fix: recover stuck update queue

install.sh

@@ -1012,7 +1012,10 @@ validate_listen_port() {
 }
 
 validate_public_origin() {
-  local value=$1 authority host path_part origin_port suffix
+  # Keep the optional origin port defined under `set -u`. Origins without an
+  # explicit port (for example https://example.test) are valid and should
+  # proceed to the default-port handling below.
+  local value=$1 authority host path_part origin_port='' suffix
   case "$value" in
     http://*|https://*) ;;
     *) die '公开访问地址必须是 http:// 或 https:// 地址' ;;

package.json

@@ -1,6 +1,6 @@
 {
   "name": "tallynote",
-  "version": "1.1.15",
+  "version": "1.1.19",
   "private": true,
   "type": "module",
   "packageManager": "pnpm@9.0.6",

scripts/tallynote-update-runner.sh

@@ -190,6 +190,15 @@ recover_stale_state() {
   [[ "$state_job" =~ ^[0-9a-f-]{36}$ ]] || die 'update state job id is invalid'
   [[ "$state_old" == "$PREFIX/releases/"* && -d "$state_old" && ! -L "$state_old" ]] || die 'update state target is invalid'
   current_target=$(readlink -f -- "$CURRENT_LINK" 2>/dev/null || true)
+  if [[ "$state_phase" == download && "$current_target" == "$state_old" ]]; then
+    # Downloading never changes the active release. If the runner was killed
+    # after the CLI staged its payload but before it removed the recovery
+    # marker, keep the request available for an idempotent retry. Treating
+    # every stale download marker as a failed apply would discard a usable
+    # staged payload and leave the browser showing a misleading failure.
+    clear_update_state || true
+    return 0
+  fi
   if [[ "$state_phase" == finalizing && "$current_target" != "$state_old" ]]; then
     recovery_node="$CURRENT_LINK/runtime/bin/node"
     [[ -x "$recovery_node" ]] || recovery_node=$(command -v node || true)

scripts/test-installer.sh

@@ -88,6 +88,12 @@ bash -c '
   chmod 700 "$mode_dir"
   [[ "$(stat_mode_bits "$mode_dir")" == 448 ]]
   mkdir -p "$owner_parent"
+  # CI runs this shell suite as root. Make the parent genuinely non-root in
+  # that environment so the assertion exercises the ownership guard instead
+  # of accidentally passing because root-owned parents are allowed.
+  if [[ "${EUID:-$(id -u)}" == 0 ]]; then
+    chown 65534:65534 "$owner_parent"
+  fi
   if (assert_path_chain "$owner_parent/child") >/dev/null 2>&1; then
     echo "expected non-root path parent to fail" >&2
     exit 1
@@ -189,6 +195,40 @@ grep -q -- '--finalize-job' "$runner_root/node.log"
 [[ ! -e "$runner_data/update-request.json" ]]
 [[ ! -e "$runner_prefix/.update-state" ]]
 
