/** * Unit Tests — 节点失败时原因留不留得下来（27-P3B W12，「第 12 处四环漏一环」） * * 修之前：advanceOne 的 executor catch 把 err.message 写进 outcome.output， * 而 outcome.output **一个消费方都没有** —— 不落库（graph_run_nodes 没有 output 列）、 * 不进 SSE、也不写 timeline，fail_reason 还是 null。真跑第一个 work 节点时 * AgentManager 抛的「Agent not found」就是这么消失的，操作者看到「节点 fail、无任何原因」。 * * 这里锁的是修完之后的四环： * 产生（catch / executor 返回值）→ 传递（timelineNote / node.output 事件 data） * → 落库（timeline.ndjson + stream.log，append-only）→ 读回（本用例就是按读回的那条路读的： * 前端走 GET /api/graph-runs/:id/stream?lastEventId=0 全量重放同一份 stream.log） * * 隔离手法同 graph-engine.test.js：先设 GRAPH_DB_PATH / GRAPH_RUNS_DIR 再 require， * 仓库里的 data/graph.sqlite 与 data/companies/ 全程不被碰到；executor 全是注入的假货， * 不起任何 Claude CLI。 */ import { describe, it, expect, beforeAll, afterAll, beforeEach } from 'vitest'; import fs from 'fs'; import os from 'os'; import path from 'path'; describe('graph-engine 节点失败留痕（W12）', () => { let wdb; let engine; let sse; let runsDir; const TEST_DB_PATH = '/tmp/test-graph-engine-node-error.sqlite'; beforeAll(() => { for (const f of [TEST_DB_PATH, `${TEST_DB_PATH}-shm`, `${TEST_DB_PATH}-wal`]) { try { fs.unlinkSync(f); } catch {} } process.env.GRAPH_DB_PATH = TEST_DB_PATH; runsDir = fs.mkdtempSync(path.join(os.tmpdir(), 'graph-node-error-runs-')); process.env.GRAPH_RUNS_DIR = runsDir; for (const m of [ '../../services/graph-db', '../../services/graph-service', '../../services/org-graph-service', '../../services/work-graph-db', '../../services/graph-run-sse', '../../services/graph-engine', ]) { delete require.cache[require.resolve(m)]; } wdb = require('../../services/work-graph-db'); sse = require('../../services/graph-run-sse'); engine = require('../../services/graph-engine'); }); afterAll(() => { for (const f of [TEST_DB_PATH, `${TEST_DB_PATH}-shm`, `${TEST_DB_PATH}-wal`]) { try { fs.unlinkSync(f); } catch {} } delete process.env.GRAPH_DB_PATH; try { fs.rmSync(runsDir, { recursive: true, force: true }); } catch {} delete process.env.GRAPH_RUNS_DIR; }); beforeEach(() => { const db = wdb.getDb(); db.pragma('foreign_keys = OFF'); db.exec(` DELETE FROM graph_run_loop_records; DELETE FROM graph_run_nodes; DELETE FROM graph_runs; DELETE FROM work_graph_edges; DELETE FROM work_graph_nodes; DELETE FROM work_graphs; DELETE FROM system_agents; `); db.pragma('foreign_keys = ON'); engine.setAgentExecutor(null); }); let seq = 0; const nextCompany = () => `w12-co-${++seq}`; function ensureAgents(ids) { const db = wdb.getDb(); const now = new Date().toISOString(); const ins = db.prepare(` INSERT OR IGNORE INTO system_agents (id, name, scope, owner_company_id, source_kind, legacy_path, status, created_at, updated_at) VALUES (?, ?, 'public', NULL, 'global_agent', ?, 'active', ?, ?) `); for (const id of new Set(ids)) ins.run(id, id, `system/agents/${id}`, now, now); } function buildRun(nodes, edges = []) { const companyId = nextCompany(); ensureAgents(nodes); const g = wdb.createWorkGraph({ companyId, nodes, edges }); expect(g.errors || []).toEqual([]); const run = wdb.createRun({ graphKind: 'work', graphId: g.id, companyId, executionMode: 'manual' }); engine.markRunReady(run.id); return run.id; } /** 读回轨 1：审计轨 timeline.ndjson（append-only） */ const timeline = (runId) => fs.readFileSync(sse.timelinePath(runId), 'utf8') .split('\n').filter(Boolean).map((l) => JSON.parse(l)); /** 读回轨 2：stream.log —— 前端 lastEventId=0 全量重放读的就是这一份 */ const streamEvents = (runId) => fs.readFileSync(path.join(runsDir, runId, 'stream.log'), 'utf8') .split('\n\n').filter(Boolean) .map((block) => { const ev = /^event: (.