fix(phase-5.2): 消项 CR 第二轮 P1/P2/P3(通道/证据链/交付时序/后管证据页)

This commit is contained in:
2026-10-31 21:12:18 +08:00
parent a4e4456b50
commit 91d36327f9
61 changed files with 2833 additions and 311 deletions
+156 -21
View File
@@ -1,38 +1,172 @@
import { http, unwrap } from '@/api/http'
/** 大档分片阈值(AC-11):8MB 以上不再一次 PUT 直传,断点续传由服务端 chunk 通道承担。 */
export const CHUNK_UPLOAD_THRESHOLD = 8 * 1024 * 1024
/**
* 大档分片阈值与片大小(AC-11):后管这一侧只有这一个出口,阈值由片大小派生。
*
* ⚠ 同族的 8MB 数还另有两处:用户端上传出口一处、桌面客户端分片入口一处。两边是各自
* 独立的构建根(各自的 `@` 别名与 tsconfig include),客户端是 Python 打包物 ⇒
* 没有共享包就合不成同一个字面量(P3-5 的边界,已登记)。
*/
export const CHUNK_UPLOAD_PART_SIZE = 8 * 1024 * 1024
export const CHUNK_UPLOAD_THRESHOLD = CHUNK_UPLOAD_PART_SIZE
/** 每片重试次数(不含首次):单片 502/超时不该让 512MB 从 0 再来。 */
const PART_RETRY_LIMIT = 2
/**
* 逐片超时(P2-8):预算只看**这一片**有多大——60s 保底 + 每 MB 30s(8MB 片=300s)。
* 过去按整包体积算(120s + MB×15s),512MB 包里的一个 8MB 片能拿到 128 分钟,
* 慢链路真卡死时没人知道,进度条还会一直爬。
*/
export function partTimeoutOf(partBytes: number): number {
return 60_000 + Math.ceil(partBytes / (1024 * 1024)) * 30_000
}
/** 会话登记/查询只发元数据,固定预算即可;失败要快快回来,别陪整包一起放大。 */
const REQUEST_TIMEOUT_MS = 60_000
/** 只有拼装这一步真随整包体积走:服务端要读全部分片、复算 sha256。 */
const COMPLETE_TIMEOUT_MS = 600_000
function partUrl(uploadId: string, index: number): string {
return `/api/files/chunk/${encodeURIComponent(uploadId)}/part/${index}`
}
/** 后管大档上传(版本包/教程包共用):init → GET 已达片号 → 逐片 POST → complete。 */
/** 分片会话跨尝试/刷新续用:同一文件身份复用 upload_id,服务端没回收就从已达片号继续(P2-7)。 */
interface StoredChunkSession {
uploadId: string;
size: number;
lastModified: number;
}
const CHUNK_SESSION_PREFIX = 'shufu:admin:chunk:'
function lastModifiedOf(file: Blob): number {
const value = (file as File).lastModified
return typeof value === 'number' && Number.isFinite(value) ? value : 0
}
/** 会话键:文件名+体积+修改时间,三者同一条上传才算同一个件。 */
export function chunkSessionKeyOf(fileName: string, file: Blob): string {
return `${CHUNK_SESSION_PREFIX}${fileName}:${file.size}:${lastModifiedOf(file)}`
}
function readStoredSession(key: string, file: Blob): string {
if (typeof localStorage === 'undefined') return ''
try {
const raw = localStorage.getItem(key)
if (!raw) return ''
const parsed = JSON.parse(raw) as StoredChunkSession
return parsed.size === file.size && parsed.lastModified === lastModifiedOf(file) ? parsed.uploadId : ''
} catch {
return ''
}
}
function writeStoredSession(key: string, file: Blob, uploadId: string): void {
if (typeof localStorage === 'undefined') return
try {
localStorage.setItem(key, JSON.stringify({
uploadId, size: file.size, lastModified: lastModifiedOf(file),
} satisfies StoredChunkSession))
} catch {
// 存储不可用(隐私模式/配额满)只影响跨刷新续传,不影响本次上传
}
}
function clearStoredSession(key: string): void {
if (typeof localStorage === 'undefined') return
try {
localStorage.removeItem(key)
} catch {
// 同上:清不掉只会让下次多问一次 status
}
}
/**
* 服务端对「这个会话还在不在」的三种答复:还在(附带已到片号)/确实没了/问不出来。
* 第三态是 P2-7 与 P1-11 的共同判据——过去它被当成"没会话",瞬态抖动=整包作废。
*/
type ProbeOutcome
= { state: 'ok'; parts: number[] }
| { state: 'gone' }
| { state: 'transient' }
/** 后端回收会话的原文案(ChunkedUploadService.readMeta);BusinessException 不带 code ⇒ HTTP 200 出。 */
function messageSaysSessionGone(text: unknown): boolean {
const message = typeof text === 'string' ? text : ''
return ['不存在', '过期'].some((marker) => message.includes(marker))
}
function classifyProbe(response: unknown): ProbeOutcome {
const status = Number((response as { status?: unknown } | null | undefined)?.status)
if (status === 404) return { state: 'gone' }
if (status >= 500 || status === 408 || status === 429) return { state: 'transient' }
