Files
maestro/test/orchestrator.test.ts
T
wangjia a00091329d merge: maestro/tsk_N5wO3Armums2 [重试与超时策略细化(可配 + 退避)]
# Conflicts:
#	src/api/server.ts
#	test/store.test.ts
2026-06-13 10:20:20 +08:00

477 lines
19 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { test } from 'node:test';
import assert from 'node:assert/strict';
import { Store } from '../src/store/index.js';
import { createOrchestrator, DEFAULT_MAX_RETRIES, computeBackoffMs, type OrchestratorDeps } from '../src/daemon/orchestrator.js';
import type { RunnerResult } from '../src/executor/runner.js';
import type { ReviewResult } from '../src/executor/reviewer.js';
import type { Autonomy } from '../src/model/types.js';
const noopLog = { info: (): void => undefined, error: (): void => undefined };
function deferred<T>(): { promise: Promise<T>; resolve: (v: T) => void } {
let resolve!: (v: T) => void;
const promise = new Promise<T>((r) => { resolve = r; });
return { promise, resolve };
}
const okRun: RunnerResult = { ok: true, transcriptRef: '/tmp/fake.jsonl', sessionId: 'sess-mock-1', finalText: '执行自述:改了 README' };
const okReview: ReviewResult = {
summary: '## 做了什么\nmock 复审通过',
verdict: 'approve',
transcriptRef: '/tmp/fake-review.jsonl',
sessionId: 'sess-review-1',
};
const okSecurity: ReviewResult = {
summary: '## 安全审计\nmock 审计通过',
verdict: 'approve',
transcriptRef: '/tmp/fake-security.jsonl',
sessionId: 'sess-security-1',
};
/** 全 mock 依赖(不真起 CC、不动 git):可按用例覆盖 */
function mockDeps(overrides: Partial<OrchestratorDeps> = {}): OrchestratorDeps {
return {
createWorktree: async (_repo, taskId) => ({ dir: `/tmp/fake-wt/${taskId}`, branch: `maestro/${taskId}` }),
worktreeDiff: async () => ({ diffSummary: ' README.md | 1 +', commits: ['abc1234 hello maestro'] }),
verify: async () => ({ ok: true, exitCode: 0, logRef: null }),
runner: async () => okRun,
reviewCode: async () => okReview,
reviewSecurity: async () => okSecurity,
nowMs: Date.now,
...overrides,
};
}
/** 可前进的测试时钟(用于退避相关测试) */
function makeClock(initialMs = 0): { nowMs: () => number; advance: (ms: number) => void } {
let t = initialMs;
return { nowMs: () => t, advance: (ms) => { t += ms; } };
}
function setup(autonomy: Autonomy, concurrency = 1): { store: Store; projectId: string } {
const store = new Store(':memory:');
const p = store.createProject({
name: 'orch', repoPath: '/tmp/orch-repo-' + Math.random(), autonomy, concurrency,
});
return { store, projectId: p.id };
}
async function settle(): Promise<void> {
await new Promise((r) => setImmediate(r));
}
test('autonomy=manual:编排器不领取任何任务', async () => {
const { store, projectId } = setup('manual');
const t = store.createTask({ projectId, title: 'easy task', complexity: 'easy' });
let calls = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { calls++; return okRun; },
}));
orch.tick();
await orch.drain();
assert.equal(calls, 0);
assert.equal(store.getTask(t.id)!.status, 'ready');
store.close();
});
test('autonomy=auto-easy:只领 easymedium ready 不动', async () => {
const { store, projectId } = setup('auto-easy', 5);
const easy = store.createTask({ projectId, title: 'small', complexity: 'easy' });
const medium = store.createTask({ projectId, title: 'mid', complexity: 'medium' });
store.setSpec(medium.id, '方案');
store.transition(medium.id, 'spec_review');
store.decide(medium.id, 'accept', 'user'); // medium → ready
assert.equal(store.getTask(medium.id)!.status, 'ready');
const ran: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { ran.push(task.id); return okRun; },
}));
orch.tick();
await orch.drain();
assert.deepEqual(ran, [easy.id]);
assert.equal(store.getTask(easy.id)!.status, 'exec_review');
assert.equal(store.getTask(medium.id)!.status, 'ready'); // 不自动跑
store.close();
});
test('autonomy=auto-approved:领全部 ready(含 medium),依赖未满足/非叶子不领', async () => {
const { store, projectId } = setup('auto-approved', 5);
