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

+168
View File
@@ -0,0 +1,168 @@
// Anthropic 出站适配器:IR → Anthropic Messages API(SSE 事件流)。
package transform
import (
"bufio"
"bytes"
"context"
"encoding/json"
"fmt"
"net/http"
"strings"
"time"
"hunjing/shenshi/internal/ir"
)
// AnthropicAdaptor Anthropic Messages 端点适配器。
type AnthropicAdaptor struct {
EndpointName string
BaseURL string
APIKey string
UpstreamModel string
Client *http.Client
}
// Name 实现 Adaptor。
func (a *AnthropicAdaptor) Name() string { return a.EndpointName }
type antMessage struct {
Role string `json:"role"`
Content string `json:"content"`
}
// 可选标量一律指针 + omitempty(shenshi-gateway.md §5 DTO 纪律)。
type antRequest struct {
Model string `json:"model"`
Messages []antMessage `json:"messages"`
MaxTokens *int `json:"max_tokens,omitempty"`
Temperature *float64 `json:"temperature,omitempty"`
Stream bool `json:"stream"`
System *string `json:"system,omitempty"`
}
// antEvent Anthropic SSE 事件:message_start 的用量在 message.usage,
// message_delta 的用量在顶层 usage(两处形态不同,指针区分可选标量)。
type antEvent struct {
Type string `json:"type"`
Delta struct {
Type string `json:"type"`
Text string `json:"text"`
} `json:"delta"`
Message *struct {
Usage *struct {
InputTokens int `json:"input_tokens"`
OutputTokens int `json:"output_tokens"`
} `json:"usage"`
} `json:"message"`
Usage *struct {
OutputTokens int `json:"output_tokens"`
} `json:"usage"`
}
// Stream 实现 Adaptor:Anthropic SSE(message_delta / content_block_delta 等)转 IR Delta。
func (a *AnthropicAdaptor) Stream(ctx context.Context, req *ir.ChatRequest, out chan<- ir.Delta) error {
body := antRequest{
Model: a.UpstreamModel,
Messages: make([]antMessage, 0, len(req.Messages)),
MaxTokens: req.MaxTokens,
Temperature: req.Temperature,
Stream: true,
}
// Anthropic 的 system 是顶层字段而非消息序列首条
var system strings.Builder
for _, m := range req.Messages {
if m.Role == "system" {
if system.Len() > 0 {
system.WriteString("\n\n")
}
system.WriteString(m.Content)
continue
}
role := m.Role
if role == "tool" {
role = "user" // 最小映射:tool 结果按用户侧内容回传(工具循环随百工 M5 完整化)
}
body.Messages = append(body.Messages, antMessage{Role: role, Content: m.Content})
}
if system.Len() > 0 {
s := system.String()
body.System = &s
}
if body.MaxTokens == nil {
def := 4096 // Anthropic 要求 max_tokens 必填
body.MaxTokens = &def
}
payload, err := json.Marshal(body)
if err != nil {
return err
}
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost,
strings.TrimRight(a.BaseURL, "/")+"/v1/messages", bytes.NewReader(payload))
if err != nil {
return err
}
httpReq.Header.Set("content-type", "application/json")
httpReq.Header.Set("x-api-key", a.APIKey)
httpReq.Header.Set("anthropic-version", "2023-06-01")
client := a.Client
if client == nil {
client = &http.Client{Timeout: 120 * time.Second}
}
resp, err := client.Do(httpReq)
if err != nil {
return &UpstreamError{StatusCode: 0, Message: err.Error()}
}
defer func() { _ = resp.Body.Close() }()
if resp.StatusCode >= 400 {
buf := new(bytes.Buffer)
_, _ = buf.ReadFrom(resp.Body)
return &UpstreamError{StatusCode: resp.StatusCode, Message: buf.String()}
}
var finish string
var usagePrompt, usageCompletion int
scanner := bufio.NewScanner(resp.Body)
scanner.Buffer(make([]byte, 0, 64*1024), 1024*1024)
for scanner.Scan() {
line := scanner.Text()
if !strings.HasPrefix(line, "data:") {
continue
}
data := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
if data == "" {
continue
}
var ev antEvent
if err := json.Unmarshal([]byte(data), &ev); err != nil {
continue
}
switch ev.Type {
case "content_block_delta":
if ev.Delta.Type == "text_delta" && ev.Delta.Text != "" {
out <- ir.Delta{Delta: ev.Delta.Text}
}
case "message_delta":
// 结束与用量在 message_delta 顶层 usage 汇总(output_tokens)
if ev.Usage != nil {
usageCompletion = ev.Usage.OutputTokens
}
finish = "stop"
case "message_start":
if ev.Message != nil && ev.Message.Usage != nil {
usagePrompt = ev.Message.Usage.InputTokens
}
case "error":
return fmt.Errorf("上游流内错误: %s", data)
}
}
if err := scanner.Err(); err != nil {
return err
}
if finish != "" {
out <- ir.Delta{FinishReason: finish, UsagePromptTokens: usagePrompt, UsageCompletionTokens: usageCompletion}
}
return nil
}