oasis_test 0.1.118 → 0.1.119

package/dist/index.js

@@ -4801,6 +4801,11 @@ function createPolicy(cfg) {
     }
   };
 }
+function policyGetter(source, fallback = CODE_DEFAULT_POLICY) {
+  if (typeof source === "function") return source;
+  if (source) return () => source;
+  return () => fallback;
+}
 var MINUTE, HOUR, DAY, AGENT_WORK_DEADLINE_MS, AGENT_REVIEW_DEADLINE_MS, AGENT_REVIEW_MAX_RETRIES, AGENT_WORK_MAX_RETRIES, RETRY_BASE_MS, RUNTIME_MAX_RETRIES, DEFAULT_IDLE_BY_RUNTIME, DEFAULT_IDLE_MS, DEFAULT_WALL_CLOCK_BY_RUNTIME, DEFAULT_WALL_CLOCK_MS, DEFAULT_HUMAN_CADENCE, CODE_DEFAULT_POLICY;
 var init_policy = __esm({
   "../timeout-policy/src/policy.ts"() {
@@ -7571,6 +7576,10 @@ function inputsMatch(state, node, latestWork) {
   const latestInputs = new Set(state.workInputsOf(latestWork.id).map((i) => i.upstreamWorkId));
   return currInputs.size === latestInputs.size && [...currInputs].every((x2) => latestInputs.has(x2));
 }
+function reviewMaxRetriesOf(policy) {
+  const act = policy.resolve({ executor: "agent", scenario: "review" }).timeoutAction;
+  return act.kind === "kill-rebuild" ? act.maxRetries : AGENT_REVIEW_MAX_RETRIES;
+}
 function processReviews(state, nodeId, workId, ctx, events) {
   const policy = ctx.policy ?? CODE_DEFAULT_POLICY;
   const reqs = state.requirementsOf(nodeId);
@@ -7620,7 +7629,7 @@ function processReviews(state, nodeId, workId, ctx, events) {
       }
     } else if (info.status === "failed") {
       const req = reqs.find((r) => r.reviewGroup === g2);
-      const overLimit = countFailedForReviewGroup(g2, reviews) > AGENT_REVIEW_MAX_RETRIES;
+      const overLimit = countFailedForReviewGroup(g2, reviews) > reviewMaxRetriesOf(policy);
       if (overLimit && classify(req.reviewerActorId) === "agent") {
         const lrId = req.latestReviewId;
         if (lrId) events.push({
@@ -15427,7 +15436,8 @@ var init_bus = __esm({
       io;
       clock;
       config;
-      policy;
+      /** 取当前策略（不是策略本身）——见 {@link EngineBusOptions.policy} 的热载不变量。 */
+      policyOf;
       registry;
       maxEffectAttempts;
       stuckAppliedRecoveryMs;
@@ -15439,7 +15449,7 @@ var init_bus = __esm({
         this.io = opts.io ?? NOOP_IO;
         this.clock = opts.clock ?? SYSTEM_CLOCK;
         this.config = opts.config ?? DEFAULT_CONFIG;
-        this.policy = opts.policy ?? CODE_DEFAULT_POLICY;
+        this.policyOf = policyGetter(opts.policy);
         this.registry = new HandlerRegistry(opts.handlers);
         this.maxEffectAttempts = opts.maxEffectAttempts ?? 3;
         this.stuckAppliedRecoveryMs = opts.stuckAppliedRecoveryMs ?? DEFAULT_STUCK_APPLIED_RECOVERY_MS;
@@ -15828,7 +15838,7 @@ var init_bus = __esm({
           if (!snapshot) return { emitted: 0, terminated: false };
           if (snapshot.workorder.lifecycle !== "active") return { emitted: 0, terminated: false };
