feat: v0.8.0 流语义补全 · 会话可靠 · 恢复力 — finish_reason 全链路贯通根治"思考中停止" · 2445 用例全量回归
P0 会话可靠性收口（根治"模型思考着会话就停止"）： - P0-1 finish_reason 全链路贯通：DONE 事件与 IterationStep 新增 finishReason，OpenAI 共享 SSE / Anthropic message_delta.stop_reason / Ollama done_reason 三路采集，TRACE 层弃用硬编码 'stop' 记录真值 - P0-2 空响应守卫 + 降级重试：零产出流→可重试错误走退避；思考耗尽输出预算（reasoning-only + length）→自动关闭思考降级重试一次；仍失败→OUTPUT_LENGTH_EXCEEDED 结构化错误 + 故障转移；附带根治 abort 恰逢零工具调用轮被 COMPLETED 抢占的真实缺陷 - P0-3 思考×能力×预算三对齐：DeepSeek/MiMo/Agnes/Ollama 四家 supportsThinking=false 强制不发思考参数；小输出预算告警；设置页联动提示 - P0-4 渲染层可见性：截断/空完成/友好错误三类提示，i18n 全部出层 - P0-5 回归四件套：reasoning-only 终止判定、集成级空闲超时、504 引擎重试归类、思考中 abort→USER_INTERRUPT、P4-2 强制收尾路径 FEAT-1：LLM 设置新增「最大输出上限」——Provider 支持矩阵显隐 + 模型上限钳制提示 + 超限保存警告 + llm.maxTokens 热生效 P1 修复面收口： - 渲染层三缺陷根治：后台会话回放缓冲（2000 条/4MB 有界 + agent:getReplayState + 事件总线）+ abort 双层自愈 + sendMessage 收尾兜底 + 中断卡片清扫 - 工具 abort 信号全覆盖：web_search/web_fetch/http_request/code_search/git 系列/delegate_task 全部接入引擎中断；web_search 时间预算收敛（720s→≤240s）；移除伪造 ToolExecutionContext 与死代码 - 安全：本地 Pinned CONNECT 代理根治浏览器通道 DNS rebinding（校验期 IP pinning，可注入 resolver 表测）；配置 URL 域名解析深校验（DeepCheckSoftFailure 软失败）；SSE 空 error 帧防御修复；Ollama generate/embed AbortSignal.any 合并 - 缺陷清单：UTF-16 BOM 读取、tmp 同毫秒碰撞（nanoid 后缀）、code_search JS 回退参数对称（case_sensitive/前后文独立）、list_directory include_node_modules、崩溃自愈退避（60s 窗 ≥3 次停 reload）、MemoryViewer/Sidebar i18n 收口 P2 能力演进： - 会话回收站：SCHEMA_VERSION 3 + 迁移 10（deleted_at，存在性守卫），软删除/恢复/彻底删除/30 天自动清理（启动+24h），searchMessages 聚合剔除，Sidebar 回收站面板 - 会话回放播放器：sessions:listRecordings/readRecording（白名单+目录边界+20MB 上限），SessionReplayPlayer 时间轴/步进/变速，Trace 面板入口 - electron-updater 自动更新：双轨（手动 feed 比对保留），生产环境启动静默检查 + update:status 广播 + app:updateInstall + LogsSettings UpdatePanel + builder publish 配置 - @ 文件提及：workspace.listFiles/readFileClip（边界/512KB/NUL 拒绝/MEMORY.md 保护），ChatInput Fuse 联想+键盘导航+附件管线注入 - MCP Resources/Prompts 发现：可选能力 try/catch 降级，mcp:listServerContents，MCPSettings 展开视图 - 文档对齐：内部 API 标准 HTML（Adapter 清单补 MiMo/已实现注记/STREAM_RESET/DONE.finishReason/ repetition_truncation 映射）；README v0.8.0 亮点表 P3 测试基建： - 新增 4 个测试文件：engine-stream-contract（6）、engine-stream-reliability（4：集成空闲超时/504 重试/思考中 abort/P4-2 强制收尾）、thinking-capability-gate（7）、pinned-proxy（9，含深校验 5）、session-trash（5，DB 域）、use-agent-stream hook 级（5）、agent.test 回放缓冲（2） - 契约更新：orchestrator 被中断 SubAgent success=false（abort 优先级修复语义）、SSE 空 error 帧、UTF-16 正常读取、DeepSeek 未配置思考显式 disabled、迁移矩阵 v2→3 - 弱断言根治：registry WEBP 单向断言、hooks-contracts 自比恒真、memory 空 token 补强 全量验证：typecheck 0 错误 / lint 0 问题 / 系统 Node 2144 通过（301 DB 用例按 ABI 跳过）/ Electron ABI 2445/2445 全量通过 0 跳过
This commit is contained in:
@@ -0,0 +1,243 @@
|
||||
/**
|
||||
* v0.8.0 P0-5 回归测试（第二批）：流可靠性四件套的剩余项。
|
||||
*
|
||||
* 1. parseSSEStream 集成级空闲超时：挂死 ReadableStream → 60s 无数据 →
|
||||
* SseUpstreamError(504)（P1-2 防线的集成真实性，此前仅纯函数层覆盖）；
