@sleep2agi/commhub-server 0.9.0-preview.141 → 0.9.0-preview.142

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sleep2agi/commhub-server",
-  "version": "0.9.0-preview.141",
+  "version": "0.9.0-preview.142",
   "description": "CommHub Server — AI Agent communication hub with MCP protocol, multi-network isolation, user auth, and MCP tools (17 collaboration-core + node/provider ops tools; authoritative list: docs-site/docs/api/mcp-tools.md).",
   "type": "module",
   "main": "src/index.ts",

package/src/stop-delete-node.test.ts

@@ -961,6 +961,65 @@ describe("RFC-027 PR1.2a — restart_node resets lifecycle_state to 'active'", (
   });
 });
 
+// ── #1511 — restart_node 在途任务默认拒绝（只提示，不自动重启） ──────
+//
+// 节点侧 exit-75 的 drain 只等 60 秒，超时会硬打断当前回合；所以「正在跑
+// 任务」时 Hub 端在派发前就拒绝，计数规则与 stop_node 完全同一份
+// （#2022/#2029 修订版，实现在 countInFlightTasks）。客户端语义：保存密钥
+// 后若节点忙，不安排重启，只提示等任务完成或去节点机器上手动重启。
+describe("restart_node — in-flight default-refuse (#1511)", () => {
+  test("live task row → node_busy_in_flight, no dispatch row, state untouched", async () => {
+    setupAlphaNetwork();
+    seedTaskWithInbox(NET_A, CHILD_A_ALIAS, { task_id: "task_1511_live", status: "delivered" });
+    const r = await call(buildHandlers(USER_A_ID).restart_node, { node_id: CHILD_A_ID, network_id: NET_A });
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_busy_in_flight");
+    expect(r.in_flight_count).toBe(1);
+    expect(typeof r.hint).toBe("string");
+    const rows = db.all<{ update_id: string }>(`SELECT update_id FROM node_config_updates WHERE node_id = ?1`, [CHILD_A_ID]);
+    expect(rows.length).toBe(0);
+    expect(readNode(CHILD_A_ID)?.lifecycle_state).toBe("active");
+  });
+
+  test("unacked legacy inbox rows count too — same rule as stop", async () => {
+    setupAlphaNetwork();
+    seedInbox(NET_A, CHILD_A_ALIAS, 2);
+    const r = await call(buildHandlers(USER_A_ID).restart_node, { node_id: CHILD_A_ID, network_id: NET_A });
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_busy_in_flight");
+    expect(r.in_flight_count).toBe(2);
+  });
+
+  test("terminal-only rows → restart dispatches (parity with #2022 stop rule)", async () => {
+    setupAlphaNetwork();
+    seedTaskWithInbox(NET_A, CHILD_A_ALIAS, { task_id: "task_1511_replied", status: "replied" });
+    const r = await call(buildHandlers(USER_A_ID).restart_node, { node_id: CHILD_A_ID, network_id: NET_A });
+    expect(r.ok).toBe(true);
+    expect(r.apply_mode).toBe("restart_only");
+  });
+
+  test("2h-old live task is never aged out — status wins over age (same as stop)", async () => {
+    setupAlphaNetwork();
+    seedTaskWithInbox(NET_A, CHILD_A_ALIAS, {
+      task_id: "task_1511_running_120m", status: "running",
+      task_age: "-120 minutes", inbox_age: "-120 minutes", expires_in: "-60 minutes",
+    });
+    const r = await call(buildHandlers(USER_A_ID).restart_node, { node_id: CHILD_A_ID, network_id: NET_A });
+    expect(r.ok).toBe(false);
+    expect(r.error).toBe("node_busy_in_flight");
+  });
+
+  test("stopped node + live-row corpse → restart still dispatches (un-stop must not be blocked)", async () => {
+    setupAlphaNetwork();
+    db.run(`UPDATE nodes SET lifecycle_state = 'stopped' WHERE node_id = ?1`, [CHILD_A_ID]);
+    seedTaskWithInbox(NET_A, CHILD_A_ALIAS, { task_id: "task_1511_corpse", status: "delivered" });
+    const r = await call(buildHandlers(USER_A_ID).restart_node, { node_id: CHILD_A_ID, network_id: NET_A });
+    expect(r.ok).toBe(true);
+    expect(r.apply_mode).toBe("restart_only");
+    expect(readNode(CHILD_A_ID)?.lifecycle_state).toBe("active");
+  });
+});
+
 // ── #1286 stale-deleting 重派出口 ──────────────────────────────────
 //
 // 修复前 daemon 丢 ack 会让行永远停在 deleting：重发被挡、force 也一样。

package/src/tools.ts

@@ -4160,6 +4160,28 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
   );
 