+# A stale download marker must be recoverable without finalizing the staged
+# download as a failed apply. The next runner invocation should retry the
+# request and let the CLI preserve/refresh its staged workspace.
+download_runner_root="$tmp/download-runner"
+download_runner_prefix="$download_runner_root/prefix"
+download_runner_data="$download_runner_root/data"
+download_runner_tools="$download_runner_root/tools"
+mkdir -p "$download_runner_prefix/releases/1.0.0/runtime/bin" "$download_runner_prefix/releases/1.0.0/dist/server/cli" "$download_runner_data" "$download_runner_tools"
+ln -s "$download_runner_prefix/releases/1.0.0" "$download_runner_prefix/current"
+# The request has already been consumed; only the stale download marker is
+# left, which is the narrow recovery window covered by this fixture.
+download_runner_prefix_physical=$(cd "$download_runner_prefix" && pwd -P)
+download_runner_data_physical=$(cd "$download_runner_data" && pwd -P)
+printf '%s\n' 'job_id=00000000-0000-4000-8000-000000000002' "old_target=$download_runner_prefix_physical/releases/1.0.0" 'phase=download' > "$download_runner_prefix/.update-state"
+printf '%s\n' '#!/usr/bin/env bash' 'printf "%s\n" "$*" >> "$TALLYNOTE_DOWNLOAD_NODE_TRACE"' 'exit 0' > "$download_runner_prefix/releases/1.0.0/runtime/bin/node"
+printf '%s\n' cli > "$download_runner_prefix/releases/1.0.0/dist/server/cli/update.js"
+printf '%s\n' '#!/usr/bin/env bash' 'case "${1:-}" in is-active) exit 0;; *) exit 0;; esac' > "$download_runner_tools/systemctl"
+printf '%s\n' '#!/usr/bin/env bash' 'if [[ "${1:-}" == "-f" ]]; then shift; [[ "${1:-}" == "--" ]] && shift; /bin/realpath "$1"; else /usr/bin/readlink "$@"; fi' > "$download_runner_tools/readlink"
+cat >"$download_runner_tools/stat" <<'EOF'
+#!/usr/bin/env bash
+case "$*" in
+  *"-c %u"*|*"-f %u"*) printf '0\n' ;;
+  *"-c %a"*|*"-f %Lp"*) printf '600\n' ;;
+  *) /usr/bin/stat "$@" ;;
+esac
+EOF
+chmod 755 "$download_runner_prefix/releases/1.0.0/runtime/bin/node" "$download_runner_tools/systemctl" "$download_runner_tools/readlink" "$download_runner_tools/stat"
+download_runner_script="$download_runner_root/runner.sh"
+sed "s#PATH=/usr/sbin:/usr/bin:/sbin:/bin#PATH=$download_runner_tools:/usr/sbin:/usr/bin:/sbin:/bin#" "$root/scripts/tallynote-update-runner.sh" > "$download_runner_script"
+chmod 755 "$download_runner_script"
+env EUID=0 TALLYNOTE_INSTALL_PREFIX="$download_runner_prefix_physical" TALLYNOTE_DATA_DIR="$download_runner_data_physical" TALLYNOTE_DOWNLOAD_NODE_TRACE="$download_runner_root/node.log" bash "$download_runner_script"
+[[ ! -e "$download_runner_root/node.log" ]]
+[[ ! -e "$download_runner_prefix/.update-state" ]]
+
 # A RETURN trap installed by install_release must be cleared while its local
 # temporary variables still exist; otherwise set -u fails at the end of main.
 release_fixture="$tmp/release-fixture"
@@ -247,7 +287,11 @@ ln -s "$wrapper_prefix/releases/1.0.0" "$wrapper_prefix/current"
 printf '%s\n' '#!/usr/bin/env bash' 'pwd -P > "$TALLYNOTE_WRAPPER_LOG"' 'printf "%s\n" "$@" >> "$TALLYNOTE_WRAPPER_LOG"' > "$wrapper_prefix/releases/1.0.0/runtime/bin/node"
 chmod 755 "$wrapper_prefix/releases/1.0.0/runtime/bin/node"
 printf '%s\n' cli > "$wrapper_prefix/releases/1.0.0/dist/server/cli/admin-init.js"
-TALLYNOTE_INSTALL_PREFIX="$wrapper_prefix" TALLYNOTE_CONFIG_DIR="$tmp/no-config" TALLYNOTE_WRAPPER_LOG="$tmp/wrapper.log" \
+# This fixture verifies release-relative execution and argument forwarding.
+# Force the wrapper's non-root branch so the root CI runner does not need a
+# real `tallynote` service account or a privileged runuser hand-off; that
+# privilege boundary is validated by the production checks themselves.
+env EUID=1000 TALLYNOTE_INSTALL_PREFIX="$wrapper_prefix" TALLYNOTE_CONFIG_DIR="$tmp/no-config" TALLYNOTE_WRAPPER_LOG="$tmp/wrapper.log" \
   bash "$root/bin/tallynote-admin-init" --generate
 wrapper_expected_root=$(cd "$wrapper_prefix/releases/1.0.0" && pwd -P)
 grep -Fxq "$wrapper_expected_root" "$tmp/wrapper.log"
@@ -414,6 +458,9 @@ bash -c '
   stat_uid() { printf "0"; }
   stat_mode_bits() { printf "384"; }
   validate_public_origin "http://[2001:db8::10]:3000"
+  # A standard HTTPS origin may omit its default port; this must remain valid
+  # under the installer strict unset-variable mode.
+  validate_public_origin "https://example.test"
 ' _ "$installer_lib"
 printf '%s\n' \
   'TALLYNOTE_HOST=0.0.0.0' \

