mirror of
https://gitcode.com/JianFeeeee/ModelRouter.git
synced 2026-11-07 09:47:23 +00:00
fix: 回退两处内核改动，channels 改为插件自持持久化（照参考实现）
用户指出「你又tm改内核？」—— 核对后确认两处本可避免：
1. internal/lua/vm.go +32 行 新增 base64_decode 全局函数
2. internal/lua/plugin_http_handler.go +16 行 InvokeHandler 补 markDirty
工作区已还原到 v2.0.2（git diff v2.0.2 = 0 行）。
参考实现（dsh-our-free-model/vendor/channel-pack）对同一问题的做法：
- 解 JWT：用**语言自带**的 `Buffer.from(..., 'base64url')`，从不为此改宿主。
→ 我改成了插件的纯 Lua b64url_decode（约 20 行），实测解出真 JWT 的
nickname/sub/exp 全对。
- 状态持久化：插件**自己 writeFileSync+renameSync** 写
`channel-pack/state.json`，从不依赖宿主 settings。
→ 参考还记下了同构教训（Gitee IKI7WT）：DSH 0.1.7 移除 settings.register()
后账号列表退化为纯内存；正解是**插件探测宿主能力、自持文档**，不是要求
宿主改。我的 markDirty 修法方向相反：把插件的数据安全外包给宿主某条
分发路径的实现细节。
- `.channels-store.json`（tmp + rename 原子替换），只放**索引与配置**
（账号条目含 credentialRef 而非凭据本身、模型黑名单、设置、事件日志）。
凭据仍在 .secrets/（0600），两者分离：即便文档泄露也不含 token。
- 变更点同步落盘（不用「置标志+定时 flush」）：本插件无 hooks、45 个端点
各自变更状态，还有大量内部互调；挂在某个边界就必须保证**每条**路径都
经过它，漏一条就是「操作成功但重启丢失」——正是本次要修的缺陷形态。
状态实测 ~1.7KB，写入只发生在运维动作上，不在请求热路径。
- 黑名单写入收口到 setModelDisabled（写点曾有 5+ 处，散着写必漏）。
- 存量迁移：读内核遗留 `.channels.state.json`（v2.0.4 内核写的）→ 写自有
文档 → **写成功才删**遗留（写失败还删=用一次迁移丢数据）；自有存储已存在
时也删遗留，否则内核 restore() 会在启动时用它覆盖。
1. **测试污染生产目录**（严重）：插件 storeDir() 读 /etc/llmsproxy/config.yaml
决定落盘位置，而既有测试都没隔离 → 一次全量测试把 26 个测试账号写进了
**生产** .channels-store.json（含 fixture-a/example 等夹具 id）。
prod 文件已删除（真实账号另有来源）；kernelTestCore 现在隔离插件目录，
且尊重调用方已设的值（多次「重启」需共用同一目录）。
2. model_set_disabled 改了写入口后仍引用已删除的局部变量（nil 索引崩溃）。
- TestChannelsStoreIsPluginOwned：状态写进**插件自己的**文件、不含凭据字段、
新实例读回。变异（markDirty 空操作）→ FAIL。
- TestChannelsMigrationKeepsLegacyOnWriteFailure：**写失败必须保留遗留文件**。
变异（无条件删）→ FAIL。（这条单独测：迁移成功后的删除有两条路径互相掩盖，
只测「最终文件没了」区分不出「写失败也删了」。）
- TestChannelsMigratesLegacyKernelState：遗留账号迁入内存 + 自有存储写出 +
遗留文件删除。
- 内核对照 v2.0.2：diff 为空。
go test ./internal/core/（78s）与 ./internal/lua/（63s）全绿；
全量测试后确认生产插件目录未被写入。
This commit is contained in:
JianFeeeee
@ -338,67 +338,3 @@ return plugin
|
||||
t.Errorf("persisted n = %v after three exported calls, want 3", saved.State["n"])
|
||||
}
|
||||
}
|
||||
|
||||
// TestInvokeHandlerIsPersisted 是真实事故的回归：channels 插件声明 hooks=[]，
|
||||
// 一切状态变更（账号入池、凭据引用、模型黑名单、事件日志）都经过自己的 HTTP
|
||||
// handler。而 InvokeHandler 是唯一一条**从不 markDirty** 的分发路径 —— hook 路径
|
||||
// 有、Call 路径有，只有它漏了。后果：
|
||||
//
|
||||
// .channels.state.json 从未被写出 → 重启后账号池静默清空。
|
||||
// 运维在升级前的重启里丢掉了刚登录的账号，界面显示"已登录"但列表为空。
|
||||
//
|
||||
// billing/quota 侥幸没踩到，只因为它们有 hooks（billing）或用 Call（quota）。
|
||||
// 本测试用**无 hooks、只走 handler** 的插件复现该形态。
|
||||
func TestInvokeHandlerIsPersisted(t *testing.T) {
|
||||
ps, _ := newPersistVM(t)
|
||||
if err := ps.LoadSource("handlersonly", `
|
||||
local plugin = { name = "handlersonly", version = "1.0" }
|
||||
plugin.state = { accounts = {} }
|
||||
-- 不声明任何 hooks：状态只能经 HTTP handler 变更（channels 的形态）
|
||||
plugin.hooks = {}
|
||||
plugin.http = {
|
||||
{ path = "add", method = "POST", handler = "add" },
|
||||
}
|
||||
function plugin.add(req)
|
||||
local body = req.body or ""
|
||||
local id = body:match('"id":"([^"]+)"') or "unknown"
|
||||
plugin.state.accounts[id] = { id = id, enabled = true }
|
||||
return { status = 200, json = { ok = true } }
|
||||
end
|
||||
return plugin
|
||||
`); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
|
||||
for _, id := range []string{"acct-1", "acct-2"} {
|
||||
resp, err := ps.InvokeHandler("handlersonly", "add", &handlerRequest{
|
||||
Method: "POST", Body: `{"id":"` + id + `"}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("handler: %v", err)
|
||||
}
|
||||
if resp == nil {
|
||||
t.Fatal("handler returned nil response")
|
||||
}
|
||||
}
|
||||
// Close() waits for the final flush, so no timing is involved: anything the
|
||||
// handlers did NOT mark dirty cannot appear here.
