helloloop 0.8.6 → 0.10.0

package/src/supervisor_guardian.mjs

@@ -0,0 +1,307 @@
+import path from "node:path";
+
+import { nowIso, readJson, sleep, writeJson } from "./common.mjs";
+import { createContext } from "./context.mjs";
+import { refreshHostContinuationArtifacts } from "./host_continuation.mjs";
+import { isHostLeaseAlive } from "./host_lease.mjs";
+import { spawnNodeProcess } from "./node_process_launch.mjs";
+import {
+  FINAL_SUPERVISOR_STATUSES,
+  readJsonIfExists,
+  readSupervisorState,
+  writeActiveSupervisorState,
+  writeSupervisorState,
+} from "./supervisor_state.mjs";
+import { bindBackgroundTerminalSession } from "./terminal_session_limits.mjs";
+import { loadRuntimeSettings } from "./runtime_settings_loader.mjs";
+import { hasRetryBudget, pickRetryDelaySeconds } from "./runtime_settings.mjs";
+import { registerActiveSession, unregisterActiveSession } from "./workspace_registry.mjs";
+
+const GUARDIAN_ACTIVE_ENV = "HELLOLOOP_SUPERVISOR_GUARDIAN_ACTIVE";
+const GUARDIAN_PID_ENV = "HELLOLOOP_SUPERVISOR_GUARDIAN_PID";
+
+function refreshContinuation(context, sessionId) {
+  try {
+    refreshHostContinuationArtifacts(context, { sessionId });
+  } catch {
+    // ignore continuation snapshot refresh failures during guardian lifecycle
+  }
+}
+
+function readFinalOutcome(context, sessionId) {
+  const result = readJsonIfExists(context.supervisorResultFile);
+  if (result?.sessionId === sessionId) {
+    return {
+      sessionId,
+      ok: result.ok === true,
+      exitCode: Number(result.exitCode ?? (result.ok === true ? 0 : 1)),
+      result,
+    };
+  }
+
+  const state = readSupervisorState(context);
+  if (state?.sessionId === sessionId && FINAL_SUPERVISOR_STATUSES.has(String(state.status || ""))) {
+    return {
+      sessionId,
+      ok: state.status === "completed",
+      exitCode: Number(state.exitCode || (state.status === "completed" ? 0 : 1)),
+      result: null,
+    };
+  }
+
+  return null;
+}
+
+function writeLeaseStopped(context, request, message) {
+  const payload = {
+    sessionId: request.sessionId,
+    command: request.command,
+    exitCode: 1,
+    ok: false,
+    stopped: true,
+    error: message,
+  };
+  writeJson(context.supervisorResultFile, payload);
+  writeSupervisorState(context, {
+    sessionId: request.sessionId,
+    command: request.command,
+    status: "stopped",
+    exitCode: payload.exitCode,
+    message,
+    completedAt: nowIso(),
+  });
+}
+
+function writeKeepAliveFailure(context, request, message, restartCount, keepAliveEnabled) {
+  const payload = {
+    sessionId: request.sessionId,
+    command: request.command,
+    ok: false,
+    exitCode: 1,
+    error: message,
+  };
+  writeJson(context.supervisorResultFile, payload);
+  writeSupervisorState(context, {
+    sessionId: request.sessionId,
+    command: request.command,
+    status: "failed",
+    exitCode: 1,
+    message,
+    guardianPid: process.pid,
+    workerPid: 0,
+    guardianRestartCount: restartCount,
+    keepAliveEnabled,
+    completedAt: nowIso(),
+  });
+}
+
+function buildWorkerArgs(context, sessionFile) {
+  return [
+    path.join(context.bundleRoot, "bin", "helloloop.js"),
+    "__supervise-worker",
+    "--session-file",
+    sessionFile,
+  ];
+}
+
+function waitForChildExit(child) {
+  return new Promise((resolve, reject) => {
+    let settled = false;
+    child.once("error", (error) => {
+      if (!settled) {
+        settled = true;
+        reject(error);
+      }
+    });
+    child.once("exit", (exitCode, signal) => {
+      if (!settled) {
+        settled = true;
+        resolve({
+          exitCode: Number(exitCode ?? 0),
+          signal: signal ? String(signal) : "",
+        });
+      }
+    });
+  });
+}
+
+function writeGuardianRunningState(context, request, keepAlive, restartCount, workerPid, message) {
+  writeActiveSupervisorState(context, {
+    sessionId: request.sessionId,
+    command: request.command,
+    status: "running",
+    lease: request.lease || {},
+    pid: process.pid,
+    guardianPid: process.pid,
+    workerPid,
+    guardianRestartCount: restartCount,
+    keepAliveEnabled: keepAlive.enabled,
+    message,
+  });
+}
+
+async function runWorkerUnderGuardian(context, request, sessionFile, keepAlive, restartCount) {
+  writeGuardianRunningState(
+    context,
+    request,
+    keepAlive,
+    restartCount,
+    0,
+    restartCount > 0
+      ? `后台守护进程正在执行第 ${restartCount} 次自动重拉起。`
+      : "后台守护进程已接管，正在启动执行 worker。",
+  );
+
+  const child = spawnNodeProcess({
+    args: buildWorkerArgs(context, sessionFile),
+    cwd: context.repoRoot,