server/app.ts

@@ -119,7 +119,7 @@ function enforceUpdateCooldown(
   adminId: string,
   operation: "check" | "download" | "apply",
   reply: FastifyReply,
-): void {
+): number {
   const state = updateRateState(database, adminId);
   const now = Date.now();
   const previous = operation === "check" ? state.checkedAt : operation === "download" ? state.downloadedAt : state.appliedAt;
@@ -134,6 +134,7 @@ function enforceUpdateCooldown(
   if (operation === "check") state.checkedAt = now;
   else if (operation === "download") state.downloadedAt = now;
   else state.appliedAt = now;
+  return now;
 }
 
 function adminSelect(alias = ""): string {
@@ -957,12 +958,15 @@ export async function buildApp(database: DatabaseContext, config: AppConfig) {
   });
 
   app.post("/api/update/check", { preHandler: guard(database, config), bodyLimit: 32 * 1024 }, async (request, reply) => {
+    const rateState = updateRateState(database.sqlite, request.auth!.admin.id);
+    const previousCheckedAt = rateState.checkedAt;
+    let reservedCheckedAt: number | null = null;
     try {
       reconcileOrphanedUpdateJobs(database.sqlite, config);
       // Disabled/dev installs do not contact a release endpoint, so repeated
       // checks are local status reads and should remain immediately usable.
       if (config.updateStrategy !== "disabled") {
-        enforceUpdateCooldown(database.sqlite, config, request.auth!.admin.id, "check", reply);
+        reservedCheckedAt = enforceUpdateCooldown(database.sqlite, config, request.auth!.admin.id, "check", reply);
       }
       const result = await checkForUpdate(database.sqlite, config);
       writeAudit(database.sqlite, {
@@ -981,6 +985,10 @@ export async function buildApp(database: DatabaseContext, config: AppConfig) {
       reply.header("Cache-Control", "no-store");
       return { ...result, strategy: config.updateStrategy };
     } catch (error) {
+      // A failed upstream request is not a successful check. Release the
+      // reservation only when this request still owns it, so a concurrent
+      // successful check cannot have its cooldown overwritten.
+      if (reservedCheckedAt !== null && rateState.checkedAt === reservedCheckedAt) rateState.checkedAt = previousCheckedAt;
       writeAudit(database.sqlite, {
         requestId: request.id,
         actorAdminId: request.auth!.admin.id,

server/update-service.ts

@@ -1,4 +1,4 @@
-import { lstatSync, realpathSync, unlinkSync } from "node:fs";
+import { lstatSync, realpathSync, readFileSync, unlinkSync } from "node:fs";
 import { chmod, mkdir, rename, writeFile } from "node:fs/promises";
 import path from "node:path";
 import { createPublicKey, randomUUID, verify as verifySignature } from "node:crypto";
@@ -385,6 +385,17 @@ function removeExpiredRequest(filePath: string, now: number): void {
   }
 }
 
