | None`(网关内部字段,不进 `model_extra` → **不透传上游**)。
|
||||||
|
- **`app/adapters/openai_compat.py`**:`_req_headers(session_id)`、`chat`/`chat_stream` 新增 `session_id` 参数——传入则原样用作 `x-opencode-session`,缺省回退随机 UUID。
|
||||||
|
- **`app/main.py`**:非流式与流式主链路把 `req.conversation_id` 作为 `session_id` 传入适配器。
|
||||||
|
- **`frontend/src/views/Chat.vue`**:对话页新增 `convId`(`crypto.randomUUID`,带老浏览器回退),`buildPayload` 注入 `conversation_id`,随对话状态持久化(sessionStorage + 服务端历史);`applyChat` 沿用历史里的 id、否则新建;`clearConversation` 生成新 id(新对话不共享旧缓存)。
|
||||||
|
- **`tests/test_gateway.py`**:新增 `test_chat_conversation_id_is_internal_not_upstream`、`test_opencode_session_header_from_conversation_id`。
|
||||||
|
|
||||||
|
### 验证
|
||||||
|
- 单测 **157 passed / 0 failed**(+2);前端 `vue-tsc --noEmit` 通过。
|
||||||
|
- 真实 key 冒烟:同一 `conversation_id` 两次调用均 200,`x-opencode-session` 稳定一致;缺省回退 32 位随机、两次不同;非会话供应商不带该头。
|
||||||
|
- 说明:缓存命中是上游机会性优化,网关侧保证的是「同一对话稳定 session」。
|
||||||
|
|
||||||
|
### 上线(待老板批准)
|
||||||
|
- 含后端代码改动 → 需 `supervisorctl restart`;前端 `Settings.vue`/`Chat.vue` 改动需 `npm run build` 覆盖 `dist`;CNB 推送待确认。
|
||||||
|
|
||||||
**一句话**:给网关接入 OpenCode Go(`https://opencode.ai/zen/go/v1`)的 OpenAI 兼容 `/chat/completions` 线 17 个文本模型;按老板「wjd = 全部文本模型」规则全部并入 `routes.wjd`。
|
**一句话**:给网关接入 OpenCode Go(`https://opencode.ai/zen/go/v1`)的 OpenAI 兼容 `/chat/completions` 线 17 个文本模型;按老板「wjd = 全部文本模型」规则全部并入 `routes.wjd`。
|
||||||
|
|
||||||
|
|||||||
@@ -49,8 +49,13 @@ api-gateway/
|
|||||||
|
|
||||||
> OpenCode Go 是每月订阅的编程模型入口(不在订阅内按 token 额外计费)。官方硬性要求每个
|
> OpenCode Go 是每月订阅的编程模型入口(不在订阅内按 token 额外计费)。官方硬性要求每个
|
||||||
> `/chat/completions` 带 **`x-opencode-session`**(缺失直接 400),并建议专属 `User-Agent`。
|
> `/chat/completions` 带 **`x-opencode-session`**(缺失直接 400),并建议专属 `User-Agent`。
|
||||||
> 网关通过供应商 `requires_session_header: true` 在每次请求动态注入一个 UUID 会话头;
|
> 网关通过供应商 `requires_session_header: true` 动态注入该头,且**该头不能放静态
|
||||||
> 该头不能放静态 `extra_headers`(适配器缓存会冻结它)。
|
> `extra_headers`**(适配器缓存会把它冻结)。会话 id 的取值逻辑(方案 A):
|
||||||
|
> 前端对话页为「同一段对话」维持一个稳定 `conversation_id` 并随请求带来,网关把它
|
||||||
|
> 原样透传为 `x-opencode-session`——同一段对话的多次请求共享同一 session,可命中
|
||||||
|
> 上游提示词缓存;请求未带时回退为每次请求一个随机 UUID(仍能过强校验,只是没有
|
||||||
|
> 缓存收益)。`conversation_id` 是网关内部字段(在 `ChatRequest` 显式声明),
|
||||||
|
> **不会透传给上游**。
|
||||||
>
|
>
|
||||||
> 收录的 17 个文本模型(OpenAI 兼容 `/chat/completions` 线,真实 key 实测 200):
|
> 收录的 17 个文本模型(OpenAI 兼容 `/chat/completions` 线,真实 key 实测 200):
|
||||||
> `glm-5.3-flash`、`glm-5.3`、`glm-5.2`、`glm-5.1`、`kimi-k3`、`kimi-k2.7-code`、`kimi-k2.6`、
|
> `glm-5.3-flash`、`glm-5.3`、`glm-5.2`、`glm-5.1`、`kimi-k3`、`kimi-k2.7-code`、`kimi-k2.6`、
|
||||||
|
|||||||
@@ -38,12 +38,17 @@ class OpenAICompatAdapter(BaseAdapter):
|
