feat: add concurrency safety, caption detection, admin enhancements, and performance improvements

app/services/admin.py

@@ -7,7 +7,7 @@ from datetime import date
 from pathlib import Path
 from typing import Callable
 
-from sqlalchemy import func, select, text
+from sqlalchemy import func, select, text, update
 from sqlalchemy.orm import Session
 
 from app.config import settings
@@ -100,7 +100,9 @@ def get_admin_stats(db: Session) -> dict:
 
     # ── 活跃锁 ────────────────────────────────────────────────────────
     active_locks = (
-        db.execute(select(TaskLock).where(TaskLock.status == "running")).scalars().all()
+        db.execute(select(TaskLock).where(TaskLock.status.in_(["running", "stale"])))
+        .scalars()
+        .all()
     )
 
     return {
@@ -124,6 +126,7 @@ def get_admin_stats(db: Session) -> dict:
         "recent_logs": recent_logs,
         "active_locks": active_locks,
         "upvote_refresh_days": settings.UPVOTE_REFRESH_DAYS,
+        "config_overview": get_config_overview(),
     }
 
 
@@ -370,6 +373,7 @@ def get_logs_context(db: Session, *, page: int, per_page: int) -> dict:
         "summary_done": summary_done,
         "summary_pending": summary_pending,
         "summary_failed": summary_failed,
+        "failure_breakdown": get_failure_breakdown(db),
     }
 
 
@@ -511,3 +515,126 @@ def reset_summaries_pending(db: Session, arxiv_ids: list[str]) -> int:
             db.add(SummaryStatus(paper_id=paper_id, status=SummaryState.PENDING))
     db.commit()
     return len(paper_ids)
+
+
+# ── 任务监控 ──────────────────────────────────────────────────────────
+
+
+def query_jobs(
+    db: Session,
+    *,
+    status: str | None = None,
+    job_type: str | None = None,
+    page: int = 1,
+    per_page: int = 20,
+) -> tuple[list[dict], int]:
+    """后台任务列表查询 — 支持 status/type 过滤 + 分页，返回已 enrich 的 dict 列表。"""
+    query = select(Job)
+    if status and status != "all":
+        query = query.where(Job.status == status)
+    if job_type and job_type != "all":
+        query = query.where(Job.type == job_type)
+
+    total = db.scalar(select(func.count()).select_from(query.subquery())) or 0
+    jobs = (
+        db.execute(
+            query.order_by(Job.created_at.desc())
+            .offset((page - 1) * per_page)
+            .limit(per_page)
+        )
+        .scalars()
+        .all()
+    )
+    return [serialize_job(j) for j in jobs], total
+
+
+def _as_naive(dt):
+    """去掉 tzinfo — SQLite 读回的 datetime 是 naive UTC，与 utc_now() 运算前需统一。"""
+    if dt is not None and getattr(dt, "tzinfo", None) is not None:
+        return dt.replace(tzinfo=None)
+    return dt
+
+
+def serialize_job(job: Job) -> dict:
+    """单条 job 序列化为展示用 dict（含耗时）。"""
+    duration = None
+    started = _as_naive(job.started_at)
+    if started:
+        end = _as_naive(job.completed_at) or _as_naive(utc_now())
+        duration = round((end - started).total_seconds(), 1)
+    return {
+        "id": job.id,
+        "type": job.type,
+        "status": job.status,
+        "owner": job.owner,
+        "created_at": job.created_at,
+        "started_at": job.started_at,
+        "completed_at": job.completed_at,
+        "duration_seconds": duration,
+        "error": job.error,
+    }
+
+
+def get_job_status_counts(db: Session) -> dict:
+    """按 status 聚合 job 计数，供任务页顶部小统计行用。"""
+    rows = db.execute(
+        select(Job.status, func.count(Job.id)).group_by(Job.status)
+    ).fetchall()
+    return {row[0]: row[1] for row in rows}
+
+
+# ── 锁管理 ────────────────────────────────────────────────────────────
+
+
+def force_release_lock(db: Session, lock_id: int) -> bool:
+    """强制释放一个卡死的 TaskLock（仅对 running/stale 生效）。"""
+    result = db.execute(
+        update(TaskLock)
+        .where(TaskLock.id == lock_id, TaskLock.status.in_(["running", "stale"]))
+        .values(status="finished", released_at=utc_now())
+    )
+    db.commit()
+    return (result.rowcount or 0) > 0
+
+
+# ── 失败原因分布 ──────────────────────────────────────────────────────
+
+
+def get_failure_breakdown(db: Session) -> list[dict]:
+    """按 error_type 聚合失败/永久失败的总结，按数量降序。NULL 归 unknown。"""
+    error_expr = func.coalesce(SummaryStatus.error_type, "unknown")
+    rows = db.execute(
+        select(error_expr, func.count(SummaryStatus.id))
+        .where(
+            SummaryStatus.status.in_(
+                [SummaryState.FAILED, SummaryState.PERMANENT_FAILURE]
+            )
+        )
+        .group_by(error_expr)
+        .order_by(func.count(SummaryStatus.id).desc())
+    ).fetchall()
+    return [{"error_type": row[0], "count": row[1]} for row in rows]
+
+
+# ── 运行配置概览 ──────────────────────────────────────────────────────
+
+
+def get_config_overview() -> dict:
+    """聚合非敏感配置，供仪表盘展示。敏感字段只标是否已配置，不显示值。"""
+    return {
+        "summary_backend": settings.SUMMARY_BACKEND,
+        "summary_pdf_mode": settings.SUMMARY_PDF_MODE,
+        "summary_concurrency": settings.SUMMARY_CONCURRENCY,
+        "summary_timeout_seconds": settings.SUMMARY_TIMEOUT_SECONDS,
+        "summary_max_retries": settings.SUMMARY_MAX_RETRIES,
+        "scheduler_enabled": settings.SCHEDULER_ENABLED,
+        "schedule_time": f"{settings.SCHEDULE_HOUR:02d}:{settings.SCHEDULE_MINUTE:02d}",
+        "chroma_enabled": settings.CHROMA_ENABLED,
+        "embed_model": settings.EMBED_MODEL or "（未配置）",
+        "top_n": settings.TOP_N,
+        "upvote_refresh_days": settings.UPVOTE_REFRESH_DAYS,
+        "app_workers": settings.APP_WORKERS,
+        "layout_model": Path(settings.LAYOUT_MODEL_PATH).name,
+        "database_url": settings.DATABASE_URL,
+        "api_key_configured": bool(settings.EMBED_API_KEY),
+    }