+function requestJobId(filePath: string): string | null {
+  try {
+    const info = lstatSync(filePath);
+    if (!info.isFile() || info.isSymbolicLink()) return null;
+    const value = JSON.parse(readFileSync(filePath, "utf8")) as { jobId?: unknown };
+    return typeof value.jobId === "string" && /^[0-9a-f-]{36}$/.test(value.jobId) ? value.jobId : null;
+  } catch {
+    return null;
+  }
+}
+
 function currentReleaseVersion(config: AppConfig): string | null {
   try {
     const target = realpathSync(config.currentLink);
@@ -425,6 +436,7 @@ export function reconcileOrphanedUpdateJobs(database: Database.Database, config:
   // row is still safe to retry and must not block the queue forever.
   const releaseVersion = currentReleaseVersion(config);
   let reconciled = 0;
+  const reconciledIds = new Set<string>();
   for (const row of rows) {
     if (typeof row.updatedAt !== "number" || now - row.updatedAt < ORPHANED_UPDATE_TIMEOUT_MS) continue;
     // The runner refreshes the state marker while a download is in flight.
@@ -451,7 +463,10 @@ export function reconcileOrphanedUpdateJobs(database: Database.Database, config:
         });
         return true;
       })();
-      if (changed) reconciled += 1;
+      if (changed) {
+        reconciled += 1;
+        reconciledIds.add(row.id);
+      }
       continue;
     }
     if (requestFresh || stateFresh) continue;
@@ -476,7 +491,10 @@ export function reconcileOrphanedUpdateJobs(database: Database.Database, config:
       });
       return true;
     })();
-    if (changed) reconciled += 1;
+    if (changed) {
+      reconciled += 1;
+      reconciledIds.add(row.id);
+    }
   }
   // Prevent a stale request from being replayed after its DB row has been
   // marked failed. The path is fixed by the server configuration and the
@@ -484,7 +502,17 @@ export function reconcileOrphanedUpdateJobs(database: Database.Database, config:
   // A download runner refreshes the state marker while it is still using the
   // request. Keep the request until that lease also expires; otherwise a
   // long download can lose its job id and fail to finalize its row.
-  if (!stateFresh && (!requestPresent || (requestMtime !== null && now - requestMtime >= ORPHANED_UPDATE_TIMEOUT_MS))) {
+  const queuedRequestId = requestPresent ? requestJobId(config.updateRequestPath) : null;
+  const queuedRequest = queuedRequestId ? rows.find((row) => row.id === queuedRequestId) : undefined;
+  const requestStillNeeded = Boolean(
+    queuedRequest
+      && ACTIVE_UPDATE_STATUSES.includes(queuedRequest.status)
+      && !reconciledIds.has(queuedRequest.id)
+      && !(queuedRequest.status === "staged" && queuedRequest.operation === "download"),
+  );
+  if (!stateFresh
+    && (!requestPresent || (requestMtime !== null && now - requestMtime >= ORPHANED_UPDATE_TIMEOUT_MS))
+    && !requestStillNeeded) {
     removeExpiredRequest(config.updateRequestPath, now);
   }
   return reconciled;

tests/update-api.test.ts

@@ -76,6 +76,12 @@ describe("更新 API", () => {
     expect(tooSoon.statusCode).toBe(429);
     expect(tooSoon.headers["retry-after"]).toBeDefined();
 
