test: 测试按行为命名、server 测试统一到 test/、PG 测试分组并行
- server 的 src/modules/*/__tests__ 挪到 test/<模块>/;e2e 去掉 eN- 前缀, w5a2 / v2 / search2 / *-dh1 / *-md1 / *-md2、web 的 agui3 / search2、domain 的 id2 / id3 改为按行为命名 - vitest 分三组:unit、pg(真实 PG + 假引擎,文件并行,最多 6 个)、serial(真实引擎 / worker / e2e,串行); pg 组用例超时放宽到 60s(迁移类用例并行时排队等 schema 锁)。server 测试 615s → 约 350s - CLAUDE.md 补测试约定;待办补 harness 去重与 continuous-series 偶发超时(疑似 Slot 评估竞争) - ADR-004 补充:维持 Deno,记录实际只用到 --allow-run 白名单 Co-Authored-By: Claude Opus 5.5 <[email protected]>
This commit is contained in:
rootandClaude Opus 5.5 committed
1 parent
b343dddbf6
commit
b61335eaf6
75 files changed
+159
-115
No files matched your search
@@ -0,0 +1,148 @@
|
||||
import { TMDB_FIXTURES, type TmdbFixture } from "@umt/testkit";
|
||||
import { afterAll, beforeAll, describe, expect, it } from "vitest";
|
||||
import { INVESTIGATE_CAPABILITIES } from "../../src/modules/agent/runner.ts";
|
||||
import { investigateScript } from "./fixtures/scripts.ts";
|
||||
import { createE2E, type E2E, pgAvailable, runEvents, seedBase, waitFor } from "./harness.ts";
|
||||
|
||||
// E5 后台唤醒:识别歧义产生 decision.created → outbox → resolveDecision → autoInvestigate 开启时入队有界 runAgent
|
||||
// (plan_only、白名单只读 + identity.resolve、小预算、trigger=event);同一 decision 只一个 run(dedupe_key)。
|
||||
|
||||
const hasPg = await pgAvailable();
|
||||
|
||||
const twin = (id: number): TmdbFixture => ({
|
||||
type: "movie",
|
||||
id,
|
||||
original_title: "Twins",
|
||||
original_language: "en",
|
||||
origin_country: ["US"],
|
||||
date: "1988-12-09",
|
||||
titles: { "zh-CN": "龙兄鼠弟", "en-US": "Twins" },
|
||||
popularity: 10 + id / 1000,
|
||||
alternative_titles: [],
|
||||
external_ids: {},
|
||||
});
|
||||
|
||||
describe.skipIf(!hasPg)("E5 后台唤醒有界调查", () => {
|
||||
let e: E2E;
|
||||
let siteId: string;
|
||||
|
||||
beforeAll(async () => {
|
||||
e = await createE2E("e5", {
|
||||
fakeScripts: [investigateScript],
|
||||
tmdbFixtures: [...TMDB_FIXTURES, twin(111), twin(112)],
|
||||
});
|
||||
siteId = (await seedBase(e)).siteId;
|
||||
await e.settings({ agent: { autoInvestigate: true, investigateMaxToolCalls: 6 } });
|
||||
}, 120_000);
|
||||
|
||||
afterAll(async () => {
|
||||
await e?.close();
|
||||
}, 60_000);
|
||||
|
||||
it("识别歧义 → 自动发起只读调查 → 结论回写待处理事项;重复事件只一个 run", async () => {
|
||||
const rel = await e.db
|
||||
.insertInto("source_release")
|
||||
.values({ site_id: siteId, site_release_id: "990001", title: "Twins.1988.1080p.BluRay.x264-GRP" })
|
||||
.returning("id")
|
||||
.executeTakeFirstOrThrow();
|
||||
await e.booted.ctx.engine.enqueue(
|
||||
"resolveObservation",
|
||||
// 惰性识别(DEC-1):不相关的资源不查 TMDB;按搜索结果触发才会查到两个同名候选
|
||||
{ sourceReleaseId: rel.id, trigger: "search" },
|
||||
{ idempotencyKey: `e5:resolve:${rel.id}`, traceId: rel.id },
|
||||
);
|
||||
const decision = await waitFor(
|
||||
async () =>
|
||||
e.db
|
||||
.selectFrom("decision")
|
||||
.selectAll()
|
||||
.where("kind", "=", "identity_ambiguous")
|
||||
.where("subject_id", "=", rel.id)
|
||||
.executeTakeFirst(),
|
||||
60_000,
|
||||
"识别歧义 decision",
|
||||
);
|
||||
const run = await waitFor(
|
||||
async () =>
|
||||
e.db
|
||||
.selectFrom("agent_run")
|
||||
.selectAll()
|
||||
.where("dedupe_key", "=", `investigate:${decision.id}`)
|
||||
.executeTakeFirst(),
|