           const state = new WorkorderState(snapshot);
-          const scanCtx = { at: this.clock.now(), actorId: SYSTEM_ACTOR2, config: this.config, policy: this.policy };
+          const scanCtx = { at: this.clock.now(), actorId: SYSTEM_ACTOR2, config: this.config, policy: this.policyOf() };
           const { events, terminated } = scan(state, scanCtx);
           await tx.applyMutations(workorderId, state.mutations);
           const allEvents = [...state.emitted, ...events];
@@ -128738,13 +128748,13 @@ ${block}
               throw err;
             }
             const delay = base * Math.pow(3, attempt - 1);
-            yield sleep4(delay);
+            yield sleep5(delay);
           }
         }
         throw lastErr;
       });
     }
-    function sleep4(ms) {
+    function sleep5(ms) {
       return new Promise((r) => setTimeout(r, ms));
     }
     var DEFAULT_CHUNK_LIMIT = 3500;
@@ -173922,7 +173932,9 @@ async function openNewEngine(opts) {
   const bus = new EngineBus({
     store,
     io: NOOP_IO,
-    clock: SYSTEM_CLOCK
+    clock: SYSTEM_CLOCK,
+    // 原样交给 bus（bus 内部归一成取值函数）——这里绝不能 call 一次存下来，那就把热载掐死了。
+    policy: opts.policy
   });
   const kernel = await EngineKernel.open(store, bus, {
     schema: opts.schemaDefs ?? [],
@@ -198389,7 +198401,10 @@ async function startServe(opts) {
     schemaDefs: schema,
     registry: kernelRegistry,
     reworkWavefront,
-    convergenceUnified
+    convergenceUnified,
+    // ★ ADR 0147：引擎 scan 的 work/review 超时与重试上限走策略中心（与 dispatcher 墙钟、
+    // waiting-inbox 催办同源）。thunk 而非实例——随 SIGHUP / POST /api/policy/reload 热载而变。
+    policy: () => activeTimeoutPolicy
   });
   const kernel = newEngine.kernel;
   const dispatchLedger = await PostgresDispatchStore.open(pgPool, pgSchema);
@@ -198998,7 +199013,9 @@ async function startServe(opts) {
         schemaDefs: schema,
         registry: kernelRegistry,
         reworkWavefront,
-        convergenceUnified
+        convergenceUnified,
+        // ★ 同默认公司：漏了这里，非默认公司的引擎照旧按代码默认 30min 判超时（策略文件形同虚设）。
+        policy: () => activeTimeoutPolicy
       });
       await ensureCompanyActors(controlPlaneStore, companyId, registryStore);
       await syncRegistryRolesToKernel(registryStore, engine2.kernel);
@@ -201950,6 +201967,9 @@ var PENDING_STREAM_MAX = 2e3;
 var AUTH_REJECT_BACKOFF_MAX_MS = 15 * 6e4;
 var AUTH_REJECT_GIVE_UP_AFTER = 3;
 var AUTH_REJECT_RE = /Unexpected server response:\s*(401|403)/;
+var DRAIN_TIMEOUT_MS = 15e3;
+var DRAIN_POLL_MS = 50;
+var sleep4 = (ms) => new Promise((r) => setTimeout(r, ms));
 async function preflightOasisAccess(job) {
   const base = job.server.url.replace(/\/+$/, "");
   try {
@@ -201994,6 +202014,8 @@ var DaemonWsClient = class {
   stopped = false;
   /** 收到 update 但有活跃会话时置真：延后到所有会话结束再自更新，避免打断执行中的 agent。 */
   pendingUpdate = false;
+  /** 停机排空中（见 {@link drain}）：不再接新派发、不再触发延后的自更新。 */
