@nowcrew/daemon 0.6.91-beta.3 → 0.6.91-beta.5

package/dist/daemon-update-controller.js

@@ -1,3 +1,4 @@
+import { dslog } from "./slog.js";
 import { z } from "zod";
 import { isDaemonReleaseVersion } from "./daemon-release-version.js";
 const DaemonUpdateMessageSchema = z.object({
@@ -42,15 +43,16 @@ export function createDaemonUpdateController(deps) {
             failed(message.updateId, preparation.errorCode);
             return;
         }
-        let released = false;
-        const releaseOnce = async () => {
-            if (released)
-                return;
-            released = true;
-            await preparation.release();
+        let releaseTask;
+        const releaseOnce = () => {
+            releaseTask ??= preparation.release().catch(error => {
+                releaseTask = undefined;
+                throw error;
+            });
+            return releaseTask;
         };
-        deps.sendStatus({ type: "daemon:update-status", updateId: message.updateId, status: "restarting" });
         try {
+            deps.sendStatus({ type: "daemon:update-status", updateId: message.updateId, status: "restarting" });
             const handoff = deps.scheduleRestart(eligibility.serviceSpec);
             const state = {
                 release: releaseOnce,
@@ -92,7 +94,13 @@ export function createDaemonUpdateController(deps) {
                     failed(message.updateId, "update_in_progress");
                 return true;
             }
-            const done = execute(message).finally(() => {
+            const done = execute(message).catch(error => {
+                dslog("daemon.update_failed", "Daemon 更新或重启准备失败", {
+                    level: "ERROR", update_id: message.updateId, operation: message.type,
+                    error_message: error instanceof Error ? error.message : String(error),
+                });
+                failed(message.updateId, message.type === "daemon:restart" ? "restart_failed" : "install_failed");
+            }).finally(() => {
                 handled.add(message.updateId);
                 if (running?.id === message.updateId)
                     running = null;

package/dist/daemon-updater.js

@@ -1,3 +1,4 @@
+import { dslog } from "./slog.js";
 import { readFile } from "node:fs/promises";
 import { resolve } from "node:path";
 import { systemCommandRunner } from "./computer-service.js";
@@ -17,6 +18,8 @@ export async function prepareDaemonRestart(input) {
     const local = input.localSlots.tryAcquireExclusive();
     if (local === null)
         return { ok: false, errorCode: "runtime_busy" };
+    const acquiredAt = Date.now();
+    dslog("daemon.maintenance_barrier_acquired", "Daemon 已暂停接收执行，准备维护");
     let released = false;
     return {
         ok: true,
@@ -25,6 +28,7 @@ export async function prepareDaemonRestart(input) {
                 return;
             released = true;
             local.release();
+            dslog("daemon.maintenance_barrier_released", "维护屏障已释放", { held_ms: Date.now() - acquiredAt });
         },
     };
 }
@@ -67,16 +71,10 @@ export async function installExactDaemonUpdate(input) {
         || installation.npmPrefix !== input.npmPrefix) {
         return { ok: false, errorCode: "ineligible" };
     }
-    const local = input.localSlots.tryAcquireExclusive();
-    if (local === null)
-        return { ok: false, errorCode: "runtime_busy" };
-    let released = false;
-    const release = async () => {
-        if (released)
-            return;
-        released = true;
-        local.release();
-    };
+    const barrier = await prepareDaemonRestart(input);
+    if (!barrier.ok)
+        return barrier;
+    const release = barrier.release;
     const lockedInstallation = daemonGlobalInstallation(input.currentEntryPath ?? process.argv[1] ?? "", input.platform ?? process.platform);
     if (lockedInstallation === null
         || lockedInstallation.packageRoot !== input.packageRoot

package/dist/host-execution-coordinator.js

@@ -286,13 +286,19 @@ export function createHostExecutionCoordinator(options = {}) {
                 ...Array.from({ length: executionSlots }, (_, slot) => join(root, "slots", String(slot))),
                 join(root, "memory-prune-slots", "0"),
             ];
-            for (const target of targets) {
-                const lease = await acquireLease(target);
-                if (lease === null) {
-                    await retryRelease("exclusive_execution_partial", () => Promise.all(leases.map((owned) => owned.close())).then(() => undefined));
-                    return null;
+            try {
+                for (const target of targets) {
+                    const lease = await acquireLease(target);
+                    if (lease === null) {
+                        await retryRelease("exclusive_execution_partial", () => Promise.all(leases.map((owned) => owned.close())).then(() => undefined));
+                        return null;
+                    }
+                    leases.push(lease);
                 }
-                leases.push(lease);
+            }
+            catch (error) {
+                await retryRelease("exclusive_execution_partial", () => Promise.all(leases.map(owned => owned.close())).then(() => undefined));
+                throw error;
             }
             let releasePromise = null;
             return {

