feat(server): add workbench API, DSH planner worker and RAG pipeline

2026-09-16 19:06:20 +08:00
parent 72fc5b47c7
commit d4b5b3f43a
122 changed files with 31906 additions and 0 deletions
@@ -0,0 +1,66 @@
+"""任务启动点数预检 + 运行中熔断（用户反馈：没点数了还在跑）。
+
+预检（readiness 阶段）：按成本预估与当前余额比较，余额必须覆盖预估 + 20% 缓冲。
+熔断（runner 每段结束后）：点数/余额类错误说明继续跑只会白跑，立刻停止整个任务。
+"""
+from __future__ import annotations
+
+from sqlmodel import Session
+
+from .billing import point_state
+
+# 启动缓冲：预估之外还要留 20%，防止输出超长导致中途断粮
+BALANCE_BUFFER_RATIO = 0.2
+
+
+def points_preflight(session: Session, campaign, estimated_points: int) -> dict:
+    """启动前点数校验。返回 {ok, blocker, info}。
+
+    ok=False 时 blocker 是可直接展示的中文原因。
+    """
+    owner_id = int(campaign.owner_user_id or 0)
+    org_id = int(campaign.organization_id or 1)
+    if not owner_id:
+        return {"ok": True, "blocker": "", "info": None}
+    state = point_state(session, owner_id, org_id)
+    balance = int(state.get("points") or 0)
+    need = int(estimated_points or 0)
+    required = int(need * (1 + BALANCE_BUFFER_RATIO))
+    info = {"balance": balance, "estimated_points": need,
+            "required_points": required, "buffer_ratio": BALANCE_BUFFER_RATIO}
+    if balance < required:
+        gap = required - balance
+        return {
+            "ok": False,
+            "blocker": (f"点数不足：本次预估需要约 {need:,} 点数（含 20% 缓冲共 {required:,}），"
+                        f"当前余额 {balance:,}，还差 {gap:,}。请联系管理员在「用量与账单」增加点数后重试"),
+            "info": info,
+        }
+    return {"ok": True, "blocker": "", "info": info}
+
+
+# 段级失败里属于「继续跑没有意义」的错误类型：钱的问题不会自愈。
+CIRCUIT_BREAK_FAILURE_TYPES = {"billing", "balance", "auth"}
+
+
+def should_break_run(failure: dict | None) -> tuple[bool, str]:
+    """根据 execute_segment 返回的 failure 判断是否熔断整个任务。
+
+    返回 (是否熔断, 原因)。
+    """
+    if not failure:
+        return False, ""
+    ftype = str(failure.get("type") or "")
+    detail = str(failure.get("detail") or "")
+    if ftype in CIRCUIT_BREAK_FAILURE_TYPES:
+        return True, detail[:160] or "模型服务拒绝（余额/鉴权问题）"
+    low = detail.lower()
+    if ("点数余额不足" in detail or "余额不足" in detail
+            or "insufficient" in low and ("balance" in low or "quota" in low)
+            or "invalid api key" in low):
+        return True, detail[:160]
+    return False, ""
+
+
+__all__ = ["points_preflight", "should_break_run", "BALANCE_BUFFER_RATIO",
+           "CIRCUIT_BREAK_FAILURE_TYPES"]