phase3(契约层): protocol.ts + 落盘状态 + store 失败策略/判活原语

多进程执行解耦的 Phase 0 共享契约(worker 与 daemon 都 import 它):
- executor/protocol.ts: JobSpec(daemon→worker 唯一输入) + OutboxRecord(worker→daemon) +
  文件 IO(job/outbox/heartbeat) + isWorkerAlive(心跳为主/boot 窗口 pid 兜底/防 pid 复用)
- schema+db+types+mappers: runs.worker_pid/last_seq、tasks.next_eligible_at(持久化退避)
- store: failTaskAttempt(失败/重试策略唯一落点) + setNextEligibleAt/setWorkerPid/setLastSeq +
  countExecuting + executingWithLatestExecutorRun + reconcileInterrupted(isAlive 注入,默认保守回收)

纯增量,140 测试全绿。worker.ts/pipeline.ts(worker 侧) 与 orchestrator/ingest/index(daemon 侧) 待接。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
wangjia
2026-06-13 12:51:09 +08:00
parent d49192344f
commit 007dffac39
7 changed files with 269 additions and 12 deletions
+162
View File
@@ -0,0 +1,162 @@
// Worker ↔ daemon 契约(Phase 3 多进程执行)。
//
// 铁律:worker 进程【完全不碰 DB】(连读都不碰)。daemon 是唯一 DB 读写者。
// 二者只经【文件 + 进程信号】通讯,天然跨 daemon 重启持久:
// daemon → workerruns/<runId>/job.json(唯一输入);SIGTERM(取消/超时)
// worker → daemonruns/<runId>/outbox.ndjson(追加、带 seq);runs/<runId>/heartbeatmtime 心跳)
//
// 本模块同时被 worker 侧(写 outbox/读 job/刷心跳)与 daemon 侧(写 job/读 outbox/判活)import
// 是唯一的共享面——双方都不应另起自己的协议常量/路径计算。
import {
existsSync, mkdirSync, readFileSync, writeFileSync, appendFileSync,
statSync, utimesSync, closeSync, openSync,
} from 'node:fs';
import { homedir } from 'node:os';
import { join } from 'node:path';
import type { Project, Task, ReviewVerdict } from '../model/types.js';
/** 数据根目录(与 worktree.ts 同约定):<MAESTRO_DATA_DIR 或 ~/.maestro> */
function dataDir(): string {
return process.env.MAESTRO_DATA_DIR ?? join(homedir(), '.maestro');
}
/** 全部 run 工作目录根:<data>/runs */
export function runsBase(): string {
return join(dataDir(), 'runs');
}
/** 单个 run 的工作目录:<runsBase>/<runId>(放 job.json / outbox.ndjson / heartbeat */
export function runDir(runId: string): string {
return join(runsBase(), runId);
}
export function jobPath(runId: string): string { return join(runDir(runId), 'job.json'); }
export function outboxPath(runId: string): string { return join(runDir(runId), 'outbox.ndjson'); }
export function heartbeatPath(runId: string): string { return join(runDir(runId), 'heartbeat'); }
// ───────────────────────── daemon → workerJobSpec ─────────────────────────
/**
* worker 的唯一输入。daemon 在 spawn 前写 runs/<runId>/job.jsonworker 读它即可执行,
* 【无需访问 DB】。携带完整 Task / Project(均 JSON 可序列化),worker 自行 pickModel/buildPrompt。
*/
export interface JobSpec {
runId: string;
task: Task;
project: Project;
/** 确定性 worktree 路径与分支(daemon 预填;worker 也能由 worktree.ts 算出,传入避免重复) */
worktreeDir: string;
branch: string;
}
export function writeJobSpec(job: JobSpec): void {
mkdirSync(runDir(job.runId), { recursive: true });
writeFileSync(jobPath(job.runId), JSON.stringify(job));
}
export function readJobSpec(runId: string): JobSpec {
return JSON.parse(readFileSync(jobPath(runId), 'utf8')) as JobSpec;
}
// ───────────────────────── worker → daemonOutboxRecord ────────────────────
/** 一次复审(code review / 安全审计)的产出,worker 报给 daemon,由 daemon 落成 run 行 + result 字段 */
export interface ReviewReport {
summary: string | null;
verdict: ReviewVerdict | null;
transcriptRef: string | null;
}
/**
* worker 追加进 outbox 的记录。`seq` 单调递增(每 run 内),`at` ISO 时间,均由 appendOutbox 填。
* - startedworker 启动,报自己的 pid / worktree / branch / model
* - phase :进度阶段(executing|verifying|reviewing…),daemon 落成 event 推看板(可选)
* - failed :本次尝试终态失败(executor 报错或 verify 失败)。daemon 据此 finishRun(failed)+重试决策
* - result :成功终态。daemon 据此建 reviewer/security run + setResult(四字段) + 转 exec_review
* - done :worker 即将退出(成功或失败都发,daemon 据此停止 tail