+    // Cooldown is scoped to the authenticated administrator, not the whole
+    // database or release endpoint.
+    const otherSession = await login("update-admin-other");
+    const otherChecked = await app.inject({ method: "POST", url: "/api/update/check", headers: { origin: config.publicOrigin, cookie: otherSession.cookies, "x-csrf-token": otherSession.csrf }, payload: {} });
+    expect(otherChecked.statusCode).toBe(200);
+
     const applied = await app.inject({ method: "POST", url: "/api/update/apply", headers: { origin: config.publicOrigin, cookie: session.cookies, "x-csrf-token": session.csrf }, payload: { version: "1.2.0", confirm: true } });
     expect(applied.statusCode).toBe(202);
     const jobId = applied.json().job.id as string;
@@ -157,6 +163,12 @@ describe("更新 API", () => {
     globalThis.fetch = (async () => new Response("upstream unavailable", { status: 503 })) as typeof fetch;
     const response = await app.inject({ method: "POST", url: "/api/update/apply", headers: { origin: config.publicOrigin, cookie: session.cookies, "x-csrf-token": session.csrf }, payload: { version: "1.2.0", confirm: true } });
     expect(response.statusCode).toBe(502);
+    // A failed upstream check must not reserve the per-admin cooldown; an
+    // operator can retry immediately after fixing the release endpoint.
+    const check = await app.inject({ method: "POST", url: "/api/update/check", headers: { origin: config.publicOrigin, cookie: session.cookies, "x-csrf-token": session.csrf }, payload: {} });
+    expect(check.statusCode).toBe(502);
+    const retry = await app.inject({ method: "POST", url: "/api/update/check", headers: { origin: config.publicOrigin, cookie: session.cookies, "x-csrf-token": session.csrf }, payload: {} });
+    expect(retry.statusCode).toBe(502);
     const audit = database.sqlite.prepare("SELECT outcome FROM audit_events WHERE action='update.apply_requested' ORDER BY id DESC LIMIT 1").get() as { outcome: string } | undefined;
     expect(audit?.outcome).toBe("failure");
   });

tests/update.test.ts

@@ -372,6 +372,47 @@ describe("更新安全工具", () => {
     }
   });
 
+  it("队列任务有新请求标记时可被重新检查，标记过期后才回收", async () => {
+    const root = await mkdtemp(path.join(tmpdir(), "tallynote-update-queued-marker-"));
+    let database: ReturnType<typeof openDatabase> | undefined;
+    try {
+      const dataDir = path.join(root, "data");
+      const installPrefix = path.join(root, "install");
+      process.env.TALLYNOTE_DATA_DIR = dataDir;
+      process.env.TALLYNOTE_INSTALL_PREFIX = installPrefix;
+      process.env.TALLYNOTE_PUBLIC_ORIGIN = "http://127.0.0.1:3998";
+      process.env.TALLYNOTE_COOKIE_SECURE = "false";
+      process.env.TALLYNOTE_UPDATE_STRATEGY = "systemd";
+      process.env.TALLYNOTE_UPDATE_METADATA_URL = "https://updates.example/latest";
+      process.env.TALLYNOTE_UPDATE_ALLOWED_HOSTS = "updates.example";
+      process.env.TALLYNOTE_UPDATE_REQUIRE_SIGNATURE = "false";
+      const config = loadConfig();
+      prepareDataDirectories(config);
+      database = openDatabase(config);
+      const staleAt = Date.now() - ORPHANED_UPDATE_TIMEOUT_MS - 1;
+      const jobId = randomUUID();
+      database.sqlite.prepare(`
+        INSERT INTO update_jobs(id, operation, status, version, platform, asset_url, created_at, updated_at)
+        VALUES (?, 'apply', 'queued', '1.2.0', 'linux-x64', ?, ?, ?)
+      `).run(jobId, "https://updates.example/queued.tar.gz", staleAt, staleAt);
+      await writeFile(config.updateRequestPath, JSON.stringify({ jobId, operation: "apply" }));
+      const now = Date.now();
+      await utimes(config.updateRequestPath, new Date(now), new Date(now));
+
+      expect(reconcileOrphanedUpdateJobs(database.sqlite, config, now)).toBe(0);
+      expect(database.sqlite.prepare("SELECT status FROM update_jobs WHERE id=?").get(jobId)).toEqual({ status: "queued" });
+      expect(await stat(config.updateRequestPath)).toBeTruthy();
+
+      const expiredNow = now + ORPHANED_UPDATE_TIMEOUT_MS + 1;
+      expect(reconcileOrphanedUpdateJobs(database.sqlite, config, expiredNow)).toBe(1);
+      expect(database.sqlite.prepare("SELECT status FROM update_jobs WHERE id=?").get(jobId)).toEqual({ status: "failed" });
+      await expect(stat(config.updateRequestPath)).rejects.toThrow();
+    } finally {
+      if (database) database.sqlite.close();
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
   it("流式解包在展开大小上限前拒绝高压缩比归档，并修正发布树权限", async () => {
     const root = await mkdtemp(path.join(tmpdir(), "tallynote-update-stream-"));
     try {

