@nowcrew/daemon 0.6.102-beta.5 → 0.6.102-beta.6

package/dist/execution-telemetry-journal.js

@@ -1,9 +1,8 @@
-import { access, open, mkdir, readdir, readFile, rename, unlink, stat } from "node:fs/promises";
+import { access, open, mkdir, readdir, readFile, rename, unlink } from "node:fs/promises";
 import { join } from "node:path";
 import { randomUUID } from "node:crypto";
 import { DaemonToServerExecutionFrameSchema, } from "./execution-protocol.js";
 const MAX_PENDING_FILES = 4_096;
-const MAX_TELEMETRY_FRAME_BYTES = 256 * 1_024;
 const fileName = (frame) => `${frame.executionId}.${frame.type === "execution:activity" ? "activity" : "console"}.${frame.seq}.json`;
 export class ExecutionTelemetryJournal {
     directory;
@@ -11,10 +10,6 @@ export class ExecutionTelemetryJournal {
         this.directory = join(agentsRoot, ".execution-telemetry");
     }
     async append(frame) {
-        const serialized = JSON.stringify(frame);
-        if (Buffer.byteLength(serialized) > MAX_TELEMETRY_FRAME_BYTES) {
-            throw new Error("execution telemetry frame exceeds 256 KiB");
-        }
         await mkdir(this.directory, { recursive: true, mode: 0o700 });
         const path = join(this.directory, fileName(frame));
         if (await access(path).then(() => true, () => false))
@@ -27,7 +22,7 @@ export class ExecutionTelemetryJournal {
         let handle;
         try {
             handle = await open(tempPath, "wx", 0o600);
-            await handle.writeFile(serialized, "utf8");
+            await handle.writeFile(JSON.stringify(frame), "utf8");
             await handle.sync();
             await handle.close();
             handle = undefined;
@@ -48,36 +43,29 @@ export class ExecutionTelemetryJournal {
                 throw error;
         });
     }
-    async *replay(onInvalid) {
+    async replay() {
         const names = await readdir(this.directory).catch((error) => {
             if (error.code === "ENOENT")
                 return [];
             throw error;
         });
-        for (const name of names.filter((entry) => entry.endsWith(".json")).sort((a, b) => a.localeCompare(b, undefined, { numeric: true }))) {
+        const frames = [];
+        for (const name of names.filter((entry) => entry.endsWith(".json")).sort()) {
             let raw;
             try {
-                const path = join(this.directory, name);
-                if ((await stat(path)).size > MAX_TELEMETRY_FRAME_BYTES) {
-                    throw new Error("execution telemetry frame exceeds 256 KiB");
-                }
-                raw = JSON.parse(await readFile(path, "utf8"));
-                const parsed = DaemonToServerExecutionFrameSchema.safeParse(raw);
-                if (!parsed.success || (parsed.data.type !== "execution:activity" && parsed.data.type !== "execution:console")) {
-                    throw new Error("invalid execution telemetry frame");
-                }
-                yield parsed.data;
+                raw = JSON.parse(await readFile(join(this.directory, name), "utf8"));
             }
             catch (error) {
-                // ACK 可与目录扫描并发删除条目。
-                if (error.code === "ENOENT")
-                    continue;
-                const invalid = new Error(`invalid execution telemetry journal entry: ${name}`, { cause: error });
-                if (!onInvalid)
-                    throw invalid;
-                onInvalid(invalid);
+                throw new Error(`invalid execution telemetry journal entry: ${name}`, { cause: error });
+            }
+            const parsed = DaemonToServerExecutionFrameSchema.safeParse(raw);
+            if (!parsed.success || (parsed.data.type !== "execution:activity" && parsed.data.type !== "execution:console")) {
+                throw new Error(`invalid execution telemetry journal entry: ${name}`);
             }
+            frames.push(parsed.data);
         }
+        return frames.sort((left, right) => left.at.localeCompare(right.at)
+            || left.type.localeCompare(right.type) || left.seq - right.seq);
     }
 }
 export const createExecutionTelemetryJournal = (agentsRoot) => new ExecutionTelemetryJournal(agentsRoot);