const body = (response as { data?: unknown } | null | undefined)?.data
if (body && typeof body === 'object' && (body as { success?: unknown }).success === false) {
return messageSaysSessionGone((body as { message?: unknown }).message)
? { state: 'gone' }
: { state: 'transient' }
}
const core = (unwrap<Record<string, unknown>>(body) ?? {}) as Record<string, unknown>
const arrived = core.uploaded_parts
return { state: 'ok', parts: Array.isArray(arrived) ? arrived.map((value) => Number(value)) : [] }
}
/**
* 探测已达片号(P2-7/与用户端 upload.ts 同一口径)。
* validateStatus 恒真:状态码要自己判——真 404/5xx 一旦被拦截器折成 Error 就分不出原因了;
* 能走到 catch 的就只剩网络层失败与超时,铁定瞬态。
*/
async function probeUploadedParts(uploadId: string): Promise<ProbeOutcome> {
try {
const response = await http.get(
`/api/files/chunk/${encodeURIComponent(uploadId)}`,
{ validateStatus: () => true },
)
return classifyProbe(response)
} catch (error) {
const status = (error as { response?: { status?: unknown } } | null | undefined)?.response?.status
if (status === 404) return { state: 'gone' }
if (messageSaysSessionGone((error as { message?: unknown } | null | undefined)?.message)) {
return { state: 'gone' }
}
return { state: 'transient' }
}
}
/** 后管大档上传(版本包/教程包共用):会话续用 → GET 已达片号 → 逐片 POST → complete。 */
export async function uploadFileInChunks(
file: Blob,
fileName: string,
onProgress?: (percent: number) => void,
): Promise<Record<string, unknown>> {
const totalParts = Math.max(1, Math.ceil(file.size / CHUNK_UPLOAD_PART_SIZE))
const initResponse = await http.post('/api/files/chunk/init', null, {
params: { fileName, totalParts, totalBytes: file.size },
timeout: 120_000 + Math.ceil(file.size / (1024 * 1024)) * 15_000,
})
const init = (unwrap(initResponse.data) ?? {}) as Record<string, unknown>
const uploadId = String(init.upload_id ?? '')
if (!uploadId) {
throw new Error('后端未返回分片会话,请重试')
const sessionKey = chunkSessionKeyOf(fileName, file)
let uploadId = readStoredSession(sessionKey, file)
const arrived = new Set<number>()
if (uploadId) {
const probed = await probeUploadedParts(uploadId)
if (probed.state === 'gone') {
// 服务端确实说会话没了(404/「不存在或已过期」):清掉重开,否则这个件每次都在同一个死会话上失败。
clearStoredSession(sessionKey)
uploadId = ''
} else if (probed.state === 'ok') {
probed.parts.forEach((index) => arrived.add(index))
}
// transient:片号未知但会话还在 ⇒ 照原会话逐片重传(后端同片覆盖,幂等),不重开 init。
}
if (!uploadId) {
const initResponse = await http.post('/api/files/chunk/init', null, {
params: { fileName, totalParts, totalBytes: file.size },
timeout: REQUEST_TIMEOUT_MS,
})
const init = (unwrap(initResponse.data) ?? {}) as Record<string, unknown>
uploadId = String(init.upload_id ?? '')
if (!uploadId) {
throw new Error('后端未返回分片会话,请重试')
}
writeStoredSession(sessionKey, file, uploadId)
}
const statusResponse = await http.get(`/api/files/chunk/${encodeURIComponent(uploadId)}`)
const status = (unwrap(statusResponse.data) ?? {}) as Record<string, unknown>
const arrived = new Set<number>(
Array.isArray(status.uploaded_parts) ? status.uploaded_parts.map((value) => Number(value)) : [],
)
for (let index = 0; index < totalParts; index += 1) {
if (arrived.has(index)) {
@@ -47,7 +181,7 @@ export async function uploadFileInChunks(
try {
const partResponse = await http.post(partUrl(uploadId, index), form, {
headers: { 'Content-Type': 'multipart/form-data' },
timeout: 120_000 + Math.ceil(file.size / (1024 * 1024)) * 15_000,
timeout: partTimeoutOf(blob.size),
onUploadProgress: (event) => {
if (!onProgress) return
const perPart = Math.round(((index + (event.loaded || 0) / Math.max(blob.size, 1)) / totalParts) * 100)