const medium = store.createTask({ projectId, title: 'mid', complexity: 'medium' });
store.setSpec(medium.id, '方案');
store.transition(medium.id, 'spec_review');
store.decide(medium.id, 'accept', 'user');
// 依赖未 done 的任务:建在 medium 上 → blocked,不可领
const dep = store.createTask({ projectId, title: 'after-mid', complexity: 'easy', deps: [medium.id] });
assert.equal(store.getTask(dep.id)!.status, 'blocked');
const ran: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { ran.push(task.id); return okRun; },
}));
orch.tick();
await orch.drain();
assert.deepEqual(ran, [medium.id]);
assert.equal(store.getTask(medium.id)!.status, 'exec_review');
assert.equal(store.getTask(dep.id)!.status, 'blocked');
store.close();
});
test('concurrency=1:同项目同轮只领 1 个,跑完下一轮再领', async () => {
const { store, projectId } = setup('auto-approved', 1);
const t1 = store.createTask({ projectId, title: 'a', complexity: 'easy' });
const t2 = store.createTask({ projectId, title: 'b', complexity: 'easy' });
const gate = deferred<void>();
const started: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { started.push(task.id); await gate.promise; return okRun; },
}));
orch.tick();
await settle();
assert.deepEqual(started, [t1.id], '并发=1 只应启动第一个任务');
assert.equal(store.getTask(t1.id)!.status, 'executing');
assert.equal(store.getTask(t2.id)!.status, 'ready');
assert.equal(orch.inflight.size, 1);
orch.tick(); // 在途占满 → 本轮不领
await settle();
assert.deepEqual(started, [t1.id]);
gate.resolve();
await orch.drain();
assert.equal(store.getTask(t1.id)!.status, 'exec_review');
orch.tick(); // 槽位释放 → 领第二个
gate.resolve();
await orch.drain();
assert.deepEqual(started, [t1.id, t2.id]);
assert.equal(store.getTask(t2.id)!.status, 'exec_review');
store.close();
});
test('成功路径:状态流转 + setResult(四字段) + executor/reviewer/security 三 run succeeded', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'tweak', complexity: 'easy' });
store.setOperations(t.id, '在 README.md 追加一行');
const seen: string[] = [];
store.subscribe((e) => { if (e.type === 'status.changed' && e.taskId === t.id) seen.push(String(e.payload.to)); });
let codeReportSeen: string | null = null;
let secReportSeen: string | null = null;
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async (_task, _project, _wt, _runId, executorReport) => {
codeReportSeen = executorReport;
return okReview;
},
reviewSecurity: async (_task, _project, _wt, _runId, executorReport) => {
secReportSeen = executorReport;
return okSecurity;
},
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.deepEqual(seen, ['queued', 'executing', 'exec_review']);
assert.deepEqual(done.result, {
branch: `maestro/${t.id}`,
worktree: `/tmp/fake-wt/${t.id}`,
diffSummary: ' README.md | 1 +',
commits: ['abc1234 hello maestro'],
prUrl: null,
summary: '## 做了什么\nmock 复审通过',
verdict: 'approve',
securitySummary: '## 安全审计\nmock 审计通过',
securityVerdict: 'approve',
mergeTaskId: null,
});
assert.equal(codeReportSeen, '执行自述:改了 README'); // runner finalText 传给两个复审作执行者自述
assert.equal(secReportSeen, '执行自述:改了 README');
const runs = store.listRuns(t.id);
assert.equal(runs.length, 3);
const executor = runs.find((r) => r.kind === 'executor')!;
assert.equal(executor.status, 'succeeded');
assert.equal(executor.branch, `maestro/${t.id}`);
assert.equal(executor.transcriptRef, '/tmp/fake.jsonl');
assert.equal(executor.claudeSessionId, 'sess-mock-1');
const reviewer = runs.find((r) => r.kind === 'reviewer')!;
assert.equal(reviewer.status, 'succeeded');
assert.equal(reviewer.transcriptRef, '/tmp/fake-review.jsonl');
assert.equal(reviewer.claudeSessionId, 'sess-review-1');
const security = runs.find((r) => r.kind === 'security')!;
assert.equal(security.status, 'succeeded');
assert.equal(security.transcriptRef, '/tmp/fake-security.jsonl');
assert.equal(security.claudeSessionId, 'sess-security-1');
store.close();
});