package/dist/serve.js

@@ -1,4 +1,3 @@
-import { ExecutionTelemetrySender } from "./execution-telemetry-sender.js";
 /** 常驻控制面桥接；按请求拉起本地 runtime。 */
 import { handleAgentEnvironmentFrame } from "./agent-environment-control.js";
 import { handleMachineEnvironmentFrame } from "./machine-environment-control.js";
@@ -81,11 +80,6 @@ export function serve(config, opts = {}) {
     const agentMemory = opts.execution?.dependencies?.agentMemory
         ?? (config.agentMemory === null ? undefined : createAgentMemoryBridge(config.agentMemory));
     const executionTelemetry = createExecutionTelemetryJournal(config.agentsRoot);
-    const telemetrySender = new ExecutionTelemetrySender(executionTelemetry, (error) => {
-        dslog("execution.telemetry_replay_failed", "execution telemetry 重放失败", {
-            level: "ERROR", error_message: error.message,
-        });
-    });
     const executeProtocol = opts.execution?.runExecution ?? runExecution;
     const localHome = opts.localHome ?? resolveDaemonLocalStateRootFromEnv();
     const hostCoordinator = opts.execution?.hostCoordinator
@@ -240,8 +234,6 @@ export function serve(config, opts = {}) {
         if (frame.type === "execution:activity" || frame.type === "execution:console") {
             try {
                 await executionTelemetry.append(frame);
-                telemetrySender.notify();
-                return;
             }
             catch (error) {
                 dslog("execution.telemetry_journal_failed", "execution telemetry 持久化失败", {
@@ -262,6 +254,11 @@ export function serve(config, opts = {}) {
                 completionRetransmitter.track(entry.completion);
         }
     };
+    const replayExecutionTelemetry = async () => {
+        const frames = await executionTelemetry.replay();
+        for (const frame of frames)
+            safeExecutionSend(frame);
+    };
     const snapshotEntry = (entry) => {
         if ((entry.state === "completed" || entry.state === "interrupted") && entry.completion !== null) {
             return { executionId: entry.executionId, state: entry.state, completion: entry.completion, updatedAt: entry.updatedAt };
@@ -377,7 +374,11 @@ export function serve(config, opts = {}) {
                     completionRetransmitter.resume();
                 }
             });
-            telemetrySender.start((frame) => ws === openedSocket && safeExecutionSend(frame), () => openedSocket.close(1013, "telemetry ACK timeout"));
+            void replayExecutionTelemetry().catch((error) => {
+                dslog("execution.telemetry_replay_failed", "execution telemetry 重放失败", {
+                    level: "ERROR", error_message: error.message,
+                });
+            });
         });
         connectionSocket.on("message", async (data) => {
             if (stopped || ws !== connectionSocket)
@@ -449,7 +450,7 @@ export function serve(config, opts = {}) {
                         return;
                     }
                     if (frame.type === "execution:event-ack") {
-                        await telemetrySender.acknowledge(frame.executionId, frame.kind, frame.seq).catch((error) => {
+                        await executionTelemetry.acknowledge(frame.executionId, frame.kind, frame.seq).catch((error) => {
                             dslog("execution.telemetry_ack_failed", "execution telemetry ACK 处理失败", {
                                 level: "ERROR", execution_id: frame.executionId,
                                 error_message: error.message,
@@ -1262,7 +1263,6 @@ export function serve(config, opts = {}) {
             // server 能力随连接失效;下个连接的 ready 帧重新声明(防降级重连后沿用过期能力)。
             serverCapabilities = new Set();
             completionRetransmitter.pause();
-            telemetrySender.stop();
             // 4001 = 控制面应用级「鉴权失败」关闭码(见 server control-plane.ts)。即使上面的
             // error 帧因 close 抢先而丢失,也能据关闭码识别这是凭证失效——退避拉满,不再每秒热循环。
             if (reportDaemonConnectionClose(code, log))
@@ -1323,7 +1323,6 @@ export function serve(config, opts = {}) {
             stopped = true;
             concurrencyTelemetry.stop();