+)$/m.exec(block); const data = /^data: (.+)$/m.exec(block); return { event: ev && ev[1], data: JSON.parse(data[1]) }; }); const nodeByName = (runId, name) => wdb.getRunState(runId).nodes.find((n) => n.name === name); it('executor 抛异常：节点 fail，且 timeline 里查得到原因（不是空的）', async () => { const runId = buildRun(['prd-planning']); engine.setAgentExecutor(async () => { // 真跑第一个 work 节点时 AgentManager.getAgentConfig 抛的就是这一句 throw new Error('Agent not found: prd-planning'); }); engine.startRun(runId); await engine.nextStep(runId); const node = nodeByName(runId, 'prd-planning'); // 没有 fail 边的死路：fail 之后就地 blocked —— 真跑第一个 work 节点时看到的就是这个 expect(node.status).toBe('blocked'); // W12 补完（本轮改）：以前这里断言 `failReason` 恒 null，理由写的是「引擎不发明 fail_reason」。 // 那条理由只对**归因语义**成立（不许引擎替 validator 说「这是 impl_bug」）， // 结果却让概览 tab 上的失败节点永远显示「原因 —」。现在引擎自判的失败署自己的名， // 取值来自与归因域**不重叠**的 ENGINE_FAIL_REASONS，选边行为与 null 完全一致（都走兜底边）。 // 详见 graph-engine-fail-reason.test.js。 expect(node.failReason).toBe('node_error'); const errs = timeline(runId).filter((e) => e.kind === 'node.error'); expect(errs).toHaveLength(1); expect(errs[0].summary).toContain('Agent not found: prd-planning'); expect(errs[0].payload.node_run_id).toBe(node.id); expect(errs[0].payload.attempt).toBe(0); expect(errs[0].payload.node_name).toBe('prd-planning'); expect(errs[0].payload.system_agent_id).toBe('prd-planning'); expect(errs[0].payload.error).toBe('Agent not found: prd-planning'); expect(errs[0].payload.error_type).toBe('Error'); expect(errs[0].payload.error_stack).toContain('Error: Agent not found'); // ref = 节点 run id：前端 NodeRunLog 按 node_run_id 归组，这条落在失败节点自己的卡片里 expect(errs[0].ref).toBe(node.id); // 同一条事件也在实时轨上（emitEvent 保证两轨不分家） const onStream = streamEvents(runId).filter((e) => e.event === 'node.error'); expect(onStream).toHaveLength(1); expect(onStream[0].data.error).toBe('Agent not found: prd-planning'); }); it('抛异常时 node.output 事件带 output_excerpt，错误原因同样能读回', async () => { const runId = buildRun(['a']); engine.setAgentExecutor(async () => { throw new Error('boom: 配置缺失'); }); engine.startRun(runId); await engine.nextStep(runId); const out = streamEvents(runId).filter((e) => e.event === 'node.output'); expect(out).toHaveLength(1); expect(out[0].data.status).toBe('fail'); // 同上：W12 补完后 executor 抛异常这条路径署名 node_error（不再是 null） expect(out[0].data.fail_reason).toBe('node_error'); expect(out[0].data.output_excerpt).toBe('boom: 配置缺失'); }); it('executor 正常返回的 output 也进 output_excerpt，超长按定长截断', async () => { const runId = buildRun(['long']); const huge = 'x'.repeat(5000); engine.setAgentExecutor(async () => ({ status: 'success', output: huge })); engine.startRun(runId); await engine.nextStep(runId); const out = streamEvents(runId).find((e) => e.event === 'node.output'); expect(out.data.status).toBe('success'); expect(out.data.output_excerpt.startsWith('xxxx')).toBe(true); // 截断了，且短得多（完整输出只在 output_ref 指的产物目录里，不进事件也不进库） expect(out.data.output_excerpt.length).toBeLessThan(huge.length); expect(out.data.output_excerpt).toContain('已截断'); }); it('executor 没有 output 时 output_excerpt 是 null，不塞空串', async () => { const runId = buildRun(['quiet']); engine.setAgentExecutor(async () => ({ status: 'success' })); engine.startRun(runId); await engine.nextStep(runId); const out = streamEvents(runId).find((e) => e.event === 'node.output'); expect(out.data.output_excerpt).toBeNull(); // 没抛异常就不该有 node.error expect(timeline(runId).filter((e) => e.kind === 'node.error')).toHaveLength(0); }); });