*/
export type OutboxRecord =
| { seq: number; at: string; type: 'started'; pid: number; worktree: string; branch: string; model: string }
| { seq: number; at: string; type: 'phase'; phase: string }
| { seq: number; at: string; type: 'failed'; error: string; transcriptRef: string | null; sessionId: string | null }
| {
seq: number; at: string; type: 'result';
branch: string; worktree: string; diffSummary: string; commits: string[];
executor: { transcriptRef: string | null; sessionId: string | null };
code: ReviewReport;
security: ReviewReport;
}
| { seq: number; at: string; type: 'done' };
/** OutboxRecord 去掉 seq/at(由 appendOutbox 填) */
export type OutboxPayload =
| Omit<Extract<OutboxRecord, { type: 'started' }>, 'seq' | 'at'>
| Omit<Extract<OutboxRecord, { type: 'phase' }>, 'seq' | 'at'>
| Omit<Extract<OutboxRecord, { type: 'failed' }>, 'seq' | 'at'>
| Omit<Extract<OutboxRecord, { type: 'result' }>, 'seq' | 'at'>
| Omit<Extract<OutboxRecord, { type: 'done' }>, 'seq' | 'at'>;
/** 追加一条 outbox 记录(worker 侧调用)。seq = 现有行数+1(worker 单线程,无并发写)。返回写入的完整记录。 */
export function appendOutbox(runId: string, payload: OutboxPayload): OutboxRecord {
mkdirSync(runDir(runId), { recursive: true });
const prev = readOutboxAll(runId);
const seq = prev.length + 1;
const rec = { seq, at: new Date().toISOString(), ...payload } as OutboxRecord;
appendFileSync(outboxPath(runId), JSON.stringify(rec) + '\n');
return rec;
}
/** 读全部 outbox 记录(坏行跳过,容忍写一半的尾行)。 */
export function readOutboxAll(runId: string): OutboxRecord[] {
const p = outboxPath(runId);
if (!existsSync(p)) return [];
const out: OutboxRecord[] = [];
for (const line of readFileSync(p, 'utf8').split('\n')) {
const s = line.trim();
if (!s) continue;
try { out.push(JSON.parse(s) as OutboxRecord); } catch { /* 半截尾行:跳过,下轮再读 */ }
}
return out;
}
/** 读 seq > lastSeq 的新记录(daemon ingest 侧调用,幂等去重靠 lastSeq)。 */
export function readOutboxSince(runId: string, lastSeq: number): OutboxRecord[] {
return readOutboxAll(runId).filter((r) => r.seq > lastSeq);
}
// ───────────────────────── worker → daemon:心跳 + 存活判定 ──────────────────
/** worker 刷心跳:touch runs/<runId>/heartbeat(文件不存在则建)。daemon 用其 mtime 判活。 */
export function touchHeartbeat(runId: string): void {
mkdirSync(runDir(runId), { recursive: true });
const p = heartbeatPath(runId);
if (!existsSync(p)) { closeSync(openSync(p, 'w')); return; }
const now = new Date();
utimesSync(p, now, now);
}
/** 心跳文件的「年龄」毫秒(now - mtime);无心跳文件返回 null。 */
export function heartbeatAgeMs(runId: string, now = Date.now()): number | null {
try { return now - statSync(heartbeatPath(runId)).mtimeMs; }
catch { return null; }
}
/** pid 是否存活(signal 0 探测;EPERM 视为存活——进程在但无权) */
export function pidAlive(pid: number): boolean {
if (!pid || pid <= 0) return false;
try { process.kill(pid, 0); return true; }
catch (e) { return (e as NodeJS.ErrnoException).code === 'EPERM'; }
}
export const HEARTBEAT_INTERVAL_MS = 10_000; // worker 刷心跳间隔
export const HEARTBEAT_GRACE_MS = 60_000; // 心跳超过此年龄视为可疑(须 > interval,容忍 GC/慢盘)
export const BOOT_GRACE_MS = 30_000; // started 后此窗口内额外认 pid 存活(覆盖启动到首次心跳空窗)
/**
* worker 是否存活。主信号=心跳新鲜;boot 窗口内额外接受 pid 存活(覆盖启动空窗);
* 窗口后仅认心跳——规避 pid 复用误判(被复用的 pid 不会更新本 run 的心跳)。
*/
export function isWorkerAlive(args: { pid: number | null; heartbeatAgeMs: number | null; startedAgeMs: number }): boolean {
const { pid, heartbeatAgeMs: hb, startedAgeMs } = args;
if (hb !== null && hb < HEARTBEAT_GRACE_MS) return true; // 心跳新鲜 → 活
if (startedAgeMs < BOOT_GRACE_MS && pid !== null && pidAlive(pid)) return true; // 启动空窗 → pid 兜底
return false;
}