+  draining = false;
   /** 已回 dispatch_received、尚未完成 runtime spawn 的派发；也随 hello 上报供重连对账。 */
   pendingDispatchIds = /* @__PURE__ */ new Set();
   /** setup 期间收到 kill 的派发。spawn 返回后先杀句柄，不得再宣告 session_started。 */
@@ -202050,6 +202072,73 @@ var DaemonWsClient = class {
     if (this.pingTimer) clearInterval(this.pingTimer);
     this.ws?.close();
   }
+  /** 排空中？（观测/测试用） */
+  isDraining() {
+    return this.draining;
+  }
+  /**
+   * 停机排空——在进程退出前，把「在途会话已经死了」这件事告诉服务端。
+   *
+   * ## 为什么必须有这一步（2026-08-25 生产实证）
+   *
+   * 引擎里一条 work 从 `running` 变成别的状态，全后端只有三个入口：收到 `session_exited` 帧走主路
+   * （`serve.ts` 的 work.response 桥）、同一帧走 journal 反查的兜底、以及 scan 里 **30 分钟**的
+   * `work.timeout`。前两个都要求**这一帧真的被发出来**。
+   *
+   * 而这一帧的生成者，是本进程 dispatch 分支里注册的那个 `handle.onExit` 回调——**它住在 daemon
+   * 进程内部**。子进程被杀不会自动产生任何帧；进程一死，回调就再没机会跑。
+   *
+   * 旧行为：SIGTERM handler 只删 pid 文件就 `process.exit(0)`；systemd unit 又没设 `KillMode`，
+   * 默认 `control-group` 把 daemon 和它拉起的所有会话**同时** SIGTERM——父子同死，一帧都发不出去。
+   * 实测 node-6745008c1fc7：5389 次会话起跑里 48 个没有配对退出帧，其中 43 个（90%）是重启带走的；
+   * 每一个都让对应的 work 空转到 30 分钟超时才被判死，控制台全程显示"运行中"。
+   *
+   * ## 做法：让真实退出帧走正路，只有两类才合成
+   *
+   * 不图省事直接合成假帧——`kill()` 掉在途会话，adapter 的 `killReason` 归因成 `cancelled`，既有
+   * `onExit` 正路会带着 `transcriptRef` 把**真实**帧发出去。只有两类兜底才合成：还没 spawn 的派发
+   * （没有子进程可 kill，不可能有真实帧），以及超时后仍未退出的会话（宁可给服务端一帧不精确的死讯，
+   * 也好过让它等满 30 分钟）。
+   *
+   * 排空期间 `draining` 置真：不再接新派发（见 dispatch 分支），也不再触发延后的自更新。
+   */
+  async drain(reason, timeoutMs = DRAIN_TIMEOUT_MS) {
+    if (this.draining) return;
+    this.draining = true;
+    const active = this.sessions.activeSessions();
+    const pending = [...this.pendingDispatchIds];
+    if (active.length === 0 && pending.length === 0) {
+      this.stop();
+      return;
+    }
+    log2("[node-cli]", `\u505C\u673A\u6392\u7A7A\uFF08${reason}\uFF09\uFF1A\u5728\u9014\u4F1A\u8BDD ${active.length} \u4E2A\u3001\u5F85\u8D77\u6D3E\u53D1 ${pending.length} \u4E2A\u2014\u2014\u53D1\u5B8C\u9000\u51FA\u5E27\u518D\u9000\u51FA`);
+    for (const dispatchId of pending) {
+      this.pendingDispatchIds.delete(dispatchId);
+      this.sendKeyFrame(dispatchId, {
+        type: "session_exited",
+        dispatchId,
+        sessionId: "node-shutdown",
+        info: { code: null, reason: "cancelled", errorMessage: `\u8282\u70B9 daemon \u505C\u673A\uFF08${reason}\uFF09\uFF1A\u6D3E\u53D1\u5C1A\u672A\u8D77\u4F1A\u8BDD` }
+      });
+    }
+    await Promise.all(active.map((s2) => this.sessions.kill(s2.dispatchId).catch(() => void 0)));
+    const deadline = Date.now() + timeoutMs;
+    while (Date.now() < deadline && (this.sessions.activeCount() > 0 || this.outbox.size > 0)) await sleep4(DRAIN_POLL_MS);
+    const stuck = this.sessions.activeSessions();
+    for (const s2 of stuck) {