 
+  // #1511 — stop_node（§4.3 D4）与 restart_node 共用的「在途任务」计数。
+  // 计数规则见 stop_node 内的长注释（#2022/#2029 修订：能成为「本节点正在做的
+  // 工作」的行才算——tasks 行按状态判、终态不算且不设龄上限；无状态行按 60 分钟
+  // 龄判）。抽成一处后，两个闸门共享同一份 SQL，不再各自漂移。
+  const IN_FLIGHT_MAX_AGE_MINUTES = 60;
+  function countInFlightTasks(alias: string, networkId: string | null): number {
+    const row = db.get<{ n: number }>(
+      `SELECT COUNT(*) AS n FROM inbox i
+         LEFT JOIN tasks t
+           ON i.type = 'task'
+          AND t.task_id = COALESCE(i.task_id, i.id)
+        WHERE i.session_name = ?1 AND i.acked = 0
+          AND COALESCE(i.network_id, ?2) = ?2
+          AND ((t.task_id IS NOT NULL
+                AND t.status NOT IN ('replied', 'failed', 'cancelled', 'expired'))
+            OR (t.task_id IS NULL
+                AND i.created_at >= datetime('now', ?3)))`,
+      alias, networkId, `-${IN_FLIGHT_MAX_AGE_MINUTES} minutes`,
+    );
+    return row?.n ?? 0;
+  }
+
   server.tool(
     "restart_node",
     "Trigger a node restart without changing config. RFC-024 Vincent 2026-06-28 increment. Network-scoped (SEC-1); member+ role suffices (lifecycle ops are not privilege elevation).",
@@ -4189,6 +4211,27 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
         return { content: [{ type: "text" as const, text: JSON.stringify({ ok: false,
           error: "adopted_restart_requires_daemon", message: "Use Stop, then Start for this adopted node." }) }] };
       }
+      // #1511 — 正在跑任务时不自动重启，只提示。节点侧 exit-75 的 drain 只等
+      // 60 秒、超时会硬打断当前回合（agent-node cli.ts drainInFlightThink），
+      // 所以 Hub 端在派发前就按 stop_node 同一条在途计数规则拒绝，客户端拿到
+      // 明确二选一：等任务完成后重试，或到节点所在机器手动重启。
+      //
+      // 只对 active 节点生效。stopped 节点的进程已死、任务不可能正在跑，其
+      // unacked 行是 stop 留下的尸体；restart_node 在 stopped 节点上是把
+      // lifecycle 翻回 active 的出口，不能用尸体行把这条路堵死。
+      // lifecycle_state 需单独取：resolveTargetNode 的 SELECT 不含该列。
+      const restartLifecycle = db.get<{ lifecycle_state: string | null }>(
+        "SELECT lifecycle_state FROM nodes WHERE node_id = ?1", nodeId,
+      )?.lifecycle_state ?? null;
+      if (restartLifecycle === "active") {
+        const busyCount = countInFlightTasks(node.alias, node.network_id);
+        if (busyCount > 0) {
+          return { content: [{ type: "text" as const, text: JSON.stringify({
+            ok: false, error: "node_busy_in_flight", in_flight_count: busyCount,
+            hint: "节点正在执行任务，本次未安排重启；等它完成后再试，或到节点所在机器手动重启",
+          }) }] };
+        }
+      }
       // F-B reaper: same stale-supersede semantics as update_node_config,
       // with same acked_at-anchored liveness clock (see update_node_config
       // for the 通信龙 polish reasoning).
@@ -5449,21 +5492,8 @@ export function registerTools(server: McpServer, clientIP?: string, enforceNetwo
     //      ever retire them, so they get the age ceiling
     //      IN_FLIGHT_MAX_AGE_MINUTES on inbox.created_at.
     // The count stays a per-alias inbox count: one PK join per row.
-    const IN_FLIGHT_MAX_AGE_MINUTES = 60;
-    const inFlightRow = db.get<{ n: number }>(
-      `SELECT COUNT(*) AS n FROM inbox i
-         LEFT JOIN tasks t
-           ON i.type = 'task'
-          AND t.task_id = COALESCE(i.task_id, i.id)
-        WHERE i.session_name = ?1 AND i.acked = 0
-          AND COALESCE(i.network_id, ?2) = ?2
-          AND ((t.task_id IS NOT NULL
-                AND t.status NOT IN ('replied', 'failed', 'cancelled', 'expired'))
-            OR (t.task_id IS NULL
-                AND i.created_at >= datetime('now', ?3)))`,
-      node.alias, node.network_id, `-${IN_FLIGHT_MAX_AGE_MINUTES} minutes`,
-    );
-    const inFlight = inFlightRow?.n ?? 0;
+    // 计数实现在上方 countInFlightTasks。#1511 让 restart_node 共用同一份规则，判据本身未变。
+    const inFlight = countInFlightTasks(node.alias, node.network_id);
     if (inFlight > 0 && !args.force) {
       return { ok: false, error: "node_busy_in_flight", in_flight_count: inFlight, hint: "set force=true to override (audit logged)" };
     }