             completionRetransmitter.stop();
-            telemetrySender.stop();
             const deadline = createShutdownDeadline(shutdownTimeoutMs);
             if (reconnectTimer !== null) {
                 clearTimeout(reconnectTimer);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nowcrew/daemon",
-  "version": "0.6.102-beta.5",
+  "version": "0.6.102-beta.6",
   "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",
@@ -18,9 +18,17 @@
   "publishConfig": {
     "access": "public"
   },
+  "scripts": {
+    "daemon": "pnpm --filter @nowcrew/cli build && tsx src/main.ts",
+    "build": "tsc -p tsconfig.json",
+    "prepublishOnly": "pnpm build && node ../scripts/daemon-release-artifact.mjs --strict-registry",
+    "test": "vitest run",
+    "test:release-critical": "vitest run test/automatic-daemon-installer.test.ts test/automatic-install-target.test.ts test/automatic-service-migration.test.ts test/computer-cli.test.ts test/computer-service.test.ts test/computer-setup-client.test.ts test/computer-setup-orchestration.test.ts test/computer-setup.test.ts test/daemon-release-version.test.ts test/daemon-update-controller.test.ts test/daemon-update-eligibility.test.ts test/daemon-updater.test.ts test/execution-protocol.test.ts test/machine-info.test.ts --maxWorkers=2 --minWorkers=1",
+    "typecheck": "tsc --noEmit"
+  },
   "dependencies": {
     "@agentclientprotocol/sdk": "1.2.1",
-    "@nowcrew/cli": "0.4.19-beta.1",
+    "@nowcrew/cli": "0.4.17",
     "cross-spawn": "^7.0.6",
     "semver": "7.8.5",
     "smol-toml": "1.8.0",
@@ -39,12 +47,5 @@
     "tsx": "^4.19.0",
     "typescript": "^5.6.0",
     "vitest": "^2.1.0"
-  },
-  "scripts": {
-    "daemon": "pnpm --filter @nowcrew/cli build && tsx src/main.ts",
-    "build": "tsc -p tsconfig.json",
-    "test": "vitest run",
-    "test:release-critical": "vitest run test/automatic-daemon-installer.test.ts test/automatic-install-target.test.ts test/automatic-service-migration.test.ts test/computer-cli.test.ts test/computer-service.test.ts test/computer-setup-client.test.ts test/computer-setup-orchestration.test.ts test/computer-setup.test.ts test/daemon-release-version.test.ts test/daemon-update-controller.test.ts test/daemon-update-eligibility.test.ts test/daemon-updater.test.ts test/execution-protocol.test.ts test/machine-info.test.ts --maxWorkers=2 --minWorkers=1",
-    "typecheck": "tsc --noEmit"
   }
-}
+}

package/dist/execution-telemetry-sender.js

@@ -1,151 +0,0 @@
-import { dslog } from "./slog.js";
-import {} from "./execution-telemetry-journal.js";
-const MAX_IN_FLIGHT = 16;
-const MAX_IN_FLIGHT_BYTES = 512 * 1_024;
-const ACK_TIMEOUT_MS = 30_000;
-const keyOf = (executionId, kind, seq) => `${executionId}.${kind}.${seq}`;
-/** 复用 journal 与 event-ack。窗口只保留游标，帧内容按需从磁盘读取。 */
-export class ExecutionTelemetrySender {
-    journal;
-    onError;
-    inFlight = new Map();
-    generation = 0;
-    running = false;
-    requested = false;
-    windowBlocked = false;
-    lastProgress = 0;
-    lastReportAt = 0;
-    lastErrorAt = -Infinity;
-    sentFrames = 0;
-    acknowledgedFrames = 0;
-    send;
-    timer;
-    constructor(journal, onError) {
-        this.journal = journal;
-        this.onError = onError;
-    }
-    start(send, onTimeout) {
-        this.stop();
-        this.send = send;
-        this.sentFrames = 0;
-        this.acknowledgedFrames = 0;
-        this.lastReportAt = Date.now();
-        this.logState("execution.telemetry_connected", "遥测补传连接已建立");
-        this.timer = setInterval(() => {
-            // 单个失败帧不能打断仍在正常确认的连接；只有整个窗口停滞才重连。
-            if (this.windowBlocked && Date.now() - this.lastProgress >= ACK_TIMEOUT_MS) {
-                this.logState("execution.telemetry_stalled", "遥测确认窗口停滞，准备重连", {
-                    level: "WARN", stalled_ms: Date.now() - this.lastProgress,
-                });
-                this.stop();
-                onTimeout();
-            }
-            else if (this.inFlight.size > 0 && Date.now() - this.lastReportAt >= ACK_TIMEOUT_MS) {
-                this.lastReportAt = Date.now();
-                this.logState("execution.telemetry_progress", "遥测确认窗口状态");