|||||||
# 所以在每次请求时动态生成一个 UUID 会话头。
|
# 所以在每次请求时动态生成一个 UUID 会话头。
|
||||||
self._requires_session = requires_session_header
|
self._requires_session = requires_session_header
|
||||||
|
|
||||||
def _req_headers(self) -> dict[str, str]:
|
def _req_headers(self, session_id: str | None = None) -> dict[str, str]:
|
||||||
"""每次请求的 headers。需要会话头的供应商动态注入 x-opencode-session。"""
|
"""每次请求的 headers。需要会话头的供应商动态注入 x-opencode-session。
|
||||||
|
|
||||||
|
session_id 由调用方传入(网关把对话的 conversation_id 透传过来):
|
||||||
|
- 有 → 同一段对话的所有请求共用同一个 session(吃到上游提示词缓存);
|
||||||
|
- 无 → 回退为每请求一个随机 UUID(仍能过上游的强校验,只是没有缓存收益)。
|
||||||
|
"""
|
||||||
if not self._requires_session:
|
if not self._requires_session:
|
||||||
return self._headers
|
return self._headers
|
||||||
headers = dict(self._headers)
|
headers = dict(self._headers)
|
||||||
headers["x-opencode-session"] = uuid.uuid4().hex
|
headers["x-opencode-session"] = session_id or uuid.uuid4().hex
|
||||||
return headers
|
return headers
|
||||||
|
|
||||||
def _build_payload(
|
def _build_payload(
|
||||||
@@ -94,6 +99,7 @@ class OpenAICompatAdapter(BaseAdapter):
|
|||||||
tools: list[dict] | None = None,
|
tools: list[dict] | None = None,
|
||||||
tool_choice: str | dict | None = None,
|
tool_choice: str | dict | None = None,
|
||||||
extra: dict[str, Any] | None = None,
|
extra: dict[str, Any] | None = None,
|
||||||
|
session_id: str | None = None,
|
||||||
) -> dict[str, Any]:
|
) -> dict[str, Any]:
|
||||||
payload = self._build_payload(
|
payload = self._build_payload(
|
||||||
model, messages, temperature, top_p, max_tokens,
|
model, messages, temperature, top_p, max_tokens,
|
||||||
@@ -101,7 +107,7 @@ class OpenAICompatAdapter(BaseAdapter):
|
|||||||
)
|
)
|
||||||
# Bug 4 修复: 用 lease 占用 client,热重载 close 会等本次请求结束才真正关连接
|
# Bug 4 修复: 用 lease 占用 client,热重载 close 会等本次请求结束才真正关连接
|
||||||
async with self._lease() as client:
|
async with self._lease() as client:
|
||||||
resp = await client.post("/chat/completions", json=payload, headers=self._req_headers())
|
resp = await client.post("/chat/completions", json=payload, headers=self._req_headers(session_id))
|
||||||
resp.raise_for_status()
|
resp.raise_for_status()
|
||||||
data = json_or_raise(resp)
|
data = json_or_raise(resp)
|
||||||
|
|
||||||
@@ -126,6 +132,7 @@ class OpenAICompatAdapter(BaseAdapter):
|
|||||||
tools: list[dict] | None = None,
|
tools: list[dict] | None = None,
|
||||||
tool_choice: str | dict | None = None,
|
tool_choice: str | dict | None = None,
|
||||||
extra: dict[str, Any] | None = None,
|
extra: dict[str, Any] | None = None,
|
||||||
|
session_id: str | None = None,
|
||||||
) -> AsyncIterator[dict]:
|
) -> AsyncIterator[dict]:
|
||||||