+      log2("[node-cli]", `  \u26A0 \u4F1A\u8BDD ${s2.sessionId} \u5728 ${timeoutMs}ms \u5185\u6CA1\u9000\u51FA\uFF0C\u6309\u505C\u673A\u5408\u6210\u9000\u51FA\u5E27`);
+      this.sendKeyFrame(s2.dispatchId, {
+        type: "session_exited",
+        dispatchId: s2.dispatchId,
+        sessionId: s2.sessionId,
+        info: { code: null, reason: "cancelled", errorMessage: `\u8282\u70B9 daemon \u505C\u673A\uFF08${reason}\uFF09\uFF1A\u4F1A\u8BDD\u672A\u5728\u6392\u7A7A\u7A97\u53E3\u5185\u9000\u51FA` }
+      });
+    }
+    if (stuck.length > 0) await sleep4(DRAIN_POLL_MS * 4);
+    if (this.outbox.size > 0) log2("[node-cli]", `\u26A0 \u505C\u673A\u6392\u7A7A\u7ED3\u675F\uFF1A\u4ECD\u6709 ${this.outbox.size} \u4E2A\u9000\u51FA\u5E27\u672A\u786E\u8BA4\u9001\u8FBE\uFF08\u8FD9\u4E9B work \u4F1A\u9000\u56DE 30 \u5206\u949F\u8D85\u65F6\u515C\u5E95\uFF09`);
+    else log2("[node-cli]", "\u505C\u673A\u6392\u7A7A\u5B8C\u6210\uFF1A\u5728\u9014\u4F1A\u8BDD\u7684\u9000\u51FA\u5E27\u5747\u5DF2\u786E\u8BA4\u9001\u8FBE");
+    this.stop();
+  }
   connect() {
     if (this.stopped) return;
     const ws = new import_websocket.default(this.serverUrl, {
@@ -202126,6 +202215,16 @@ var DaemonWsClient = class {
         const { dispatchId, job } = msg;
         const t0 = Date.now();
         log2("[node-cli]", `\u2190 dispatch  ${dispatchId}  artifact=${job.artifactId}`);
+        if (this.draining) {
+          log2("[node-cli]", `  \u62D2\u6536\uFF1A\u8282\u70B9\u6B63\u5728\u505C\u673A\u6392\u7A7A \u2192 \u56DE\u9000\u51FA\u5E27\uFF0C\u670D\u52A1\u7AEF\u53EF\u7ACB\u5373\u6539\u6D3E`);
+          this.sendKeyFrame(dispatchId, {
+            type: "session_exited",
+            dispatchId,
+            sessionId: "node-draining",
+            info: { code: null, reason: "cancelled", errorMessage: "\u8282\u70B9\u6B63\u5728\u505C\u673A\u6392\u7A7A\uFF0C\u672C\u6B21\u6D3E\u53D1\u672A\u8D77\u4F1A\u8BDD" }
+          });
+          break;
+        }
         if (this.pendingDispatchIds.has(dispatchId) || this.sessions.get(dispatchId)) {
           this.send({ type: "dispatch_received", dispatchId });
           break;
@@ -202243,6 +202342,10 @@ var DaemonWsClient = class {
           log2("[node-cli]", "\u2190 update \u5FFD\u7565\uFF1A\u672C\u8FDB\u7A0B\u672A\u6CE8\u5165\u81EA\u66F4\u65B0\u903B\u8F91\uFF08\u975E node-daemon \u6258\u7BA1\uFF1F\uFF09");
           break;
         }
+        if (this.draining) {
+          log2("[node-cli]", "\u2190 update \u5FFD\u7565\uFF1A\u672C\u8FDB\u7A0B\u6B63\u5728\u505C\u673A\u6392\u7A7A");
+          break;
+        }
         const active = this.sessions.activeCount();
         if (active > 0) {
           this.pendingUpdate = true;
@@ -202330,6 +202433,7 @@ var DaemonWsClient = class {
   }
   /** 会话结束后调用：若之前收到过 update 但因忙延后了,且现已空闲,则触发自更新。 */
   maybeRunPendingUpdate() {
+    if (this.draining) return;
     if (this.pendingUpdate && this.sessions.activeCount() === 0) {