|
||||
* 2. SseUpstreamError(504) 经引擎 isRetryableError 归类 → 退避重试后成功；
|
||||
* 3. REASONING_DELTA 持续输出中 abort → USER_INTERRUPT（"思考中点停止"场景）；
|
||||
* 4. 引擎 P4-2：上一 run 卡死（不响应 abort）→ 新 run 等待 30s → 强制 abort
|
||||
* → 再等 5s → 抛错（防"会话永久排队"）。
|
||||
*/
|
||||
|
||||
import { describe, it, expect, vi, afterEach } from 'vitest';
|
||||
import { parseSSEStream, SseUpstreamError } from '../../adapters/shared/sse-stream';
|
||||
import { AgentLoopEngine } from '../engine';
|
||||
import type { MetonaRequest, MetonaStreamEvent, IMetonaProviderAdapter } from '../../types';
|
||||
import { MetonaStreamEventType } from '../../types';
|
||||
|
||||
const SYSTEM_PROMPT = {
|
||||
roleDefinition: 'test',
|
||||
outputConstraints: '',
|
||||
safetyGuidelines: '',
|
||||
};
|
||||
|
||||
afterEach(() => {
|
||||
vi.useRealTimers();
|
||||
});
|
||||
|
||||
// ===== 1. 集成级空闲超时 =====
|
||||
|
||||
describe('parseSSEStream integrated idle timeout (P0-5 #2)', () => {
|
||||
function makeHungBody(): ReadableStream<Uint8Array> {
|
||||
return {
|
||||
getReader: () => ({
|
||||
// 永不 resolve 的 read —— 模拟"服务器保活但不再推数据"
|
||||
read: () => new Promise<never>(() => {}),
|
||||
releaseLock: () => {},
|
||||
}),
|
||||
} as unknown as ReadableStream<Uint8Array>;
|
||||
}
|
||||
|
||||
it('挂死流 60s 无数据 → SseUpstreamError(504)', async () => {
|
||||
vi.useFakeTimers();
|
||||
const iterator = parseSSEStream(makeHungBody(), 'r', 's', 1);
|
||||
let thrown: unknown = null;
|
||||
const firstRead = iterator.next().catch((e: unknown) => {
|
||||
thrown = e;
|
||||
return { done: true, value: undefined } as const;
|
||||
});
|
||||
// 推进 61s：IDLE_TIMEOUT_MS=60_000 触发 idle controller → read() 竞速失败
|
||||
await vi.advanceTimersByTimeAsync(61_000);
|
||||
await firstRead;
|
||||
expect(thrown).toBeInstanceOf(SseUpstreamError);
|
||||
expect((thrown as SseUpstreamError).status).toBe(504);
|
||||
expect((thrown as Error).message).toContain('idle timeout');
|
||||
});
|
||||
});
|
||||
|
||||
// ===== 2. SseUpstreamError(504) → 引擎重试 =====
|
||||
|
||||
describe('engine: SseUpstreamError(504) retry classification (P0-5 #3)', () => {
|
||||
it('504 首败 → 退避重试 → 第二次流成功 → COMPLETED', async () => {
|
||||
let call = 0;
|
||||
const adapter = {
|
||||
providerId: 'mock' as const,
|
||||
supportedModels: ['m'],
|
||||
supportsToolCalling: true,
|
||||
supportsThinking: true,
|
||||
getContextWindow: () => 128_000,
|
||||
async send(): Promise<unknown> {
|
||||
throw new Error('not used');
|
||||
},
|
||||
async *sendStream(_r: MetonaRequest): AsyncIterable<MetonaStreamEvent> {
|
||||
call++;
|
||||
if (call === 1) {
|
||||
throw new SseUpstreamError('Stream idle timeout after 60000ms (no data received)', {
|
||||
status: 504,
|
||||
});
|
||||
}
|
||||
yield {
|
||||
