fix: saasStore require() bug + health check pool formula + DEV error details

- saasStore.ts: replace require('./chat/conversationStore') with await import()
  to fix ReferenceError in Vite ESM environment (P1)
- main.rs: fix health check pool usage formula from max_connections - num_idle
  to pool.size() - num_idle, preventing false "degraded" status (P1)
- error.rs: show detailed error messages in ZCLAW_SAAS_DEV=true mode
- Update bug tracker with BUG-003 through BUG-007

crates/zclaw-saas/src/error.rs

@@ -127,12 +127,16 @@ impl IntoResponse for SaasError {
     fn into_response(self) -> Response {
         let status = self.status_code();
         let (error_code, message) = match &self {
-            // 500 错误不泄露内部细节给客户端
+            // 500 错误不泄露内部细节给客户端 (开发模式除外)
             Self::Database(_) | Self::Internal(_) | Self::Io(_)
             | Self::Jwt(_) | Self::Config(_) => {
                 tracing::error!("内部错误 [{}]: {}", self.error_code(), self);
+                if std::env::var("ZCLAW_SAAS_DEV").as_deref() == Ok("true") {
+                    (self.error_code().to_string(), format!("[DEV] {}", self))
+                } else {
                     (self.error_code().to_string(), "服务内部错误".to_string())
                 }
+            }
             _ => (self.error_code().to_string(), self.to_string()),
         };
         let body = json!({

crates/zclaw-saas/src/main.rs

@@ -160,8 +160,9 @@ async fn main() -> anyhow::Result<()> {
                 interval.tick().await;
                 let pool = &metrics_db;
                 let total = pool.options().get_max_connections() as usize;
+                let size = pool.size() as usize;
                 let idle = pool.num_idle() as usize;
-                let used = total.saturating_sub(idle);
+                let used = size.saturating_sub(idle);
                 let usage_pct = if total > 0 { used * 100 / total } else { 0 };
                 tracing::info!(
                     "[PoolMetrics] total={} idle={} used={} usage_pct={}%",
@@ -248,9 +249,10 @@ async fn health_handler(
     let pool = &state.db;
     let total = pool.options().get_max_connections() as usize;
     if total > 0 {
+        let size = pool.size() as usize;
         let idle = pool.num_idle() as usize;
-        let used = total - idle;
-        let ratio = used * 100 / total;
+        let used = size.saturating_sub(idle);
+        let ratio = if size > 0 { used * 100 / total } else { 0 };
         if ratio >= 80 {
             return (
                 axum::http::StatusCode::SERVICE_UNAVAILABLE,

desktop/src/store/saasStore.ts

@@ -478,7 +478,7 @@ export const useSaaSStore = create<SaaSStore>((set, get) => {
         // switch to the first available model to prevent 404 errors
         if (models.length > 0) {
           try {
-            const { useConversationStore } = require('./chat/conversationStore');
+            const { useConversationStore } = await import('./chat/conversationStore');
             const current = useConversationStore.getState().currentModel;
             const modelIds = models.map(m => m.alias || m.id);
             if (current && !modelIds.includes(current)) {

docs/test-results/2026-04-09-exploratory/bug-tracker.md

@@ -5,3 +5,37 @@
 
 | Bug ID | 场景 | 严重度 | 标题 | 状态 | 修复提交 |
 |--------|------|--------|------|------|----------|
+| BUG-001 | Relay 全场景 | P0 | SaaS Relay DATABASE_ERROR: SUM(token_count) 返回 NUMERIC 而非 bigint | FIXED | bd6cf8e |
+| BUG-002 | 场景1.2 | P2 | 旧对话错误消息缓存: 修复后刷新页面旧错误仍显示，需点"重试"或新建对话 | WONTFIX | — |
+| BUG-003 | 启动 | P1 | saasStore.ts fetchAvailableModels 使用 require() 导致模型同步失败 | FIXED | require → await import |
+| BUG-004 | Health | P1 | Health check 连接池使用率公式错误 (max-idle 而非 size-idle) | FIXED | pool.size() 替代 max_connections |
+| BUG-005 | 启动 | P2 | OfflineStore 无模型配置时重连循环过于频繁 | KNOWN | 首次启动预期行为 |
+| BUG-006 | 启动 | P2 | WebMCP 注册失败 TypeError: Required member is undefined | KNOWN | 需 Chrome 146+ flag |
+| BUG-007 | Admin | P2 | Admin V2 authStore 测试 19 个失败 (113 passed) | OPEN | 测试代码与实现不同步 |
+
+## BUG-001 详细
+
+**根因**: PostgreSQL `SUM(bigint)` 返回 `NUMERIC` 类型，但 sqlx Rust 绑定期望 `i64` (INT8)。`key_pool.rs` 的 `select_best_key()` 查询中 `COALESCE(SUM(uw.token_count), 0)` 缺少 `::bigint` 转换。
+
+**影响**: 所有 SaaS Relay 聊天请求 100% 失败，返回 `500 DATABASE_ERROR`。
+
+**修复**: 4 处 SUM() 添加 `::bigint` 转换:
+- `relay/key_pool.rs` — 根因
+- `relay/service.rs` — sort_candidates_by_quota
+- `account/handlers.rs` — dashboard stats
+- `workers/aggregate_usage.rs` — usage aggregation
+
+## BUG-002 详细
+
+**现象**: 修复 BUG-001 后刷新页面，旧对话仍显示 "Relay error: 500 DATABASE_ERROR"。需手动点"重试"或新建对话才能恢复正常。
+
+**原因**: 前端 IndexedDB 缓存了历史错误消息，刷新不会自动清除。
+
+**决定**: WONTFIX — 这是预期行为（保留历史消息真实性），用户可通过"重试"恢复。
+
+## 测试通过场景
+
+| 场景 | 结果 | 备注 |
+|------|------|------|
+| 1.2 首次对话 (医务科) | PASS | Relay 完整链路: Desktop→SaaS→Key Pool→kimi-for-coding |
+| 离线队列恢复 | PASS | "已恢复连接, 发送中 4 条" 自动重发 |