payload = self._build_payload(
|
payload = self._build_payload(
|
||||||
model, messages, temperature, top_p, max_tokens,
|
model, messages, temperature, top_p, max_tokens,
|
||||||
@@ -134,7 +141,7 @@ class OpenAICompatAdapter(BaseAdapter):
|
|||||||
# Bug 4 修复: lease 在整个流式期间持有占用,热重载/关闭不会掐断在途流
|
# Bug 4 修复: lease 在整个流式期间持有占用,热重载/关闭不会掐断在途流
|
||||||
async with self._lease() as client:
|
async with self._lease() as client:
|
||||||
async with client.stream(
|
async with client.stream(
|
||||||
"POST", "/chat/completions", json=payload, headers=self._req_headers()
|
"POST", "/chat/completions", json=payload, headers=self._req_headers(session_id)
|
||||||
) as resp:
|
) as resp:
|
||||||
resp.raise_for_status()
|
resp.raise_for_status()
|
||||||
async for line in resp.aiter_lines():
|
async for line in resp.aiter_lines():
|
||||||
|
|||||||
@@ -1309,6 +1309,7 @@ async def _chat_fallback(cfg, rid, caller, request, req, chain, route, messages,
|
|||||||
max_tokens=req.max_tokens, stop=req.stop,
|
max_tokens=req.max_tokens, stop=req.stop,
|
||||||
tools=tools, tool_choice=tool_choice,
|
tools=tools, tool_choice=tool_choice,
|
||||||
extra=req.upstream_extra() or None,
|
extra=req.upstream_extra() or None,
|
||||||
|
session_id=req.conversation_id,
|
||||||
)
|
)
|
||||||
ok = True
|
ok = True
|
||||||
break
|
break
|
||||||
@@ -1412,6 +1413,7 @@ def _stream_response(request, req, chain, route, messages, tools):
|
|||||||
max_tokens=req.max_tokens, stop=req.stop,
|
max_tokens=req.max_tokens, stop=req.stop,
|
||||||
tools=tools, tool_choice=req.tool_choice,
|
tools=tools, tool_choice=req.tool_choice,
|
||||||
extra=req.upstream_extra() or None,
|
extra=req.upstream_extra() or None,
|
||||||
|
session_id=req.conversation_id,
|
||||||
)
|
)
|
||||||
async for chunk in stream_gen:
|
async for chunk in stream_gen:
|
||||||
if chunk.get("done"):
|
if chunk.get("done"):
|
||||||
|
|||||||
@@ -108,6 +108,10 @@ class ChatRequest(BaseModel):
|
|||||||
top_logprobs: int | None = None
|
top_logprobs: int | None = None
|
||||||
parallel_tool_calls: bool | None = None
|
parallel_tool_calls: bool | None = None
|
||||||
stream_options: dict[str, Any] | None = None
|
stream_options: dict[str, Any] | None = None
|
||||||
|
# 网关内部字段(显式声明 → 不进 model_extra → 不透传上游):
|
||||||
|
# 由前端对话页为「同一段对话」提供一个稳定 id;后端只拿它当 OpenCode Go 的
|
||||||
|
# x-opencode-session(同一对话共享,吃到上游提示词缓存)。非 opencode 供应商忽略。
|
||||||
|
conversation_id: str | None = None
|
||||||
# 透传供应商特有参数
|
# 透传供应商特有参数
|
||||||
extra: dict[str, Any] = Field(default_factory=dict)
|
extra: dict[str, Any] = Field(default_factory=dict)
|
||||||
|
|
||||||
|
|||||||
@@ -332,6 +332,15 @@ const params = reactive({
|
|||||||