web-next/src/pages/update/UpdatePage.tsx

@@ -109,7 +109,7 @@ export default function UpdatePage({ timezone = "Asia/Shanghai", notify }: { tim
   const progress = job ? ({ queued: 8, downloading: 28, verifying: 48, staged: 65, backing_up: 80, applying: 92 } as Partial<Record<JobStatus, number>>)[job.status] ?? 100 : 0;
   const notes = latest ? notesFor(latest) : null;
 
-  return <Page title="系统更新" subtitle="检查受信任的 Release；更新前会校验文件并保护现有数据。" actions={<Button variant="outline" onClick={() => void check()} disabled={checking || loading || hasActiveJob} icon={<RefreshCw size={15} />}>{checking ? "检查中…" : "检查更新"}</Button>}>
+  return <Page title="系统更新" subtitle="检查受信任的 Release；更新前会校验文件并保护现有数据。" actions={<Button variant="outline" onClick={() => void check()} disabled={checking || loading} icon={<RefreshCw size={15} />}>{checking ? "检查中…" : "检查更新"}</Button>}>
     {error && <ErrorBanner message={error} onRetry={() => void load()} />}{pollError && <ErrorBanner message={pollError} />}
     {loading ? <div className="tn-empty" role="status" aria-live="polite">正在读取版本信息…</div> : info && <>
       <div className="tn-update-grid"><Surface className="tn-update-block"><Server size={20} /><span className="tn-eyebrow">当前版本</span><strong className="tn-update-value">v{info.currentVersion}</strong><small>运行平台：{info.platform.target}</small></Surface><Surface className="tn-update-block"><ShieldCheck size={20} /><span className="tn-eyebrow">更新方式</span><strong>{info.strategy === "systemd" ? "后台一键更新" : "手动命令行更新"}</strong><small>{info.strategy === "systemd" ? (info.configured ? "由 systemd 更新服务执行" : "尚未配置发布源") : "当前安装未启用后台更新"}</small></Surface></div>

web/src/main.tsx

@@ -771,7 +771,7 @@ function Update({ notify }: { notify: (message: string, kind?: Notice["kind"]) =
   const canApply = Boolean(info?.strategy === "systemd" && latest?.isNewer && latest.compatible && latest.integrityReady && !hasActiveJob && (!job || job.status === "failed" || job.version !== latest.version));
 
   return <div className="page update-page">
-    <div className="page-head"><div><div className="eyebrow">系统</div><h1>系统更新</h1></div><div className="head-actions"><Button onClick={() => void load()} disabled={loading || checking}><RotateCcw size={15} />刷新</Button><Button kind="primary" onClick={() => void check()} disabled={loading || checking || hasActiveJob}><RefreshCw size={15} className={checking ? "spin" : undefined} />检查更新</Button></div></div>
+    <div className="page-head"><div><div className="eyebrow">系统</div><h1>系统更新</h1></div><div className="head-actions"><Button onClick={() => void load()} disabled={loading || checking}><RotateCcw size={15} />刷新</Button><Button kind="primary" onClick={() => void check()} disabled={loading || checking}><RefreshCw size={15} className={checking ? "spin" : undefined} />检查更新</Button></div></div>
     {error && <div className="error banner" role="alert"><AlertCircle size={16} />{error}<button className="text-button" onClick={() => void load()}>重试</button></div>}
     {loading ? <div className="update-loading"><Loader2 className="spin" size={22} />正在读取版本信息</div> : info && <>
       <div className="update-overview">