+    detached: false,
+    stdio: "inherit",
+    env: {
+      HELLOLOOP_SUPERVISOR_ACTIVE: "1",
+      [GUARDIAN_ACTIVE_ENV]: "1",
+      [GUARDIAN_PID_ENV]: String(process.pid),
+    },
+  });
+
+  writeGuardianRunningState(
+    context,
+    request,
+    keepAlive,
+    restartCount,
+    child.pid ?? 0,
+    restartCount > 0
+      ? `后台守护进程已完成第 ${restartCount} 次自动重拉起，worker 正在运行。`
+      : "后台守护进程正在运行。",
+  );
+
+  return waitForChildExit(child);
+}
+
+export async function runSupervisorGuardianFromSessionFile(sessionFile) {
+  const request = readJson(sessionFile);
+  const context = createContext(request.context || {});
+  const keepAlive = loadRuntimeSettings({
+    globalConfigFile: request.options?.globalConfigFile,
+  }).supervisorKeepAlive;
+
+  bindBackgroundTerminalSession(request.terminalSessionFile || "", {
+    command: request.command,
+    repoRoot: context.repoRoot,
+    sessionId: request.sessionId,
+  });
+  registerActiveSession({
+    sessionId: request.sessionId,
+    repoRoot: context.repoRoot,
+    configDirName: context.configDirName,
+    command: request.command,
+    pid: process.pid,
+    lease: request.lease || {},
+    startedAt: nowIso(),
+    supervisorStateFile: context.supervisorStateFile,
+    supervisorResultFile: context.supervisorResultFile,
+    statusFile: context.statusFile,
+  });
+
+  try {
+    let restartCount = 0;
+    refreshContinuation(context, request.sessionId);
+
+    while (true) {
+      const finalOutcome = readFinalOutcome(context, request.sessionId);
+      if (finalOutcome) {
+        unregisterActiveSession(request.sessionId);
+        return;
+      }
+
+      if (!isHostLeaseAlive(request.lease || {})) {
+        writeLeaseStopped(context, request, "检测到宿主窗口已关闭，HelloLoop 守护进程未继续执行。");
+        refreshContinuation(context, request.sessionId);
+        unregisterActiveSession(request.sessionId);
+        return;
+      }
+
+      let workerExit;
+      try {
+        workerExit = await runWorkerUnderGuardian(context, request, sessionFile, keepAlive, restartCount);
+      } catch (error) {
+        workerExit = {
+          exitCode: 1,
+          signal: "",
+          error: String(error?.stack || error || ""),
+        };
+      }
+
+      const outcomeAfterExit = readFinalOutcome(context, request.sessionId);
+      if (outcomeAfterExit) {
+        unregisterActiveSession(request.sessionId);
+        return;
+      }
+
+      if (!keepAlive.enabled) {
+        writeKeepAliveFailure(
+          context,
+          request,
+          "后台守护进程检测到 worker 异常退出，但当前设置已禁用自动保活重拉起。",
+          restartCount,
+          keepAlive.enabled,
+        );
+        refreshContinuation(context, request.sessionId);
+        unregisterActiveSession(request.sessionId);
+        return;
+      }
+
+      const nextRestartCount = restartCount + 1;
+      if (!hasRetryBudget(keepAlive.maxRestartCount, nextRestartCount)) {
+        writeKeepAliveFailure(
+          context,
+          request,
+          `后台守护进程自动重拉起额度已耗尽（已尝试 ${restartCount} 次），worker 仍未恢复。`,
+          restartCount,
+          keepAlive.enabled,
+        );
+        refreshContinuation(context, request.sessionId);
+        unregisterActiveSession(request.sessionId);
+        return;
+      }
+
+      const delaySeconds = pickRetryDelaySeconds(keepAlive.restartDelaysSeconds, nextRestartCount);
+      const exitSummary = workerExit.error
+        ? `spawn 失败：${workerExit.error}`
+        : `exit=${workerExit.exitCode}${workerExit.signal ? `, signal=${workerExit.signal}` : ""}`;
+
+      writeSupervisorState(context, {
+        sessionId: request.sessionId,
+        command: request.command,
+        status: "running",
+        lease: request.lease || {},
+        pid: process.pid,
+        guardianPid: process.pid,
+        workerPid: 0,
+        guardianRestartCount: nextRestartCount,
+        keepAliveEnabled: keepAlive.enabled,
+        lastWorkerExitCode: Number(workerExit.exitCode ?? 1),
+        lastWorkerSignal: workerExit.signal || "",
+        lastWorkerExitedAt: nowIso(),
+        message: `后台守护进程检测到 worker 异常退出（${exitSummary}），将在 ${delaySeconds} 秒后自动重拉起。`,
+      });
+      refreshContinuation(context, request.sessionId);
+
+      if (delaySeconds > 0) {
+        await sleep(delaySeconds * 1000);
+      }
+
+      restartCount = nextRestartCount;
+    }
+  } catch (error) {
+    writeKeepAliveFailure(
+      context,
+      request,
+      String(error?.stack || error || ""),
+      0,
+      keepAlive.enabled,
+    );
+    refreshContinuation(context, request.sessionId);
+    unregisterActiveSession(request.sessionId);
+  }
+}