test('任一 verdict=reject 也照常落进 result(最终裁决仍归用户)', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'risky', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async () => ({ ...okReview, summary: '发现问题', verdict: 'reject' as const }),
reviewSecurity: async () => ({ ...okSecurity, summary: '发现密钥泄露', verdict: 'reject' as const }),
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.equal(done.result!.verdict, 'reject');
assert.equal(done.result!.summary, '发现问题');
assert.equal(done.result!.securityVerdict, 'reject');
assert.equal(done.result!.securitySummary, '发现密钥泄露');
store.close();
});
test('code review 失败不挡任务:summary 记失败原因、verdict=null,安全审计照常跑', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'review-broken', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async () => { throw new Error('复审 CC 崩了'); },
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review'); // 不挡结果闸
assert.equal(done.result!.verdict, null);
assert.equal(done.result!.summary, '自动复审失败:复审 CC 崩了');
assert.equal(done.result!.securityVerdict, 'approve'); // 另一个复审不受影响
assert.equal(done.result!.securitySummary, '## 安全审计\nmock 审计通过');
const runs = store.listRuns(t.id);
assert.equal(runs.find((r) => r.kind === 'executor')!.status, 'succeeded');
const reviewer = runs.find((r) => r.kind === 'reviewer')!;
assert.equal(reviewer.status, 'failed');
assert.match(reviewer.error ?? '', /复审 CC 崩了/);
assert.equal(runs.find((r) => r.kind === 'security')!.status, 'succeeded');
store.close();
});
test('安全审计失败不挡任务:securitySummary 记失败原因、securityVerdict=nullcode review 不受影响', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'security-broken', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewSecurity: async () => { throw new Error('审计 CC 崩了'); },
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.equal(done.result!.verdict, 'approve');
assert.equal(done.result!.summary, '## 做了什么\nmock 复审通过');
assert.equal(done.result!.securityVerdict, null);
assert.equal(done.result!.securitySummary, '自动复审失败:审计 CC 崩了');
const runs = store.listRuns(t.id);
assert.equal(runs.find((r) => r.kind === 'executor')!.status, 'succeeded');
assert.equal(runs.find((r) => r.kind === 'reviewer')!.status, 'succeeded');
const security = runs.find((r) => r.kind === 'security')!;
assert.equal(security.status, 'failed');
assert.match(security.error ?? '', /审计 CC 崩了/);
store.close();
});
test('verify 不过:按失败处理(run failed + 重新入队)', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'v', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
verify: async () => ({ ok: false, exitCode: 1, logRef: '/tmp/v.log', error: 'verify 失败(exit 1' }),
}));
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued'); // 第一次失败 → 重新入队
const runs = store.listRuns(t.id);
assert.equal(runs.length, 1);
assert.equal(runs[0].status, 'failed');
assert.match(runs[0].error ?? '', /verify 失败/);
store.close();
});
test(`失败重试:重试 ${DEFAULT_MAX_RETRIES} 次后 → needs_attention(共 ${DEFAULT_MAX_RETRIES + 1} 次失败 run`, async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'flaky', complexity: 'easy' });
let attempts = 0;
// 使用快进时钟跳过退避冷却(每次 tick 前将时钟推进 10min
const clock = makeClock();
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: `boom #${attempts}` }; },
nowMs: clock.nowMs,
}));
for (let i = 1; i <= DEFAULT_MAX_RETRIES; i++) {
clock.advance(10 * 60_000); // 跳过退避冷却
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued', `${i} 次失败后应重新入队`);
}
clock.advance(10 * 60_000); // 跳过最后一次退避
orch.tick(); // 最后一次重试也失败
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'needs_attention');
assert.equal(attempts, DEFAULT_MAX_RETRIES + 1);
const failed = store.listRuns(t.id).filter((r) => r.status === 'failed');
assert.equal(failed.length, DEFAULT_MAX_RETRIES + 1);
clock.advance(10 * 60_000);
orch.tick(); // needs_attention 不会再被领取
await orch.drain();
assert.equal(attempts, DEFAULT_MAX_RETRIES + 1);
store.close();
});
