M4 收口:神识用户面与计量闭环——BYOK/配额三段式/usage 落库/Anthropic+Gemini 适配器/神识阁

- 神识 Go:store 双实现(PG 自动建表+60s 就绪重试 / 内存回退)承载 usage_logs·byok_keys·quota_balances;BYOK 密钥库(AES-GCM 加密、API 脱敏、撤回即失效、主密钥未配置 fail-closed);配额三段式(预扣 max_tokens 钳位 1M→结算实际 token→全败幂等退款,HJ-2004);计量归属(IR user_id→usage_logs,/me/usage 用户维度聚合+配额总览,/admin/usage/sample 对账抽样);Anthropic/Gemini 出站适配器(SSE 解析、system 提升、用量映射);stub 端点补名修复计量空 endpoint
- IR 契约:user_id/byok_key_id 双端同步(Go types.go + Py types.py)
- 灵台:runs.owner_id(迁移 005,server_default 兼容存量)经 kernel 全链透传为 IR user_id;shenshi_client 扩展用户面方法(X-Hunjing-User-Id 代理,Echo 桩带内存实现);/me/byok-keys 登记·列表·撤回与 /me/usage 端点(4 用例)
- 前端神识阁(/models 重建):偏好(生成参数表单)/自带密钥(登记加密保存、脱敏列表、撤回二次确认)/用量(聚合表+配额进度条)三区;me API 库+mock+5 用例;侧栏恢复入口
- 配置与部署:.env.example 增 SHENSHI_DATABASE_URL/SHENSHI_BYOK_MASTER_KEY/SHENSHI_USER_BUDGET_TOKENS;compose 神识容器接自有库(init-shenshi.sql 预建 hunjing_shenshi,与灵台不共库共表)
- 证据:Go 34 用例(BYOK 独占/配额拒绝/结算退款/归属聚合)、后端 532、前端 210、BFF 4 全绿;跨进程 E2E 四场景(灵台 HttpShenshiClient→真 Go 神识:登记脱敏/BYOK 独占不可达整体失败/用量归属聚合/撤回即失效)通过;rewrite-plan M4 标记完成、能力矩阵神识行升已验证
This commit is contained in:
chenxin committed 2026-08-27 18:39:17 +08:00
1 parent 462d570751
commit 144415247e
51 files changed
+3533 -90