package/dist/pre-admission-rejection.js

@@ -2,7 +2,7 @@ import { ExecutionCompletedSchema, ExecutionRejectedSchema } from "./execution-p
 import { hashExecutionSpec } from "./execution-runner.js";
 import { dslog } from "./slog.js";
 /** Queue rejection is admission work too: own the ID before any asynchronous journal write. */
-export async function rejectQueuedAdmission(context, spec, executionClass, facts) {
+export async function rejectQueuedAdmission(context, spec, executionClass, facts, rejectionReason = "machine_queue_full") {
     const id = spec.executionId;
     const hash = hashExecutionSpec(spec);
     if (context.knownExecutionHashes.has(id) || context.executionRuns.has(id)) {
@@ -16,10 +16,11 @@ export async function rejectQueuedAdmission(context, spec, executionClass, facts
     context.knownExecutionHashes.set(id, hash);
     const rejected = ExecutionRejectedSchema.parse({
         type: "execution:rejected", protocolVersion: 1, executionId: id,
-        reason: "resource_limit", message: "Local machine execution queue is full", at: new Date().toISOString(),
+        reason: "resource_limit", message: rejectionReason === "maintenance" ? "Local machine is temporarily unavailable for maintenance"
+            : rejectionReason === "agent_queue_full" ? "Local agent execution queue is full" : "Local machine execution queue is full", at: new Date().toISOString(),
     });
-    dslog("execution.machine_queue_rejected", "机器执行队列已满", {
-        level: "WARN", execution_id: id, agent_handle: spec.agent.handle, execution_class: executionClass, ...facts,
+    dslog("execution.machine_queue_rejected", rejectionReason === "maintenance" ? "机器维护期间暂停接收执行" : "执行等待队列已满", {
+        level: "WARN", execution_id: id, agent_handle: spec.agent.handle, execution_class: executionClass, admission_reason: rejectionReason, ...facts,
     });
     const rejecting = persistPreAdmissionRejection(context.journal, spec, rejected)
         .then(async (entry) => {

package/dist/serve.js

@@ -488,7 +488,7 @@ export function serve(config, opts = {}) {
                     let reservation = sharedSlots.reserve(spec.agent.handle, "execution", executionClass);
                     if (!reservation.accepted) {
                         await rejectQueuedAdmission({ journal: executionJournal, knownExecutionHashes, executionRuns,
-                            report: reportExecutionFrame, replay: sendJournalStatus }, spec, executionClass, reservation.facts);
+                            report: reportExecutionFrame, replay: sendJournalStatus }, spec, executionClass, reservation.facts, reservation.rejectionReason);
                         return;
                     }
                     knownExecutionHashes.set(spec.executionId, hash);

package/dist/shared-execution-slots.js

@@ -59,6 +59,7 @@ export function createSharedSlotManager(limits) {
             if (exclusive) {
                 return {
                     accepted: false,
+                    rejectionReason: "maintenance",
                     facts,
                     ready: Promise.resolve(),
                     isQueued: () => false,
@@ -77,6 +78,7 @@ export function createSharedSlotManager(limits) {
                 || laneQueuedTotal >= limits.maxQueuedTotal)) {
                 return {
                     accepted: false,
+                    rejectionReason: queuedForAgent >= limits.maxQueuedPerAgent ? "agent_queue_full" : "machine_queue_full",
                     facts,
                     ready: Promise.resolve(),
                     isQueued: () => false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nowcrew/daemon",
-  "version": "0.6.91-beta.3",
+  "version": "0.6.91-beta.5",
   "type": "module",
   "description": "crew daemon — 运行在用户机器:拉起/管理 agent 进程,注入 crew CLI,归一化 runtime 事件",
   "license": "Apache-2.0",