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
+3
View File
@@ -29,6 +29,9 @@ export function openDb(file: string): Database.Database {
ensureColumn(db, 'projects', 'max_retries', 'max_retries INTEGER NOT NULL DEFAULT 2'); // 失败后最大自动重试次数
ensureColumn(db, 'projects', 'timeout_ms', 'timeout_ms INTEGER NOT NULL DEFAULT 1800000'); // 单次执行超时毫秒
ensureColumn(db, 'tasks', 'retry_baseline', 'retry_baseline INTEGER NOT NULL DEFAULT 0'); // 手动重投时的失败基线
ensureColumn(db, 'tasks', 'next_eligible_at', 'next_eligible_at TEXT'); // 持久化退避(多进程执行)
ensureColumn(db, 'runs', 'worker_pid', 'worker_pid INTEGER'); // worker 进程 pid
ensureColumn(db, 'runs', 'last_seq', 'last_seq INTEGER NOT NULL DEFAULT 0'); // outbox ingest 游标
const schema = readFileSync(join(HERE, 'schema.sql'), 'utf8');
db.exec(schema);
return db;
+4
View File
@@ -16,6 +16,7 @@ export interface TaskRow {
title: string; complexity: string; status: string; priority: number;
deps: string; plan: string | null; spec: string | null; operations: string | null;
result: string | null; assignee: string | null; retry_baseline: number;
next_eligible_at: string | null;
created_at: string; updated_at: string;
source_ref: string | null;
}
@@ -27,6 +28,7 @@ export interface RunRow {
id: string; task_id: string; kind: string; worktree: string | null; branch: string | null;
status: string; started_at: string; ended_at: string | null;
transcript_ref: string | null; claude_session_id: string | null; error: string | null;
worker_pid: number | null; last_seq: number;
}
export interface EventRow {
id: string; project_id: string; task_id: string | null; type: string; payload: string; at: string;
@@ -68,6 +70,7 @@ export function rowToTask(r: TaskRow, approvals: ApprovalRecord[] = []): Task {
plan: r.plan, spec: r.spec, operations: r.operations,
approvals, result: r.result ? parseResult(r.result) : null,
assignee: r.assignee as Task['assignee'], retryBaseline: r.retry_baseline ?? 0,
nextEligibleAt: r.next_eligible_at ?? null,
createdAt: r.created_at, updatedAt: r.updated_at,
};
}
@@ -85,6 +88,7 @@ export function rowToRun(r: RunRow): Run {
worktree: r.worktree, branch: r.branch, status: r.status as RunStatus,
startedAt: r.started_at, endedAt: r.ended_at,
transcriptRef: r.transcript_ref, claudeSessionId: r.claude_session_id, error: r.error,
workerPid: r.worker_pid ?? null, lastSeq: r.last_seq ?? 0,
};
}
+4 -1
View File
@@ -36,6 +36,7 @@ CREATE TABLE IF NOT EXISTS tasks (
result TEXT, -- JSON TaskResult
assignee TEXT, -- agent | human
retry_baseline INTEGER NOT NULL DEFAULT 0, -- 上次手动重投时已有的失败 run 数(重置重试计数用)
next_eligible_at TEXT, -- 持久化退避:早于此时间不被领取(null=即刻可领)
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL,
source_ref TEXT -- 旧 todo 来源标识(todo:17 / todo:17/1A),项目内唯一
@@ -66,7 +67,9 @@ CREATE TABLE IF NOT EXISTS runs (
ended_at TEXT,
transcript_ref TEXT,
claude_session_id TEXT,
error TEXT
error TEXT,
worker_pid INTEGER, -- 多进程执行:worker 进程 piddaemon 写,判活用)
last_seq INTEGER NOT NULL DEFAULT 0 -- 已 ingest 的 outbox 最大 seq(幂等游标)
);