       this.pendingUpdate = false;
       log2("[node-cli]", "\u6240\u6709\u4F1A\u8BDD\u5DF2\u7ED3\u675F,\u6267\u884C\u5EF6\u540E\u7684\u66F4\u65B0\uFF1A\u62C9\u53D6\u6700\u65B0\u7248\u672C\u5E76\u91CD\u542F");
@@ -202493,6 +202597,7 @@ async function startNode(opts) {
   });
   console.log(`[oasis node] ${opts.nodeId} \u2192 ${opts.serverUrl}`);
   client.start();
+  opts.onReady?.({ drain: (reason) => client.drain(reason) });
   const { workRoot, legacyRoot } = resolveWorkRoots(opts.workRoot);
   const workRootIsEphemeral = workRoot.startsWith(import_node_os11.default.tmpdir());
   if (opts.gcEnabled === false && !workRootIsEphemeral) {
@@ -207460,6 +207565,37 @@ function gcVersions(versionsDirPath, opts) {
   }
   return deleted;
 }
+function systemdUnitText(o) {
+  return [
+    "[Unit]",
+    `Description=Oasis Node Daemon (${o.name})`,
+    "After=network.target",
+    "",
+    // OASIS_CLI_HOME 由 daemon 继承给它 spawn 的 agent 会话——非 default 实例的会话因此
+    // 用的是**本实例**的 CLI 版本，而不是 PATH 上那个默认指向 default 的 shim。
+    // （台阶 3 的 D4 后半截会把它细到「每个会话钉 spawn 那一刻的版本」，这里先钉到实例。）
+    "[Service]",
+    `ExecStart=${o.cmd}`,
+    `Environment=PATH=${o.daemonPath}`,
+    `Environment=OASIS_CLI_HOME=${o.cliHome}`,
+    `StandardOutput=append:${o.logFile}`,
+    `StandardError=append:${o.logFile}`,
+    "Restart=on-failure",
+    "RestartSec=5",
+    // 🚨 停机时只 SIGTERM 主进程，子会话留给 daemon 自己收——它要先给每个在途会话发 session_exited，
+    // 那是服务端唯一的死亡通知渠道（生成者是 daemon 进程内的 onExit 回调）。
+    // systemd 默认的 KillMode=control-group 会把 daemon 和它拉起的所有 claude 会话**同时**杀掉：
+    // 父子同死，一帧都发不出去，对应的 work 只能空转到 scan 的 30 分钟超时才被判死。
+    // 实测占本机孤儿会话的 90%（43/48）。详见 change-record 20260825-daemon-restart-swallows-exit-frames。
+    // TimeoutStopSec 必须明显大于排空预算 DRAIN_TIMEOUT_MS(15s)，否则排空没做完就被 SIGKILL。
+    "KillMode=mixed",
+    "TimeoutStopSec=30",
+    "",
+    "[Install]",
+    "WantedBy=default.target",
+    ""
+  ].join("\n");
+}
 function shimScript() {
   return [
     "#!/bin/sh",
@@ -207472,7 +207608,7 @@ function shimScript() {
 }
 
 // src/index.ts
-var PKG_VERSION = true ? "0.1.118" : "dev";
+var PKG_VERSION = true ? "0.1.119" : "dev";
 var LOCAL_BIN = localBin();
 var NPM_PREFIX = npmPrefix();
 var INSTANCE = DEFAULT_INSTANCE;
@@ -207586,27 +207722,13 @@ function setupAutostart(name) {
     fs38.mkdirSync(unitDir, { recursive: true });
     const unit = systemdUnitName(name);
     fs38.mkdirSync(path32.dirname(inst.logFile), { recursive: true });
-    fs38.writeFileSync(path32.join(unitDir, unit), [
-      "[Unit]",
-      `Description=Oasis Node Daemon (${name})`,
-      "After=network.target",
-      "",
-      // OASIS_CLI_HOME 由 daemon 继承给它 spawn 的 agent 会话——非 default 实例的会话因此
-      // 用的是**本实例**的 CLI 版本，而不是 PATH 上那个默认指向 default 的 shim。
-      // （台阶 3 的 D4 后半截会把它细到「每个会话钉 spawn 那一刻的版本」，这里先钉到实例。）
-      "[Service]",