type: MetonaStreamEventType.TEXT_DELTA,
|
||||
requestId: 'r',
|
||||
sessionId: 's',
|
||||
iteration: 1,
|
||||
seq: 0,
|
||||
timestamp: Date.now(),
|
||||
delta: 'recovered',
|
||||
};
|
||||
yield {
|
||||
type: MetonaStreamEventType.DONE,
|
||||
requestId: 'r',
|
||||
sessionId: 's',
|
||||
iteration: 1,
|
||||
seq: 1,
|
||||
timestamp: Date.now(),
|
||||
finishReason: 'stop',
|
||||
};
|
||||
},
|
||||
} as unknown as IMetonaProviderAdapter;
|
||||
|
||||
const engine = new AgentLoopEngine(
|
||||
{ maxIterations: 3, totalTimeoutMs: 15_000, retryCount: 2, contextWindow: 128_000 },
|
||||
adapter,
|
||||
);
|
||||
const events: MetonaStreamEvent[] = [];
|
||||
engine.on('streamEvent', (e: MetonaStreamEvent) => events.push(e));
|
||||
const output = await engine.runStream(
|
||||
{ role: 'user', content: 'hi', timestamp: Date.now() },
|
||||
's1',
|
||||
[],
|
||||
SYSTEM_PROMPT,
|
||||
);
|
||||
expect(output.terminationReason).toBe('completed');
|
||||
expect(output.finalAnswer).toBe('recovered');
|
||||
expect(call).toBe(2);
|
||||
// 重试经过 RETRY 事件（引擎消费后向渲染层转 STREAM_RESET）
|
||||
expect(events.some((e) => e.type === MetonaStreamEventType.STREAM_RESET)).toBe(true);
|
||||
});
|
||||
});
|
||||
|
||||
// ===== 3. REASONING_DELTA 中途 abort =====
|
||||
|
||||
describe('engine: abort during reasoning stream (P0-5 #4)', () => {
|
||||
it('思考增量持续输出中 abort → USER_INTERRUPT，迟到 reasoning 不再产出事件', async () => {
|
||||
const adapter = {
|
||||
providerId: 'mock' as const,
|
||||
supportedModels: ['m'],
|
||||
supportsToolCalling: true,
|
||||
supportsThinking: true,
|
||||
getContextWindow: () => 128_000,
|
||||
async send(): Promise<unknown> {
|
||||
throw new Error('not used');
|
||||
},
|
||||
// 持续每 10ms 推送一个 reasoning delta，直到 abort 或 500 个（挂起）
|
||||
async *sendStream(_r: MetonaRequest): AsyncIterable<MetonaStreamEvent> {
|
||||
for (let i = 0; i < 500; i++) {
|
||||
if (_r.params.thinkingEnabled === false) return;
|
||||
yield {
|
||||
type: MetonaStreamEventType.REASONING_DELTA,
|
||||
requestId: 'r',
|
||||
sessionId: 's',
|
||||
iteration: 1,
|
||||
seq: i,
|
||||
timestamp: Date.now(),
|
||||
delta: `t${i};`,
|
||||
};
|
||||
await new Promise((r) => setTimeout(r, 10));
|
||||
}
|
||||
yield {
|
||||
type: MetonaStreamEventType.DONE,
|
||||
requestId: 'r',
|
||||
sessionId: 's',
|
||||
iteration: 1,
|
||||
seq: 999,
|
||||
timestamp: Date.now(),
|
||||
};
|
||||
},
|
||||
} as unknown as IMetonaProviderAdapter;
|
||||
|
||||
const engine = new AgentLoopEngine(
|
||||
{ maxIterations: 3, totalTimeoutMs: 30_000, retryCount: 0, contextWindow: 128_000 },
|
||||
adapter,
|
||||
);
|
||||
const events: MetonaStreamEvent[] = [];
|
||||
engine.on('streamEvent', (e: MetonaStreamEvent) => events.push(e));
|
||||
|
||||
const run = engine.runStream(
