fix(sources): 站点快照加总截止时间;定时任务兜住收尾写库错误

OurBits 一次定时快照卡在过验证 / 代抓的等待里 10 小时,任务一直停在运行中。
快照整体 5 分钟截止,到点按失败收尾;调度器执行的拒绝不再变成未处理错误。

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01Cxwgnn2pbzQ7rXv3rq7fei

CHANGELOG.md

@@ -13,6 +13,10 @@
 - 作品详情头图的「订阅」改回普通主按钮（静态高光在剧照上看得清），去掉实心按钮例外 `pill-solid`；发现页海报上的评分与状态角标改成烟色小玻璃。
 - 触屏判断改为通用方案（`<html data-input>`、`touch:` 变体、`useTouchInput()`）：车机等把触屏报告成鼠标的设备也按触屏处理，悬停效果只给真鼠标。
 
+### 修复
+- 站点快照加 5 分钟总截止时间：请求在过人机验证、浏览器代抓或限流排队时卡住，快照到点记为失败，后台任务不会一直停在运行中（之前 OurBits 的一次定时快照卡了 10 小时）。
+- 定时任务在认领或收尾写库时出错（如关停时连接已断）不再成为未处理的错误，改为记一条日志。
+
 ## [0.3.37] - 2026-10-07
 
 ### 调整

apps/server/src/modules/schedules/scheduler.ts

@@ -169,10 +169,15 @@ export class Scheduler {
         release = await lanes.enterMaintenance(job.run_requested_at !== null);
         if (!release) continue;
       }
-      const p = this.execute(job, fn).finally(() => {
+      // execute 自己记录任务失败;这里兜住认领 / 收尾写库本身的错(比如关停时连接池已销毁),不留未处理的拒绝
-        release?.();
+      const p = this.execute(job, fn)
-        this.inflight.delete(job.id);
+        .catch((err) => {
-      });
+          if (!this.stopped) this.log.warn({ jobId: job.id, err: { message: errText(err) } }, "定时任务执行出错");
+        })
+        .finally(() => {
+          release?.();
+          this.inflight.delete(job.id);
+        });
       this.inflight.set(job.id, p);
       started.push(job.id);
     }

apps/server/src/modules/sources-identity/service/snapshot.ts

@@ -165,7 +165,27 @@ async function syncSnapshotFailureDecision(
   });
 }
 
-export async function snapshotSite(rt: SourcesRuntime, input: SnapshotInput): Promise<SnapshotResult> {
+/**
+ * 一次快照的总截止时间。站点请求里有几处等待不受单次请求超时约束(过人机验证、浏览器代抓、
+ * 共享限流排队),某次卡住会让任务一直停在运行中(2026-10-07 OurBits 卡了 10 小时);到点按失败收尾。
+ */
+export const SNAPSHOT_DEADLINE_MS = 5 * 60_000;
+
+/** 到截止时间就不再等 p(p 自己收到 signal 会尽量提前结束;不配合的等待也不会卡住调用方) */
+function untilDeadline<T>(p: Promise<T>, signal: AbortSignal, siteId: string, ms: number): Promise<T> {
+  return new Promise<T>((resolve, reject) => {
+    const onAbort = () => reject(new SiteError(siteId, "network", `快照 ${Math.ceil(ms / 1000)} 秒内没完成，已放弃`));
+    if (signal.aborted) return onAbort();
+    signal.addEventListener("abort", onAbort, { once: true });
+    p.then(resolve, reject).finally(() => signal.removeEventListener("abort", onAbort));
+  });
+}
+
+export async function snapshotSite(
+  rt: SourcesRuntime,
+  input: SnapshotInput,
+  opts: { deadlineMs?: number } = {},
+): Promise<SnapshotResult> {
   const { db } = rt.ctx;
   const none = { snapshotId: null, pages: [], itemCount: 0, newCount: 0, releaseIds: [] };
   const site = await db.selectFrom("site").selectAll().where("id", "=", input.siteId).executeTakeFirst();
@@ -212,9 +232,14 @@ export async function snapshotSite(rt: SourcesRuntime, input: SnapshotInput): Pr
   const pages: number[] = [];
   const releases: SiteRelease[] = [];
   let error: SiteError | Error | null = null;
+  const deadlineMs = opts.deadlineMs ?? SNAPSHOT_DEADLINE_MS;
+  const signal = AbortSignal.timeout(deadlineMs);
   for (let p = 1; p <= maxPages; p++) {
     try {
-      const res: SitePage = input.query ? await adapter.search(input.query, { page: p }) : await adapter.list(p);
+      const call = input.query
+        ? adapter.search(input.query, { page: p }, { signal })
+        : adapter.list(p, undefined, { signal });
+      const res: SitePage = await untilDeadline(call, signal, site.id, deadlineMs);
       pages.push(p);
       releases.push(...res.items);
       if (res.items.length === 0 || (res.totalPages !== null && p >= res.totalPages)) break;

apps/server/test/sources-identity/snapshot-deadline.test.ts

@@ -0,0 +1,62 @@
+import { afterAll, beforeAll, describe, expect, it } from "vitest";
+import { snapshotSite } from "../../src/modules/sources-identity/service/snapshot.ts";
+import { createHarness, type Harness, pgAvailable } from "./harness.ts";
+
+// 快照总截止时间:站点请求卡住且不理会取消信号(过人机验证、浏览器代抓这类等待)时,
+// 快照到点按失败收尾、写完 site_snapshot,任务不会一直停在运行中(2026-10-07 OurBits 卡了 10 小时)。
+
+const up = await pgAvailable();
+
+// biome-ignore lint/suspicious/noExplicitAny: 响应体断言
+const json = async (r: Response): Promise<any> => {
+  const body = await r.json();
+  if (!r.ok) throw new Error(`${r.status} ${JSON.stringify(body)}`);
+  return body.data;
+};
+
+describe.skipIf(!up)("快照截止时间", () => {
+  let h: Harness;
+  let siteId: string;
+  // 永不返回,也不理会 signal
+  const hang = (() => new Promise<Response>(() => {})) as typeof fetch;
+
+  beforeAll(async () => {
+    h = await createHarness("snapdeadline", { fetch: hang });
+    const s = await json(
+      await h.req("/sites", {
+        body: {
+          key: "audiences",
+          name: "观众",
+          adapter: "nexusphp",
+          baseUrl: "https://aud.test/",
+          rateLimit: { minIntervalMs: 0, concurrency: 10 },
+        },
+      }),
+    );
+    siteId = s.id;
+    await json(
+      await h.req(`/sites/${siteId}/credentials`, {
+        method: "PUT",
+        body: { kind: "cookie", cookie: "c_secure_uid=1; c_secure_pass=abc" },
+      }),
+    );
+  });
+  afterAll(async () => {
+    await h?.close();
+  });
+
+  it("请求卡住 → 到点失败,快照记录收尾", async () => {
+    const t0 = Date.now();
+    const r = await snapshotSite(h.rt, { siteId, kind: "periodic" }, { deadlineMs: 300 });
+    expect(Date.now() - t0).toBeLessThan(5_000);
+    expect(r.status).toBe("failed");
+    expect(r.error).toContain("没完成");
+    const snap = await h.db
+      .selectFrom("site_snapshot")
+      .selectAll()
+      .where("id", "=", r.snapshotId!)
+      .executeTakeFirstOrThrow();
+    expect(snap.status).toBe("failed");
+    expect(snap.finished_at).not.toBeNull();
+  });
+});