No files matched your search

+195
View File
@@ -0,0 +1,195 @@
// BYOK / 配额三段式 / 用量归属的服务级场景测试。
package server
import (
"bufio"
"encoding/hex"
"encoding/json"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"hunjing/shenshi/internal/byok"
"hunjing/shenshi/internal/quota"
"hunjing/shenshi/internal/route"
"hunjing/shenshi/internal/store"
)
func testMasterKey() string {
b := make([]byte, 32)
for i := range b {
b[i] = byte(i + 1)
}
return hex.EncodeToString(b)
}
// fullServer 装配 store/byok/quota 的完整服务;eps 为注册表候选(按优先级升序尝试)。
func fullServer(t *testing.T, budget int64, eps ...route.Endpoint) (*httptest.Server, store.Store, *byok.Vault) {
t.Helper()
st := store.NewMem()
vault, err := byok.NewVault(testMasterKey(), st)
if err != nil {
t.Fatal(err)
}
registry := registryWith(t, time.Second, eps...)
srv := New(Config{Store: st, Vault: vault, Quota: quota.New(st, budget)}, registry)
ts := httptest.NewServer(srv)
t.Cleanup(ts.Close)
return ts, st, vault
}
func postChat(ts *httptest.Server, body, userID string) (*http.Response, string) {
req, _ := http.NewRequest(http.MethodPost, ts.URL+"/v1/chat/completions", strings.NewReader(body))
req.Header.Set("content-type", "application/json")
if userID != "" {
req.Header.Set("X-Hunjing-User-Id", userID)
}
resp, err := ts.Client().Do(req)
if err != nil {
return nil, ""
}
var sb strings.Builder
scanner := bufio.NewScanner(resp.Body)
for scanner.Scan() {
sb.WriteString(scanner.Text())
sb.WriteString("\n")
}
return resp, sb.String()
}
func byokChatBody(run, extra string) string {
return `{"ir_version":1,"slot":"chat","messages":[{"role":"user","content":"hi"}],"run_id":"` + run + `"` + extra + `,"stream":true}`
}
// 场景:BYOK 独占路由——请求携带 byok_key_id 时只走用户端点,不回落平台候选。
func TestByokExclusiveRouting(t *testing.T) {
platform := fakeOpenAI(t, 0, 0, "平台端点")
defer platform.Close()
ts, _, vault := fullServer(t, 0, oaEndpoint("platform", platform.URL, 10))
byokUpstream := fakeOpenAI(t, 0, 0, "用户端点")
defer byokUpstream.Close()
key, err := vault.Register("u1", "自带", "openai_compatible", byokUpstream.URL, "sk-mine", "my-model")
if err != nil {
t.Fatal(err)
}
body := byokChatBody("run_byok", `,"user_id":"u1","byok_key_id":"`+key.ID+`"`)
resp, out := postChat(ts, body, "u1")
if resp == nil || resp.StatusCode != 200 {
t.Fatalf("BYOK 请求失败: %s", out)
}
if !strings.Contains(out, "用户端点") || strings.Contains(out, "平台端点") {
t.Fatalf("应独占 BYOK 端点: %s", out)
}
}
// 场景:BYOK 无效引用 → 整体失败,不回落平台端点。
func TestByokInvalidReferenceRejected(t *testing.T) {
platform := fakeOpenAI(t, 0, 0, "平台端点")
defer platform.Close()
ts, _, _ := fullServer(t, 0, oaEndpoint("platform", platform.URL, 10))
resp, out := postChat(ts, byokChatBody("run_bad", `,"user_id":"u1","byok_key_id":"byok_missing"`), "u1")
if resp == nil || resp.StatusCode != http.StatusServiceUnavailable {
t.Fatalf("应 503: %s", out)
}
if !strings.Contains(out, "HJ-2001") || strings.Contains(out, "平台端点") {
t.Fatalf("应整体失败而非回落: %s", out)
}
}
// 场景:配额不足 → HJ-2004 预拒,且不产生用量记录。
func TestQuotaRejectsWhenExhausted(t *testing.T) {
upstream := fakeOpenAI(t, 0, 0, "内容")
defer upstream.Close()
ts, st, _ := fullServer(t, 10, oaEndpoint("up", upstream.URL, 10))
resp, out := postChat(ts, byokChatBody("run_q", `,"user_id":"u1","max_tokens":100`), "u1")
if resp == nil || resp.StatusCode != http.StatusPaymentRequired {
t.Fatalf("配额不足应 402: %s", out)
}
if !strings.Contains(out, "HJ-2004") {
t.Fatalf("应携带 HJ-2004: %s", out)
}
sample, _ := st.SampleUsage(10)
if len(sample) != 0 {
t.Fatalf("预拒请求不应产生用量记录: %+v", sample)
}
}
// 场景:成功结算释放差值;全败全额退款——余额不残留。
func TestQuotaSettleAndRefundBalance(t *testing.T) {
upstream := fakeOpenAI(t, 0, 0, "内容")
bad := fakeOpenAI(t, 0, http.StatusForbidden, "")
defer upstream.Close()
defer bad.Close()
ts, st, _ := fullServer(t, 1000, oaEndpoint("up", upstream.URL, 10))
// 成功:预扣 max_tokens=200,实际 3+4=7
if resp, out := postChat(ts, byokChatBody("run_ok", `,"user_id":"u1","max_tokens":200`), "u1"); resp == nil || resp.StatusCode != 200 {
t.Fatalf("成功请求失败: %s", out)
}
b, _ := st.Balance("u1")
if b.Used != 7 || b.Reserved != 0 {
t.Fatalf("结算后账目错误: %+v", b)
}
// 已用 7,预算 1000:请求 max_tokens 994 应被拒(7+994 > 1000)
if resp, out := postChat(ts, byokChatBody("run_over", `,"user_id":"u1","max_tokens":994`), "u1"); resp == nil || resp.StatusCode != http.StatusPaymentRequired {
t.Fatalf("超预算应拒: %s", out)
}
// 全候选失败 → 全额退款
tsBad, stBad, _ := fullServer(t, 1000, oaEndpoint("bad", bad.URL, 10))
if resp, out := postChat(tsBad, byokChatBody("run_af", `,"user_id":"u2","max_tokens":500`), "u2"); resp == nil || resp.StatusCode != http.StatusServiceUnavailable {
t.Fatalf("全败应 503: %s", out)
}
b2, _ := stBad.Balance("u2")
if b2.Reserved != 0 || b2.Used != 0 {
t.Fatalf("全败后应全额退款: %+v", b2)
}
}
// 场景:用量归属计量——/me/usage 按用户过滤并附配额总览。
func TestUsageAttributionByUser(t *testing.T) {
upstream := fakeOpenAI(t, 0, 0, "内容")
defer upstream.Close()
ts, _, _ := fullServer(t, 0, oaEndpoint("up", upstream.URL, 10))
for _, b := range []string{
byokChatBody("run_u1", `,"user_id":"u1"`),
byokChatBody("run_u2", `,"user_id":"u2"`),
byokChatBody("run_u1b", `,"user_id":"u1"`),
} {
if resp, out := postChat(ts, b, ""); resp == nil || resp.StatusCode != 200 {
t.Fatalf("请求失败: %s", out)
}
}
req, _ := http.NewRequest(http.MethodGet, ts.URL+"/me/usage", nil)
req.Header.Set("X-Hunjing-User-Id", "u1")
resp, err := ts.Client().Do(req)
if err != nil {
t.Fatal(err)
}
var payload struct {
Rows []struct {
UserID string `json:"user_id"`
Requests int `json:"requests"`
} `json:"rows"`
Quota *struct {
Budget int64 `json:"budget"`
Used int64 `json:"used"`
} `json:"quota"`
}
_ = json.NewDecoder(resp.Body).Decode(&payload)
if len(payload.Rows) != 1 || payload.Rows[0].UserID != "u1" || payload.Rows[0].Requests != 2 {
t.Fatalf("u1 聚合错误: %+v", payload.Rows)
}
if payload.Quota == nil || payload.Quota.Used != 14 {
t.Fatalf("配额总览错误: %+v", payload.Quota)
}
}