|
||||
{ role: 'user', content: 'hi', timestamp: Date.now() },
|
||||
's1',
|
||||
[],
|
||||
SYSTEM_PROMPT,
|
||||
);
|
||||
// 等待至少 3 个 reasoning delta 到达（确认流确实在思考态输出中）
|
||||
await new Promise((r) => setTimeout(r, 80));
|
||||
expect(
|
||||
events.filter((e) => e.type === MetonaStreamEventType.REASONING_DELTA).length,
|
||||
).toBeGreaterThanOrEqual(3);
|
||||
engine.abort();
|
||||
const output = await run;
|
||||
expect(output.terminationReason).toBe('user_interrupt');
|
||||
const lastReasoningIdx = events
|
||||
.map((e) => e.type)
|
||||
.lastIndexOf(MetonaStreamEventType.REASONING_DELTA);
|
||||
const doneIdx = events.findIndex((e) => e.type === MetonaStreamEventType.DONE);
|
||||
// abort 生效后不再有迟到 reasoning（abort 尾巴过滤在引擎侧）
|
||||
expect(lastReasoningIdx).toBeLessThan(doneIdx);
|
||||
}, 15_000);
|
||||
});
|
||||
|
||||
// ===== 4. P4-2: 上一 run 卡死 → 30s 强制 abort + 5s 再等待 → 抛错 =====
|
||||
|
||||
describe('engine: stuck previous run force-abort path (P4-2)', () => {
|
||||
it('上一 run 不响应 abort → 新 run 30s 超时强制 abort → 5s 后仍卡 → 抛错', async () => {
|
||||
vi.useFakeTimers();
|
||||
let released = false;
|
||||
const stuckAdapter = {
|
||||
providerId: 'mock' as const,
|
||||
supportedModels: ['m'],
|
||||
supportsToolCalling: true,
|
||||
supportsThinking: true,
|
||||
getContextWindow: () => 128_000,
|
||||
async send(): Promise<unknown> {
|
||||
throw new Error('not used');
|
||||
},
|
||||
// 第一次 run：永远挂起且不响应 abort（模拟卡死的底层 fetch）；
|
||||
// 挂起生成器无 yield 属场景定义（require-yield 由下行指令豁免）
|
||||
// eslint-disable-next-line require-yield
|
||||
async *sendStream(): AsyncIterable<MetonaStreamEvent> {
|
||||
while (!released) {
|
||||
await new Promise(() => {});
|
||||
}
|
||||
return;
|
||||
},
|
||||
} as unknown as IMetonaProviderAdapter;
|
||||
|
||||
const engine = new AgentLoopEngine(
|
||||
{ maxIterations: 3, totalTimeoutMs: 600_000, retryCount: 0, contextWindow: 128_000 },
|
||||
stuckAdapter,
|
||||
);
|
||||
const firstRun = engine.runStream(
|
||||
{ role: 'user', content: 'first', timestamp: Date.now() },
|
||||
's1',
|
||||
[],
|
||||
SYSTEM_PROMPT,
|
||||
);
|
||||
firstRun.catch(() => {});
|
||||
|
||||
// 新消息进入 → 等待上一 run 结束（30s 超时）→ 强制 abort → 5s 再等待 → 抛错
|
||||
const secondRun = engine.runStream(
|
||||
{ role: 'user', content: 'second', timestamp: Date.now() },
|
||||
's1',
|
||||
[],
|
||||
SYSTEM_PROMPT,
|
||||
);
|
||||
const expectation = expect(secondRun).rejects.toThrow(/Previous run did not finish/i);
|
||||
|
||||
// 驱动 30s 等待超时 + 5s 强制 abort 等待
|
||||
await vi.advanceTimersByTimeAsync(30_500);
|
||||
await vi.advanceTimersByTimeAsync(5_500);
|
||||
// 清理：放行挂死流，让 firstRun 收敛
|
||||
released = true;
|
||||
await expectation;
|
||||
}, 20_000);
|
||||
});
|
||||
Reference in New Issue
Block a user