package/src/supervisor_runtime.mjs

@@ -1,20 +1,28 @@
 import fs from "node:fs";
 import path from "node:path";
-import { spawn } from "node:child_process";
 
 import { createContext } from "./context.mjs";
 import { nowIso, readJson, writeJson, readTextIfExists, timestampForFile } from "./common.mjs";
+import { refreshHostContinuationArtifacts } from "./host_continuation.mjs";
 import { isHostLeaseAlive, renderHostLeaseLabel, resolveHostLease } from "./host_lease.mjs";
+import { spawnNodeProcess } from "./node_process_launch.mjs";
 import { runLoop, runOnce } from "./runner.mjs";
+import {
+  FINAL_SUPERVISOR_STATUSES,
+  hasActiveSupervisor,
+  isTrackedPidAlive,
+  readJsonIfExists,
+  readSupervisorState,
+  writeActiveSupervisorState,
+  writeSupervisorState,
+} from "./supervisor_state.mjs";
 import {
   bindBackgroundTerminalSession,
   cancelPreparedTerminalSessionBackground,
   finalizePreparedTerminalSessionBackground,
   prepareCurrentTerminalSessionForBackground,
 } from "./terminal_session_limits.mjs";
-
-const ACTIVE_STATUSES = new Set(["launching", "running"]);
-const FINAL_STATUSES = new Set(["completed", "failed", "stopped"]);
+import { registerActiveSession, unregisterActiveSession } from "./workspace_registry.mjs";
 
 function removeIfExists(filePath) {
   try {
@@ -24,62 +32,66 @@ function removeIfExists(filePath) {
   }
 }
 