CREATE INDEX IF NOT EXISTS idx_runs_task ON runs(task_id, started_at);
+92 -10
View File
@@ -231,8 +231,8 @@ export class Store {
id: id('tsk'), project_id: input.projectId, parent_id: input.parentId ?? null, depth,
title: input.title, complexity: input.complexity, status, priority: input.priority ?? 1,
deps: JSON.stringify(input.deps ?? []), plan: null, spec: null, operations: null,
result: null, assignee: null, retry_baseline: 0, created_at: now(), updated_at: now(),
source_ref: null,
result: null, assignee: null, retry_baseline: 0, next_eligible_at: null,
created_at: now(), updated_at: now(), source_ref: null,
};
this.db.prepare(
`INSERT INTO tasks (id,project_id,parent_id,depth,title,complexity,status,priority,deps,plan,spec,operations,result,assignee,retry_baseline,created_at,updated_at,source_ref)
@@ -562,15 +562,86 @@ export class Store {
* 卡在 executing 的任务转 failed→queued 等编排器重新领取。
* 注:中断产生的 failed run 会计入该任务的失败次数(多次中断+真失败可能提前转 needs_attention,可接受)。
*/
reconcileInterrupted(): { runs: number; tasks: number } {
const runs = this.db.prepare(`SELECT * FROM runs WHERE status = 'started'`).all() as RunRow[];
for (const r of runs) this.finishRun(r.id, 'failed', { error: 'daemon 重启,执行中断' });
const rows = this.db.prepare(`SELECT * FROM tasks WHERE status = 'executing'`).all() as TaskRow[];
for (const t of rows) {
this.transition(t.id, 'failed', { auto: 'daemon-restart' });
this.transition(t.id, 'queued', { auto: 'daemon-restart' });
/**
* 失败/重试策略的唯一落点(ingest 的 failed 记录、reaper 判死、reconcile 判死 都调它)。
* 把(可选)执行 run 收尾为 failed,按「净失败次数 vs project.maxRetries」决定:
* 未到上限 → queued + 持久化退避(next_eligible_at = now + 指数退避)
* 到上限 → needs_attention(清退避)
* netFailed 相对 retry_baseline(手动重投时设的基线)。退避:min(30s·2^(n-1), 10min)。
*/
failTaskAttempt(taskId: string, runId: string | null, error: string): Task {
const row = this.getTaskRow(taskId);
if (!row) throw new StoreError(`任务不存在: ${taskId}`);
const project = this.getProject(row.project_id);
if (!project) throw new StoreError(`项目不存在: ${row.project_id}`);
// 收尾执行 run:给了 runId 且仍 started → 标 failed;没给 → 补记一条 failed(保证重试计数不漏)
if (runId) {
const rr = this.db.prepare(`SELECT status FROM runs WHERE id = ?`).get(runId) as { status: string } | undefined;
if (rr && rr.status === 'started') this.finishRun(runId, 'failed', { error });
} else {
const r = this.startRun(taskId, 'executor');
this.finishRun(r.id, 'failed', { error });
}
return { runs: runs.length, tasks: rows.length };
if ((this.getTaskRow(taskId)!.status as TaskStatus) !== 'failed') {
this.transition(taskId, 'failed', { by: 'failTaskAttempt', error });
}
const allFailed = (this.db.prepare(
`SELECT COUNT(*) AS n FROM runs WHERE task_id = ? AND kind = 'executor' AND status = 'failed'`,
).get(taskId) as { n: number }).n;
const netFailed = Math.max(0, allFailed - row.retry_baseline);
const priorNetFailed = Math.max(0, netFailed - 1); // 不含本次
if (priorNetFailed < project.maxRetries) {
const attempt = priorNetFailed + 1;
const backoffMs = Math.min(30_000 * 2 ** (attempt - 1), 10 * 60_000);