const paramsOpen = ref(false)
|
const paramsOpen = ref(false)
|
||||||
const streaming = ref(false)
|
const streaming = ref(false)
|
||||||
const messages = ref<ChatMessage[]>([])
|
const messages = ref<ChatMessage[]>([])
|
||||||
|
|
||||||
|
// 方案 A:为「同一段对话」维持一个稳定 id,随请求带给网关 → 网关透传为
|
||||||
|
// OpenCode Go 的 x-opencode-session,使同一段对话的多次请求命中上游提示词缓存。
|
||||||
|
// 清空/新建对话时重新生成;持久化进对话状态,刷新或换设备都续用同一段会话。
|
||||||
|
function genConvId(): string {
|
||||||
|
try { if (crypto?.randomUUID) return crypto.randomUUID() } catch { /* 老浏览器回退 */ }
|
||||||
|
return 'c-' + Math.random().toString(16).slice(2) + Date.now().toString(16)
|
||||||
|
}
|
||||||
|
const convId = ref(genConvId())
|
||||||
const text = ref('')
|
const text = ref('')
|
||||||
// 第十三轮: 视觉对话 —— 待发送消息附带的图片(上传的图片在浏览器内压成 data URI;也支持粘 URL)
|
// 第十三轮: 视觉对话 —— 待发送消息附带的图片(上传的图片在浏览器内压成 data URI;也支持粘 URL)
|
||||||
interface PendingImg { uri: string; name: string; thumb: string }
|
interface PendingImg { uri: string; name: string; thumb: string }
|
||||||
@@ -394,7 +403,7 @@ async function persist() {
|
|||||||
reasoning: m.reasoning || '', meta: m.meta || null, usage: m.usage || null,
|
reasoning: m.reasoning || '', meta: m.meta || null, usage: m.usage || null,
|
||||||
cost: m.cost ?? null, error: m.error || '',
|
cost: m.cost ?? null, error: m.error || '',
|
||||||
}))
|
}))
|
||||||
const state = { params: { ...params, model: params.model }, messages: slim }
|
const state = { params: { ...params, model: params.model }, messages: slim, convId: convId.value }
|
||||||
try {
|
try {
|
||||||
sessionStorage.setItem(SKEY, JSON.stringify(state)) // 本地留一份当缓存/离线兜底
|
sessionStorage.setItem(SKEY, JSON.stringify(state)) // 本地留一份当缓存/离线兜底
|
||||||
} catch { /* 配额满就只靠服务端 */ }
|
} catch { /* 配额满就只靠服务端 */ }
|
||||||
@@ -413,6 +422,8 @@ function applyDefaults(defs: any) {
|
|||||||
}
|
}
|
||||||
|
|
||||||
function applyChat(j: any) {
|
function applyChat(j: any) {
|
||||||
|
// 同一段对话的稳定 id:历史里带着就沿用,否则视为新对话生成一个
|
||||||
|
convId.value = (typeof j?.convId === 'string' && j.convId) ? j.convId : genConvId()
|
||||||
if (j?.params?.model) Object.assign(params, j.params)
|
if (j?.params?.model) Object.assign(params, j.params)
|
||||||
if (Array.isArray(j?.messages)) {
|
if (Array.isArray(j?.messages)) {
|
||||||
let maxId = 0
|
let maxId = 0
|
||||||
@@ -570,6 +581,8 @@ function buildPayload() {
|
|||||||
top_p: params.top_p,
|
top_p: params.top_p,
|
||||||
max_tokens: params.max_tokens ?? undefined,
|
max_tokens: params.max_tokens ?? undefined,
|
||||||
stream: params.stream,
|
stream: params.stream,
|
||||||
|
// 同一段对话不变的会话 id(网关用作 OpenCode Go 的 x-opencode-session)
|
||||||
|