-function readJsonIfExists(filePath) {
-  return fs.existsSync(filePath) ? readJson(filePath) : null;
+function toSerializableOptions(options = {}) {
+  return JSON.parse(JSON.stringify(options));
 }
 
-function isPidAlive(pid) {
-  const numberPid = Number(pid || 0);
-  if (!Number.isFinite(numberPid) || numberPid <= 0) {
-    return false;
-  }
+function refreshContinuation(context, sessionId) {
   try {
-    process.kill(numberPid, 0);
-    return true;
-  } catch (error) {
-    return String(error?.code || "") === "EPERM";
+    refreshHostContinuationArtifacts(context, { sessionId });
+  } catch {
+    // ignore continuation snapshot refresh failures
   }
 }
 
-function buildState(context, patch = {}) {
-  const current = readJsonIfExists(context.supervisorStateFile) || {};
-  return {
-    ...current,
-    ...patch,
-    updatedAt: nowIso(),
+function writeStoppedState(context, request, message) {
+  const payload = {
+    sessionId: request.sessionId,
+    command: request.command,
+    exitCode: 1,
+    ok: false,
+    stopped: true,
+    error: message,
   };
+  writeJson(context.supervisorResultFile, payload);
+  writeSupervisorState(context, {
+    sessionId: request.sessionId,
+    command: request.command,
+    status: "stopped",
+    exitCode: payload.exitCode,
+    message,
+    completedAt: nowIso(),
+  });
 }
 
-function writeState(context, patch) {
-  writeJson(context.supervisorStateFile, buildState(context, patch));
-}
-
-function toSerializableOptions(options = {}) {
-  return JSON.parse(JSON.stringify(options));
-}
-
-export function readSupervisorState(context) {
-  return readJsonIfExists(context.supervisorStateFile);
-}
-
-export function hasActiveSupervisor(context) {
-  const state = readSupervisorState(context);
-  return Boolean(state?.status && ACTIVE_STATUSES.has(String(state.status)) && isPidAlive(state.pid));
+function buildSupervisorSessionState(sessionId, command, lease, pid, message) {
+  return {
+    sessionId,
+    status: "running",
+    command,
+    lease,
+    startedAt: nowIso(),
+    pid,
+    guardianPid: pid,
+    workerPid: 0,
+    message,
+  };
 }
 
 export function renderSupervisorLaunchSummary(session) {
   return [
-    `HelloLoop supervisor 已启动：${session.sessionId}`,
+    `HelloLoop 后台守护进程已启动：${session.sessionId}`,
     `- 宿主租约：${renderHostLeaseLabel(session.lease)}`,
     "- 当前 turn 若被中断，只要当前宿主窗口仍存活，本轮自动执行会继续。",
+    "- 守护进程会在 worker 异常退出后按设置自动重拉起，尽量保持后台持续活跃。",
     "- 如需主动停止，直接关闭当前 CLI 窗口即可。",
   ].join("\n");
 }
 
 export function launchSupervisedCommand(context, command, options = {}) {
   const existing = readSupervisorState(context);
-  if (existing?.status && ACTIVE_STATUSES.has(String(existing.status)) && isPidAlive(existing.pid)) {
-    throw new Error(`已有 HelloLoop supervisor 正在运行：${existing.sessionId || "unknown"}`);
+  if (hasActiveSupervisor(context)) {
+    throw new Error(`已有 HelloLoop supervisor 正在运行：${existing?.sessionId || "unknown"}`);
   }
 
   const sessionId = timestampForFile();
@@ -96,7 +108,10 @@ export function launchSupervisedCommand(context, command, options = {}) {
       repoRoot: context.repoRoot,
       configDirName: context.configDirName,
     },
-    options: toSerializableOptions(options),
+    options: toSerializableOptions({
+      ...options,
+      supervisorSessionId: sessionId,
+    }),
     lease,
     terminalSessionFile: terminalSession?.file || "",
   };
@@ -106,47 +121,62 @@ export function launchSupervisedCommand(context, command, options = {}) {
   removeIfExists(context.supervisorStdoutFile);
   removeIfExists(context.supervisorStderrFile);
   writeJson(context.supervisorRequestFile, request);
-  writeState(context, {
+  writeActiveSupervisorState(context, {
     sessionId,
     status: "launching",
     command,
     lease,
     startedAt: nowIso(),
     pid: 0,
+    guardianPid: 0,
+    workerPid: 0,
   });
+  refreshContinuation(context, sessionId);
 