-      `ExecStart=${cmd}`,
-      `Environment=PATH=${daemonPath}`,
-      `Environment=OASIS_CLI_HOME=${inst.currentLink}`,
-      `StandardOutput=append:${inst.logFile}`,
-      `StandardError=append:${inst.logFile}`,
-      "Restart=on-failure",
-      "RestartSec=5",
-      "",
-      "[Install]",
-      "WantedBy=default.target",
-      ""
-    ].join("\n"));
+    fs38.writeFileSync(path32.join(unitDir, unit), systemdUnitText({
+      name,
+      cmd,
+      daemonPath,
+      cliHome: inst.currentLink,
+      logFile: inst.logFile
+    }));
     if (name === DEFAULT_INSTANCE) {
       try {
         (0, import_node_child_process19.execSync)(`systemctl --user disable --now ${LEGACY_SYSTEMD_UNIT} 2>/dev/null`, { stdio: "ignore" });
@@ -207876,19 +207998,32 @@ void (async () => {
     }
     const cleanup = () => {
       try {
-        fs38.unlinkSync(PID_FILE);
+        if (fs38.readFileSync(PID_FILE, "utf8").trim() === String(process.pid)) fs38.unlinkSync(PID_FILE);
       } catch {
       }
     };
     process.on("exit", cleanup);
-    process.on("SIGTERM", () => {
-      cleanup();
-      process.exit(0);
-    });
-    process.on("SIGINT", () => {
-      cleanup();
-      process.exit(0);
-    });
+    let drainNode = null;
+    let shuttingDown = false;
+    const shutdown = (signal) => {
+      if (shuttingDown) {
+        console.error(`[oasis] \u6536\u5230\u7B2C\u4E8C\u6B21 ${signal}\uFF1A\u653E\u5F03\u6392\u7A7A\uFF0C\u7ACB\u5373\u9000\u51FA`);
+        cleanup();
+        process.exit(0);
+      }
+      shuttingDown = true;
+      void (async () => {
+        try {
+          await drainNode?.(signal);
+        } catch (e) {
+          console.error("[oasis] \u505C\u673A\u6392\u7A7A\u5931\u8D25\uFF08\u5728\u9014\u4F1A\u8BDD\u7684\u9000\u51FA\u5E27\u53EF\u80FD\u6CA1\u53D1\u51FA\u53BB\uFF09:", String(e));
+        }
+        cleanup();
+        process.exit(0);
+      })();
+    };
+    process.on("SIGTERM", () => shutdown("SIGTERM"));
+    process.on("SIGINT", () => shutdown("SIGINT"));
     const cleanPath = sanitizeDaemonPath(process.env["PATH"]);
     if (cleanPath !== process.env["PATH"]) {
       console.error(`[oasis] daemon PATH \u5DF2\u6E05\u6D17\uFF1A\u5254\u9664 agent \u4F1A\u8BDD\u4E34\u65F6 wrapper \u76EE\u5F55 ${(process.env["PATH"] ?? "").split(":").length - cleanPath.split(":").length} \u6761`);
@@ -207928,6 +208063,9 @@ void (async () => {
         } catch (e) {
           console.error("[oasis] self-update failed:", String(e));
         }
+      },
+      onReady: (ctl) => {
+        drainNode = ctl.drain;
       }
     });
     return;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oasis_test",
-  "version": "0.1.118",
+  "version": "0.1.119",
   "description": "Oasis node daemon + CLI — background daemon, auto-start, full server CLI",
   "bin": {
     "oasis": "./dist/index.js"