|
||||
ps.Close()
|
||||
|
||||
b, err := os.ReadFile(ps.stateFile("handlersonly"))
|
||||
if err != nil {
|
||||
t.Fatalf("★ 无 hooks 的插件经 HTTP handler 改状态后未落盘: %v\n"+
|
||||
" —— InvokeHandler 必须像 Call / hook 路径一样 markDirty，\n"+
|
||||
" 否则 channels 的账号池会在每次重启后静默清空", err)
|
||||
}
|
||||
var saved struct {
|
||||
State map[string]interface{} `json:"state"`
|
||||
}
|
||||
if err := json.Unmarshal(b, &saved); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
accts, _ := saved.State["accounts"].(map[string]interface{})
|
||||
if len(accts) != 2 {
|
||||
t.Fatalf("落盘的 accounts = %v，应为 2 个账号", saved.State["accounts"])
|
||||
}
|
||||
}
|
||||
|
||||
@ -264,10 +264,6 @@ func (ps *Plugins) InvokeHandler(plugin, fnName string, req *handlerRequest) (*h
|
||||
L.Pop(1)
|
||||
}
|
||||
if errMsg != "" {
|
||||
// A handler that raised partway may already have mutated plugin.state
|
||||
// (an account added before a later step failed), so mark it dirty before
|
||||
// bailing: the mutation it managed to make is still real.
|
||||
ps.markDirtyLocked(p)
|
||||
return nil, fmt.Errorf("plugin %s: %s", plugin, errMsg)
|
||||
}
|
||||
if L.GetTop() < 1 || L.IsNil(-1) {
|
||||
@ -277,18 +273,6 @@ func (ps *Plugins) InvokeHandler(plugin, fnName string, req *handlerRequest) (*h
|
||||
if err := luaToJSON(L, -1, &raw); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
// A handler mutates plugin.state just as a hook or an exported Call does, and
|
||||
// this was the one dispatch path that never marked the plugin dirty.
|
||||
//
|
||||
// Consequence (real defect): a plugin with NO hooks — channels is exactly
|
||||
// that, it declares hooks=[] and drives everything through its own HTTP
|
||||
// routes — accumulated accounts, credential refs, model blacklists and its
|
||||
// event log entirely in memory. Nothing ever marked it dirty, so
|
||||
// .channels.state.json was never written and a restart silently wiped the
|
||||
// account pool: the operator saw "logged in", restarted for an upgrade, and
|
||||
// the account was gone. billing/quota escape this only because they declare
|
||||
// hooks and/or use Call, both of which already mark dirty.
|
||||
ps.markDirtyLocked(p)
|
||||
return parseHandlerResponse(raw)
|
||||
}
|
||||
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@ -1358,38 +1358,6 @@ func setupGlobals(L *golua.State) {
|
||||
L.PushString(base64.StdEncoding.EncodeToString([]byte(L.ToString(1))))
|
||||
return 1
|
||||
})
|
||||
// base64_decode accepts standard, URL-safe and unpadded alphabets. It exists
|
||||
// because a plugin reading a JWT payload (e.g. to read `exp`/`nickname` when
|
||||
// the token endpoint omits them) must decode base64url, and hardcoding that
|
||||
// in Lua would mean reimplementing base64 in a plugin. Generic, no vendor
|
||||
// knowledge: an invalid string yields (nil, err), never a partial result.
|
||||
registerFn(L, "base64_decode", func(L *golua.State) int {
|
||||
s := strings.TrimSpace(L.ToString(1))
|
||||
if s == "" {
|
||||
L.PushString("")
|
||||
L.PushNil()
|
||||
return 2
|
||||
}
|
||||
// Normalise URL-safe alphabet and re-pad to a multiple of 4.
|
||||
s = strings.NewReplacer("-", "+", "_", "/").Replace(s)
|
||||
if pad := len(s) % 4; pad != 0 {
|
||||
if pad == 1 {
|
||||
L.PushNil()
|
||||
L.PushString("invalid base64 length")
|
||||
return 2
|
||||
}
|
||||
s += strings.Repeat("=", 4-pad)
|
||||
}
|
||||
b, err := base64.StdEncoding.DecodeString(s)
|
||||
if err != nil {
|
||||
L.PushNil()
|
||||
L.PushString(err.Error())
|
||||
return 2
|
||||
}
|
||||
L.PushString(string(b))
|
||||
L.PushNil()
|
||||
return 2
|
||||
})
|
||||
registerFn(L, "tohex", func(L *golua.State) int {
|
||||
L.PushString(hex.EncodeToString([]byte(L.ToString(1))))
|
||||
return 1
|
||||
|
||||
Reference in New Issue
Block a user