||||
60_000,
|
||||
"自动调查 run",
|
||||
);
|
||||
expect(run).toMatchObject({ trigger: "event", mode: "plan_only", created_by: "system:auto-investigate" });
|
||||
expect(run.goal).toContain(decision.id);
|
||||
expect(run.scope).toMatchObject({
|
||||
capabilities: [...INVESTIGATE_CAPABILITIES],
|
||||
allowedKinds: ["read", "preview", "propose"],
|
||||
});
|
||||
expect(run.budget).toMatchObject({ maxToolCalls: 6 });
|
||||
|
||||
const finished = await waitFor(
|
||||
async () => {
|
||||
const r = await e.db.selectFrom("agent_run").selectAll().where("id", "=", run.id).executeTakeFirstOrThrow();
|
||||
return ["succeeded", "failed", "budget_exceeded"].includes(r.status) ? r : null;
|
||||
},
|
||||
60_000,
|
||||
"调查 run 结束",
|
||||
);
|
||||
expect(finished.status).toBe("succeeded");
|
||||
const events = await runEvents(e, run.id);
|
||||
const invoked = events
|
||||
.filter((x) => x.type === "tool_call_start")
|
||||
.map((x) => (x.payload as { capabilityId: string }).capabilityId);
|
||||
expect(invoked).toEqual(["catalog.lookup", "inventory.query"]);
|
||||
// 白名单外的写能力:被拒绝,没有执行
|
||||
const rejected = events.filter((x) => x.type === "tool_call_rejected");
|
||||
expect(rejected.map((x) => (x.payload as { toolName: string }).toolName)).toEqual(["download__submit"]);
|
||||
expect(
|
||||
await e.db.selectFrom("capability_invocation").select("id").where("agent_run_id", "=", run.id).execute(),
|
||||
).toHaveLength(0);
|
||||
// 结论回写:待处理事项交还人工并带调查摘要
|
||||
const d2 = await e.db.selectFrom("decision").selectAll().where("id", "=", decision.id).executeTakeFirstOrThrow();
|
||||
expect(d2.status).toBe("awaiting_user");
|
||||
expect(d2.agent_run_id).toBe(run.id);
|
||||
expect((d2.evidence as { agentInvestigation?: { runId: string } }).agentInvestigation?.runId).toBe(run.id);
|
||||
|
||||
// 同一根因的重复事件(不同 outbox 键)与事项重新打开后的再次触发:仍只有一个 run
|
||||
await e.db.updateTable("decision").set({ status: "open" }).where("id", "=", decision.id).execute();
|
||||
for (const k of ["dup-1", "dup-2"]) {
|
||||
await e.booted.ctx.outbox.emit(
|
||||
e.db,
|
||||
"decision.created",
|
||||
{ decisionId: decision.id, kind: "identity_ambiguous", severity: "warning", title: decision.title },
|
||||
{ idempotencyKey: `e5:${k}` },
|
||||
);
|
||||
}
|
||||
await waitFor(
|
||||
async () => {
|
||||
const rows = await e.db
|
||||
.selectFrom("outbox_event")
|
||||
.select("status")
|
||||
.where("idempotency_key", "in", ["e5:dup-1", "e5:dup-2"])
|
||||
.execute();
|
||||
return rows.length === 2 && rows.every((r) => r.status === "dispatched");
|
||||
},
|
||||
30_000,
|
||||
"重复事件已投递",
|
||||
);
|
||||
// 等 resolveDecision 跑完
|
||||
await waitFor(
|
||||
async () => {
|
||||
const runs = await e.booted.ctx.engine.list({ taskName: "resolveDecision" });
|
||||
return runs.length >= 3 && runs.every((r) => r.status === "completed") ? runs : null;
|
||||
},
|
||||
30_000,
|
||||
"resolveDecision 完成",
|
||||
);
|
||||
const runs = await e.db
|
||||
.selectFrom("agent_run")
|
||||
.select("id")
|
||||
.where("dedupe_key", "=", `investigate:${decision.id}`)
|
||||
.execute();
|
||||
expect(runs).toHaveLength(1);
|
||||
expect(await e.db.selectFrom("agent_run").select("id").execute()).toHaveLength(1);
|
||||
}, 180_000);
|
||||
});
|
||||
Reference in new issue
Block a user