conversation_id: convId.value || undefined,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
@@ -686,6 +699,7 @@ function stop() {
|
|||||||
|
|
||||||
function clearConversation() {
|
function clearConversation() {
|
||||||
messages.value = []
|
messages.value = []
|
||||||
|
convId.value = genConvId() // 新对话 → 新的会话 id(不再共享上一段对话的缓存)
|
||||||
persist()
|
persist()
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|||||||
@@ -3747,6 +3747,38 @@ def test_admin_disabled_flag_in_file_is_autohealed():
|
|||||||
assert _json.loads(p.read_text(encoding="utf-8"))["users"]["admin"]["disabled"] is False
|
assert _json.loads(p.read_text(encoding="utf-8"))["users"]["admin"]["disabled"] is False
|
||||||
|
|
||||||
|
|
||||||
|
def test_chat_conversation_id_is_internal_not_upstream():
|
||||||
|
"""方案 A:conversation_id 是网关内部字段,绝不能透传给上游。"""
|
||||||
|
from app.models import ChatRequest
|
||||||
|
req = ChatRequest(model="wjd", messages=[{"role": "user", "content": "hi"}],
|
||||||
|
conversation_id="conv-123")
|
||||||
|
assert req.conversation_id == "conv-123"
|
||||||
|
assert "conversation_id" not in req.upstream_extra(), "内部字段不应透传上游"
|
||||||
|
assert "conversation_id" not in req.unknown_fields()
|
||||||
|
# 未带时为 None,且同样不透传
|
||||||
|
req2 = ChatRequest(model="wjd", messages=[{"role": "user", "content": "hi"}])
|
||||||
|
assert req2.conversation_id is None
|
||||||
|
assert "conversation_id" not in req2.upstream_extra()
|
||||||
|
|
||||||
|
|
||||||
|
def test_opencode_session_header_from_conversation_id():
|
||||||
|
"""方案 A:同一对话的 conversation_id 原样作为 x-opencode-session;缺省回退随机 UUID。"""
|
||||||
|
from app.adapters import create_adapter
|
||||||
|
cfg = load_config(CONFIG)
|
||||||
|
p = cfg.providers["opencode"]
|
||||||
|
assert p.requires_session_header is True
|
||||||
|
ad = create_adapter(p, "dummy-token")
|
||||||
|
# 传会话 id → 原样使用(同一段对话稳定,吃上游提示词缓存)
|
||||||
|
assert ad._req_headers("conv-abc")["x-opencode-session"] == "conv-abc"
|
||||||
|
# 未传 → 回退 32 位随机 hex,且两次不同
|
||||||
|
h1 = ad._req_headers()["x-opencode-session"]
|
||||||
|
h2 = ad._req_headers()["x-opencode-session"]
|
||||||
|
assert len(h1) == 32 and h1 != h2
|
||||||
|
# 非会话供应商(agnes / 火山)根本不该带这个头
|
||||||
|
ag = create_adapter(cfg.providers["agnes"], "dummy-token")
|
||||||
|
assert "x-opencode-session" not in ag._req_headers("conv-abc")
|
||||||
|
|
||||||
|
|
||||||
# 直接运行(非 pytest)
|
# 直接运行(非 pytest)
|
||||||
if __name__ == "__main__":
|
if __name__ == "__main__":
|
||||||
fns = [v for k, v in list(globals().items()) if k.startswith("test_") and callable(v)]
|
fns = [v for k, v in list(globals().items()) if k.startswith("test_") and callable(v)]
|
||||||
|
|||||||
Reference in new issue
Block a user