-            }
-        }, 1_000);
-        this.timer.unref();
-        this.notify();
-    }
-    stop() {
-        if (this.send)
-            this.logState("execution.telemetry_disconnected", "遥测发送停止，未确认帧保留在 journal");
-        this.generation += 1;
-        this.send = undefined;
-        this.requested = false;
-        this.windowBlocked = false;
-        if (this.timer !== undefined)
-            clearInterval(this.timer);
-        this.timer = undefined;
-        this.inFlight.clear();
-    }
-    async acknowledge(executionId, kind, seq) {
-        const key = keyOf(executionId, kind, seq);
-        const entry = this.inFlight.get(key);
-        if (!entry)
-            return;
-        await this.journal.acknowledge(executionId, kind, seq);
-        // 重连会清空窗口；条目身份也能挡住旧连接和重复 ACK 的异步完成。
-        if (this.inFlight.get(key) !== entry)
-            return;
-        this.inFlight.delete(key);
-        this.acknowledgedFrames += 1;
-        this.lastProgress = Date.now();
-        this.notify();
-    }
-    notify() {
-        if (!this.send)
-            return;
-        this.requested = true;
-        if (!this.running)
-            void this.pump();
-    }
-    get bytes() {
-        let bytes = 0;
-        for (const entry of this.inFlight.values())
-            bytes += entry.bytes;
-        return bytes;
-    }
-    logState(event, message, fields = {}) {
-        dslog(event, message, {
-            generation: this.generation, in_flight_frames: this.inFlight.size,
-            in_flight_bytes: this.bytes, sent_frames: this.sentFrames,
-            acknowledged_frames: this.acknowledgedFrames, window_blocked: this.windowBlocked, ...fields,
-        });
-    }
-    reportError(error) {
-        // 同一个坏文件会被多次扫描；不要让错误日志形成新的无界积压。
-        if (Date.now() - this.lastErrorAt < ACK_TIMEOUT_MS)
-            return;
-        this.lastErrorAt = Date.now();
-        this.onError(error);
-    }
-    async pump() {
-        this.running = true;
-        const generation = this.generation;
-        try {
-            while (this.send && generation === this.generation && this.requested) {
-                this.requested = false;
-                let blocked = false;
-                // 本轮读取中的文件可能刚收到 ACK；不能把已删除游标的旧读取结果再发出去。
-                const alreadySent = new Set(this.inFlight.keys());
-                for await (const frame of this.journal.replay((error) => this.reportError(error))) {
-                    if (!this.send || generation !== this.generation)
-                        return;
-                    const key = keyOf(frame.executionId, frame.type === "execution:activity" ? "activity" : "console", frame.seq);
-                    if (alreadySent.has(key) || this.inFlight.has(key))
-                        continue;
-                    const bytes = Buffer.byteLength(JSON.stringify(frame));
-                    if (this.inFlight.size >= MAX_IN_FLIGHT || this.bytes + bytes > MAX_IN_FLIGHT_BYTES) {
-                        blocked = true;
-                        break;
-                    }
-                    if (this.inFlight.size === 0)
-                        this.lastProgress = Date.now();
-                    this.inFlight.set(key, { bytes });
-                    if (!this.send(frame)) {
-                        this.inFlight.delete(key);
-                        this.logState("execution.telemetry_send_failed", "遥测发送失败，帧保留在 journal", { level: "WARN" });
-                        return;
-                    }
-                    this.sentFrames += 1;
-                }
-                if (!this.send || generation !== this.generation)
-                    return;
-                blocked ||= this.inFlight.size >= MAX_IN_FLIGHT || this.bytes >= MAX_IN_FLIGHT_BYTES;
-                if (blocked && !this.windowBlocked)
-                    this.lastProgress = Date.now();
-                this.windowBlocked = blocked;
-            }
-        }
-        catch (error) {
-            this.reportError(error);
-        }
-        finally {
-            this.running = false;
-            if (this.send && this.requested)
-                this.notify();
-        }
-    }
-}