test('项目级 maxRetries=1:只重试 1 次就 → needs_attention', async () => {
const store = new Store(':memory:');
const p = store.createProject({
name: 'orch', repoPath: '/tmp/orch-repo-retries-' + Math.random(),
autonomy: 'auto-easy', maxRetries: 1,
});
const projectId = p.id;
const t = store.createTask({ projectId, title: 'fail1', complexity: 'easy' });
let attempts = 0;
const clock = makeClock();
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: 'boom' }; },
nowMs: clock.nowMs,
}));
// 第 1 次执行失败 → 重新入队(还有 1 次重试机会)
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued');
// 第 2 次失败 → 超过 maxRetries=1 → needs_attention(快进时钟跳过退避)
clock.advance(10 * 60_000);
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'needs_attention');
assert.equal(attempts, 2);
store.close();
});
test('项目级 maxRetries=0:首次失败直接 needs_attention(不重试)', async () => {
const store = new Store(':memory:');
const p = store.createProject({
name: 'orch', repoPath: '/tmp/orch-repo-zero-' + Math.random(),
autonomy: 'auto-easy', maxRetries: 0,
});
const t = store.createTask({ projectId: p.id, title: 'fail0', complexity: 'easy' });
let attempts = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: 'boom' }; },
}));
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'needs_attention');
assert.equal(attempts, 1);
store.close();
});
test('computeBackoffMs:指数退避公式,30s 基数,上限 10min', () => {
assert.equal(computeBackoffMs(1), 30_000);
assert.equal(computeBackoffMs(2), 60_000);
assert.equal(computeBackoffMs(3), 120_000);
assert.equal(computeBackoffMs(4), 240_000);
assert.equal(computeBackoffMs(10), 10 * 60_000); // capped at 10min
assert.equal(computeBackoffMs(100), 10 * 60_000);
});
test('退避冷却:首次失败重入队后,退避期内不被领取;退避过期后正常领取', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'backoff', complexity: 'easy' });
let attempts = 0;
// 使用可控时钟:初始时间 0,退避 30sattempt=1 → BACKOFF_BASE=30000ms
const clock = makeClock(0);
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: 'boom' }; },
nowMs: clock.nowMs,
}));
// 第 1 次失败(t=0)→ 重入队 + 退避(backoffUntil = 0 + 30000 = 30000ms
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued');
assert.equal(attempts, 1);
// t=29999:退避期内,不被领取
clock.advance(29_999);
orch.tick();
await orch.drain();
assert.equal(attempts, 1, '退避期内不应再次执行');
assert.equal(store.getTask(t.id)!.status, 'queued');
// t=30001:退避过期,任务应被重新领取
clock.advance(2);
orch.tick();
await orch.drain();
assert.equal(attempts, 2, '退避过期后应再次执行');
store.close();
});
test('requeueTaskneeds_attention → queued,重置重试基线,再次允许重试', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'requeue', complexity: 'easy' });
let attempts = 0;
const clock = makeClock();
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: 'boom' }; },
nowMs: clock.nowMs,
}));
// 耗尽默认重试次数 → needs_attention(每次需快进时钟跳过退避)
for (let i = 0; i <= DEFAULT_MAX_RETRIES; i++) {
clock.advance(10 * 60_000);
orch.tick();
await orch.drain();
}
assert.equal(store.getTask(t.id)!.status, 'needs_attention');
assert.equal(attempts, DEFAULT_MAX_RETRIES + 1);
// 手动重投:重置基线,转 queued
const requeued = store.requeueTask(t.id);
assert.equal(requeued.status, 'queued');
assert.equal(requeued.retryBaseline, DEFAULT_MAX_RETRIES + 1);
// 重投后编排器应能再次执行(快进时钟确保无退避阻拦)
clock.advance(10 * 60_000);
orch.tick();
await orch.drain();
assert.equal(attempts, DEFAULT_MAX_RETRIES + 2, '重投后应再次执行');
// 第 1 次净失败后应重入队(基线已重置,净失败=1 < maxRetries=2
assert.equal(store.getTask(t.id)!.status, 'queued');
store.close();
});
test('project paused:不领取', async () => {
const { store, projectId } = setup('auto-approved');
const t = store.createTask({ projectId, title: 'x', complexity: 'easy' });
store.patchProject(projectId, { status: 'paused' });
let calls = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { calls++; return okRun; },
}));
orch.tick();
await orch.drain();
assert.equal(calls, 0);
assert.equal(store.getTask(t.id)!.status, 'ready');
store.close();
});