   const stdoutFd = fs.openSync(context.supervisorStdoutFile, "w");
   const stderrFd = fs.openSync(context.supervisorStderrFile, "w");
+
   try {
-    const child = spawn(process.execPath, [
-      path.join(context.bundleRoot, "bin", "helloloop.js"),
-      "__supervise",
-      "--session-file",
-      context.supervisorRequestFile,
-    ], {
+    const child = spawnNodeProcess({
+      args: [
+        path.join(context.bundleRoot, "bin", "helloloop.js"),
+        "__supervise",
+        "--session-file",
+        context.supervisorRequestFile,
+      ],
       cwd: context.repoRoot,
       detached: true,
-      shell: false,
-      windowsHide: true,
       stdio: ["ignore", stdoutFd, stderrFd],
       env: {
-        ...process.env,
         HELLOLOOP_SUPERVISOR_ACTIVE: "1",
       },
     });
+
     finalizePreparedTerminalSessionBackground(child.pid ?? 0, {
       command,
       repoRoot: context.repoRoot,
       sessionId,
     });
     child.unref();
-    writeState(context, {
+
+    writeActiveSupervisorState(
+      context,
+      buildSupervisorSessionState(sessionId, command, lease, child.pid ?? 0, "后台守护进程正在运行。"),
+    );
+    refreshContinuation(context, sessionId);
+
+    registerActiveSession({
       sessionId,
-      status: "running",
+      repoRoot: context.repoRoot,
+      configDirName: context.configDirName,
       command,
+      pid: child.pid ?? 0,
+      guardianPid: child.pid ?? 0,
       lease,
       startedAt: nowIso(),
-      pid: child.pid ?? 0,
+      supervisorStateFile: context.supervisorStateFile,
+      supervisorResultFile: context.supervisorResultFile,
+      statusFile: context.statusFile,
     });
 
     return {
@@ -155,6 +185,7 @@ export function launchSupervisedCommand(context, command, options = {}) {
       lease,
     };
   } catch (error) {
+    unregisterActiveSession(sessionId);
     cancelPreparedTerminalSessionBackground({
       command,
       repoRoot: context.repoRoot,
@@ -177,7 +208,7 @@ export async function waitForSupervisedResult(context, session, options = {}) {
     }
 
     const state = readSupervisorState(context);
-    if (state?.sessionId === session.sessionId && FINAL_STATUSES.has(String(state.status || ""))) {
+    if (state?.sessionId === session.sessionId && FINAL_SUPERVISOR_STATUSES.has(String(state.status || ""))) {
       return {
         sessionId: session.sessionId,
         command: state.command || "",
@@ -187,7 +218,7 @@ export async function waitForSupervisedResult(context, session, options = {}) {
       };
     }
 