this.setNextEligibleAt(taskId, new Date(Date.now() + backoffMs).toISOString());
this.transition(taskId, 'queued', { by: 'failTaskAttempt', retry: attempt, backoffMs });
} else {
this.setNextEligibleAt(taskId, null);
this.transition(taskId, 'needs_attention', { by: 'failTaskAttempt', allFailed, netFailed });
}
return this.getTask(taskId)!;
}
/** 持久化退避:早于 next_eligible_at 不被领取(null=即刻可领)。 */
setNextEligibleAt(taskId: string, iso: string | null): void {
this.db.prepare(`UPDATE tasks SET next_eligible_at = ?, updated_at = ? WHERE id = ?`).run(iso, now(), taskId);
}
/** 项目内 executing 任务数(多进程执行的并发闸:每个 executing 对应一个活 worker)。 */
countExecuting(projectId: string): number {
return (this.db.prepare(
`SELECT COUNT(*) AS n FROM tasks WHERE project_id = ? AND status = 'executing'`,
).get(projectId) as { n: number }).n;
}
/** 所有 executing 任务 + 其最近一条 executor runreaper / reconcile 判活用)。run 可能为 null(异常)。 */
executingWithLatestExecutorRun(): Array<{ task: Task; run: Run | null }> {
const rows = this.db.prepare(`SELECT * FROM tasks WHERE status = 'executing'`).all() as TaskRow[];
return rows.map((t) => {
const rr = this.db.prepare(
`SELECT * FROM runs WHERE task_id = ? AND kind = 'executor' ORDER BY started_at DESC LIMIT 1`,
).get(t.id) as RunRow | undefined;
return { task: rowToTask(t), run: rr ? rowToRun(rr) : null };
});
}
/**
* 启动中断恢复(多进程执行):逐个 executing 任务按注入的 isAlive 判活——
* 活 → re-adopt(保留 executingdaemon 续 ingest 其 outbox);
* 死 → failTaskAttempt(收尾 + 重试/needs_attention)。
* isAlive 由 daemon 注入(protocol.isWorkerAlive,基于 worker_pid + 心跳);默认保守判死(无判活信息时回收)。
*/
reconcileInterrupted(isAlive: (run: Run | null) => boolean = () => false): { readopted: number; reclaimed: number } {
let readopted = 0, reclaimed = 0;
for (const { task, run } of this.executingWithLatestExecutorRun()) {
if (run && isAlive(run)) { readopted++; continue; }
this.failTaskAttempt(task.id, run?.id ?? null, 'daemon 重启时发现 worker 已退出');
reclaimed++;
}
return { readopted, reclaimed };
}
/**
@@ -687,6 +758,7 @@ export class Store {
const rr: RunRow = {
id: id('run'), task_id: taskId, kind, worktree: fields.worktree ?? null, branch: fields.branch ?? null,
status: 'started', started_at: now(), ended_at: null, transcript_ref: null, claude_session_id: null, error: null,
worker_pid: null, last_seq: 0,
};
this.db.prepare(
`INSERT INTO runs (id,task_id,kind,worktree,branch,status,started_at,ended_at,transcript_ref,claude_session_id,error)
@@ -707,6 +779,16 @@ export class Store {
return rowToRun(this.db.prepare(`SELECT * FROM runs WHERE id = ?`).get(runId) as RunRow);
}
/** 记录执行该 run 的 worker 进程 pid(多进程执行,daemon spawn 后写;判活用)。 */
setWorkerPid(runId: string, pid: number): void {
this.db.prepare(`UPDATE runs SET worker_pid = ? WHERE id = ?`).run(pid, runId);
}
/** 更新已 ingest 的 outbox 最大 seq(幂等游标;daemon ingest 后写,重启续读)。 */
setLastSeq(runId: string, seq: number): void {
this.db.prepare(`UPDATE runs SET last_seq = ? WHERE id = ?`).run(seq, runId);
}
/** 所有进行中的 runstatus='started'),联 tasks 取任务标题与项目。 */
activeRuns(): ActiveRun[] {
const rows = this.db.prepare(