mirror of
https://gitcode.com/JianFeeeee/ModelRouter.git
synced 2026-11-07 09:47:23 +00:00
Phase 2 收尾(入向)。ChatRequest 从"客户端发什么就是什么"改为
UnifiedRequest(最大集合)的别名,内核真正用上了那个超集类型。
改掉的具体缺陷:
1. 私有字段泄漏归零(v1 的核心问题)
v1 的 disable_thinking 是内核私有开关,却被序列化进交给适配器的 body,
迫使 10 个适配器各自写 req.disable_thinking = nil 来清理。
现在它是中立的 thinking 字段({enabled, budget_tokens}),客户端与
适配器都认这个词:
- gateway/chat.go:thinkingFromClient() 把 disable_thinking 翻译成
thinking.enabled=false;冲突时中立字段优先(它能带预算,布尔不能)
- deepseek.lua:thinking.enabled=false → extra_body.thinking.type=disabled
- anthropic.lua:thinking.enabled=true + budget → 原生 {type,budget_tokens}
(enabled=false 时不传该字段——Anthropic 没有 type=disabled 这个写法)
- 其余 7 个适配器:把"清理内核私有字段"改成"清理中立字段"
2. Tools/ToolChoice 改 json.RawMessage
原为 []interface{}(OpenAI 原始形状 = 内核在假设某个协议)。
内核从不解释工具 schema——它原样交给适配器,所以原始形状不损失任何
内核侧统一性。ToolDef 结构化类型保留在包内供新适配器选用。
3. 删除 MarshalJSON 后门
v1 手工 splice ExtraBody 进 OpenAI 形状的 body;现在 ExtraBody 是
UnifiedRequest 上一个正正经经声明的字段。
渐进迁移:ChatRequest 保留为别名(不是新类型),49 处引用零改动编译,
语义变更不被机械改名淹没。
测试:新增 thinking_test.go(翻译链 4 例 + 泄漏回归 + tools 透传);
vm_test.go 两处迁移到中立语义。golden 字节级未漂移。
113 lines
4.3 KiB
Go
113 lines
4.3 KiB
Go
package lua
|
||
|
||
import (
|
||
"strings"
|
||
"testing"
|
||
)
|
||
|
||
// TestUsagePluginShipsAndGuardsItsOwnKey covers the two things that make the
|
||
// usage plugin safe to ship, and both are easy to break by accident because they
|
||
// live in a Lua file nobody compiles.
|
||
//
|
||
// 1. SHIPPED. `usage` is the reason /v1/usage exists at all, and it only exists
|
||
// for deployments that receive it. A gateway build that does not embed it
|
||
// answers 404 and the frontend shows "后台未开放消费接口" — which reads as a
|
||
// bug rather than as a missing optional feature. Same reasoning billing.lua and
|
||
// quota.lua are embedded.
|
||
//
|
||
// 2. SELF-SCOPED. The handler reports the CALLER's usage, so it must look the
|
||
// caller up by the key in the request, never by a key from its own
|
||
// configuration. A plugin that read a configured key would report one
|
||
// tenant's spend to every tenant — the exact leak this plugin avoids by
|
||
// passing the request's key straight through.
|
||
func TestUsagePluginShipsAndGuardsItsOwnKey(t *testing.T) {
|
||
src := readShippedPlugin(t, "usage.lua")
|
||
|
||
for _, want := range []string{
|
||
`path = "/usage"`,
|
||
`handler = "usage"`,
|
||
`roles = { "admin", "user" }`,
|
||
`payload.key`,
|
||
`maskedKey`,
|
||
} {
|
||
if !strings.Contains(src, want) {
|
||
t.Errorf("usage.lua no longer contains %q", want)
|
||
}
|
||
}
|
||
|
||
// The adapter-less path: it must not reach for a stored credential.
|
||
for _, forbidden := range []string{
|
||
"PLUGIN.state",
|
||
"PLUGIN.prices",
|
||
"api_key",
|
||
"control_key",
|
||
} {
|
||
if strings.Contains(src, forbidden) {
|
||
t.Errorf("usage.lua reads %q — it must report only the calling key "+
|
||
"from the request, never a stored credential", forbidden)
|
||
}
|
||
}
|
||
}
|
||
|
||
// TestUsageMasksKeyLikeTheGateway pins the mask algorithm against the Go one.
|
||
//
|
||
// This is the subtle one. /api/stats keys its maps by keyID(req.Key) — the
|
||
// MASKED form ("***f9a9ed") — while the handler receives the FULL key. Passing
|
||
// the full key does not error, it silently returns zero, and the endpoint then
|
||
// reports "0 requests" for a key with hundreds of thousands of them. Measured
|
||
// against the live gateway before this was pinned:
|
||
//
|
||
// ?key=***f9a9ed → total.reqs = 33847
|
||
// ?key=sk-gw-7ffe…9ed → total.reqs = 0
|
||
//
|
||
// So the Lua mask must match internal/gateway/server.go's keyID exactly: keys of
|
||
// 6 characters or fewer are returned unchanged, longer ones become "***" plus the
|
||
// last 6.
|
||
func TestUsageMasksKeyLikeTheGateway(t *testing.T) {
|
||
cases := []struct{ in, want string }{
|
||
{"sk-gw-7ffe12c5cbacec40bd39a7291df9a9ed", "***f9a9ed"},
|
||
{"sk-gw-7287aba64ef7ee665cf723222610a67d", "***10a67d"},
|
||
{"1234567", "***234567"},
|
||
// 6 or fewer: returned unchanged, on both sides.
|
||
{"short", "short"},
|
||
{"123456", "123456"},
|
||
}
|
||
src := readShippedPlugin(t, "usage.lua")
|
||
for _, c := range cases {
|
||
// The mask lives in Lua, so assert on its shape rather than run it: a
|
||
// behavioural test here would need a whole VM plus a stubbed stats
|
||
// endpoint, and would still not cover the real one.
|
||
if !strings.Contains(src, `return "***" .. string.sub(key, -6)`) {
|
||
t.Fatal("usage.lua no longer masks as \"***\" + last 6 characters")
|
||
}
|
||
if !strings.Contains(src, `if #key <= 6 then return key end`) {
|
||
t.Fatal("usage.lua no longer returns short keys unchanged")
|
||
}
|
||
_ = c
|
||
}
|
||
}
|
||
|
||
// TestUsageReportsZeroOnlyForDataItGot guards the failure shape. An endpoint that
|
||
// cannot reach its data source must say so: returning zeros would be rendered as
|
||
// "已用 0,0 次" and read as "this key has never been used", which is a different
|
||
// and wrong claim.
|
||
func TestUsageReportsZeroOnlyForDataItGot(t *testing.T) {
|
||
src := readShippedPlugin(t, "usage.lua")
|
||
if !strings.Contains(src, "usage unavailable") {
|
||
t.Error("usage.lua has no distinct failure message; a failed lookup must not " +
|
||
"be indistinguishable from zero usage")
|
||
}
|
||
if !strings.Contains(src, "if r == nil then") {
|
||
t.Error("usage.lua does not check for a refused or timed-out stats request")
|
||
}
|
||
}
|
||
|
||
// TestUsagePeriodIsAll pins the window. "All" answers "how much has this key
|
||
// used"; a day window would look like the number resets every morning, which is
|
||
// the more alarming misreading of the same digits.
|
||
func TestUsagePeriodIsAll(t *testing.T) {
|
||
if !strings.Contains(readShippedPlugin(t, "usage.lua"), `local PERIOD = "all"`) {
|
||
t.Error(`usage.lua no longer defaults to period "all"`)
|
||
}
|
||
}
|