-    if (!isPidAlive(session.pid)) {
+    if (!isTrackedPidAlive(session.pid)) {
       return {
         sessionId: session.sessionId,
         command: state?.command || "",
@@ -208,44 +239,60 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
   const context = createContext(request.context || {});
   const command = String(request.command || "").trim();
   const lease = request.lease || {};
-  bindBackgroundTerminalSession(request.terminalSessionFile || "", {
-    command,
-    repoRoot: context.repoRoot,
-    sessionId: request.sessionId,
-  });
+  const guardianManaged = request.options?.guardianManaged === true
+    || process.env.HELLOLOOP_SUPERVISOR_GUARDIAN_ACTIVE === "1";
+  const guardianPid = guardianManaged
+    ? Number(request.options?.guardianPid || process.env.HELLOLOOP_SUPERVISOR_GUARDIAN_PID || process.pid)
+    : 0;
+  const supervisorPid = guardianPid > 0 ? guardianPid : process.pid;
+
+  if (!guardianManaged) {
+    bindBackgroundTerminalSession(request.terminalSessionFile || "", {
+      command,
+      repoRoot: context.repoRoot,
+      sessionId: request.sessionId,
+    });
+  }
+
   const commandOptions = {
     ...(request.options || {}),
     hostLease: lease,
   };
-
-  writeState(context, {
+  writeActiveSupervisorState(context, {
     sessionId: request.sessionId,
     command,
     status: "running",
     lease,
-    pid: process.pid,
+    pid: supervisorPid,
+    guardianPid,
+    workerPid: guardianManaged ? process.pid : 0,
     startedAt: nowIso(),
   });
+  refreshContinuation(context, request.sessionId);
+
+  if (!guardianManaged) {
+    registerActiveSession({
+      sessionId: request.sessionId,
+      repoRoot: context.repoRoot,
+      configDirName: context.configDirName,
+      command,
+      pid: supervisorPid,
+      guardianPid,
+      lease,
+      startedAt: nowIso(),
+      supervisorStateFile: context.supervisorStateFile,
+      supervisorResultFile: context.supervisorResultFile,
+      statusFile: context.statusFile,
+    });
+  }
 
   try {
     if (!isHostLeaseAlive(lease)) {
-      const stopped = {
-        sessionId: request.sessionId,
-        command,
-        exitCode: 1,
-        ok: false,
-        stopped: true,
-        error: "检测到宿主窗口已关闭，HelloLoop supervisor 未继续执行。",
-      };
-      writeJson(context.supervisorResultFile, stopped);
-      writeState(context, {
-        sessionId: request.sessionId,
-        command,
-        status: "stopped",
-        exitCode: stopped.exitCode,
-        message: stopped.error,
-        completedAt: nowIso(),
-      });
+      writeStoppedState(context, request, "检测到宿主窗口已关闭，HelloLoop supervisor 未继续执行。");
+      refreshContinuation(context, request.sessionId);
+      if (!guardianManaged) {
+        unregisterActiveSession(request.sessionId);
+      }
       return;
     }
 
@@ -260,7 +307,7 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
         results,
       };
       writeJson(context.supervisorResultFile, payload);
-      writeState(context, {
+      writeSupervisorState(context, {
         sessionId: request.sessionId,
         command,
         status: payload.ok ? "completed" : (payload.results.some((item) => item.kind === "host-lease-stopped") ? "stopped" : "failed"),
@@ -268,6 +315,10 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
         message: payload.ok ? "" : (payload.results.find((item) => !item.ok)?.summary || ""),
         completedAt: nowIso(),
       });
+      refreshContinuation(context, request.sessionId);
+      if (!guardianManaged) {
+        unregisterActiveSession(request.sessionId);
+      }
       return;
     }
 
@@ -281,7 +332,7 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
         result,
       };
       writeJson(context.supervisorResultFile, payload);
-      writeState(context, {
+      writeSupervisorState(context, {
         sessionId: request.sessionId,
         command,
         status: payload.ok ? "completed" : (result.kind === "host-lease-stopped" ? "stopped" : "failed"),
@@ -289,6 +340,10 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
         message: result.summary || result.finalMessage || "",
         completedAt: nowIso(),
       });
+      refreshContinuation(context, request.sessionId);
+      if (!guardianManaged) {
+        unregisterActiveSession(request.sessionId);
+      }
       return;
     }
 
@@ -302,7 +357,7 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
       error: String(error?.stack || error || ""),
     };
     writeJson(context.supervisorResultFile, payload);
-    writeState(context, {
+    writeSupervisorState(context, {
       sessionId: request.sessionId,
       command,
       status: "failed",
@@ -310,5 +365,9 @@ export async function runSupervisedCommandFromSessionFile(sessionFile) {
       message: payload.error,
       completedAt: nowIso(),
     });
+    refreshContinuation(context, request.sessionId);
+    if (!guardianManaged) {
+      unregisterActiveSession(request.sessionId);
+    }
   }
 }