@@ -63,14 +197,15 @@ export async function uploadFileInChunks(
}
}
if (lastError) {
// 后管上传是一次性的:放弃时把会话连同临时片一起清掉,不留 24h 残盘。
await http.delete(`/api/files/chunk/${encodeURIComponent(uploadId)}`).catch(() => undefined)
// 不 DELETE(P2-7):会话连同已达片号留在服务端(后端 24h 过期回收兜底),
// 下一次点开同一个文件就从断点接着传——这才是"续传"两个字的本义。
throw lastError instanceof Error ? lastError : new Error('分片上传失败')
}
}
const completeResponse = await http.post(
`/api/files/chunk/${encodeURIComponent(uploadId)}/complete`, null, { timeout: 600_000 })
`/api/files/chunk/${encodeURIComponent(uploadId)}/complete`, null, { timeout: COMPLETE_TIMEOUT_MS })
clearStoredSession(sessionKey)
if (onProgress) onProgress(100)
return (unwrap(completeResponse.data) ?? {}) as Record<string, unknown>
}
@@ -0,0 +1,61 @@
import { http } from './http'
import { unwrap } from './envelope'
/** 任务证据件列表行(AC-13:后管只读面,仅超级管理员)。 */
export interface TaskEvidenceRow {
id: number
taskId: string
uid: number | null
evidenceType: string
source: string
objectKey: string
byteSize: number
contentType: string
sha256: string
createdAt: string | null
}
export interface TaskEvidencePage {
items: TaskEvidenceRow[]
total: number
page: number
pageSize: number
}
export interface TaskEvidenceQuery {
taskId?: string
uid?: number
page: number
pageSize: number
}
export interface TaskEvidenceContent {
id: number
taskId: string
contentType: string
binary: boolean
content: string
}
export async function fetchTaskEvidenceFiles(params: TaskEvidenceQuery): Promise<TaskEvidencePage> {
const query: Record<string, string | number> = { page: params.page, pageSize: params.pageSize }
if (params.taskId) query.taskId = params.taskId
if (params.uid) query.uid = params.uid
const { data } = await http.get('/api/admin/task-evidence/files', { params: query })
return unwrap<TaskEvidencePage>(data)
}
export async function fetchTaskEvidenceContent(fileId: number): Promise<TaskEvidenceContent> {
const { data } = await http.get('/api/admin/task-evidence/content', { params: { fileId } })
return unwrap<TaskEvidenceContent>(data)
}
/** 下载走服务端流,不把对象直链/预签名 URL 交给浏览器(AC-13 的不入日志口径)。 */
export function taskEvidenceDownloadUrl(fileId: number): string {
return `/api/admin/task-evidence/download?fileId=${fileId}`
}
export async function deleteTaskEvidence(ids: number[]): Promise<number> {
const { data } = await http.post('/api/admin/task-evidence/delete', ids)
return unwrap<{ deleted: number }>(data).deleted
}
@@ -3,7 +3,7 @@
* 而护栏用例要从这里反过来读注册表,放在一起会成环。 */
/** 角色内在闸门:只有超管可见可读写,与 user_column_permission 无关。 */
export const SUPER_ADMIN_ONLY_MENU_KEYS: readonly string[] = ['admin_companies']
export const SUPER_ADMIN_ONLY_MENU_KEYS: readonly string[] = ['admin_companies', 'admin_task_evidence']
export interface CompanyScopedPage {
/** 与 routes.ts 的 menuKey 逐字一致。 */
@@ -109,6 +109,10 @@ export const OPERATION_GUIDES: Record<string, OperationGuideData> = {
text: '桌面客户端与麦象采集机上报的日志按来源、日期和关键字筛选;可查看日志尾部、下载完整文件或删除云端记录,云端日志按保留期自动清理。',
steps: ['选择来源与日期', '查看或下载日志', '必要时删除'],
},
admin_task_evidence: {
text: '任务异常现场件与窗口图按任务号关联;只有超级管理员能读取与下载,对象直链不出服务端,证据件按保留期自动清理。',
steps: ['按任务号或 uid 查询', '查看现场件或窗口图', '下载或清理过期证据'],
},
admin_user_secret_usage: {
text: '按日期范围统计各用户的密钥调用次数(货源查询与外观专利按 LLM 请求次数、代理按提取次数),用于评估用户资源损耗。',
steps: ['选择日期范围', '筛选用户或分组', '核对各模块次数'],
@@ -9,6 +9,7 @@ import { fetchSoftwareVersions, uploadSoftwareVersion, deleteSoftwareVersions }
import type { SoftwareVersionItem } from './version-dto.ts'
import { isNonEmptyVersion } from './version-dto.ts'
import { showAdminFeedback } from '@/components/admin-feedback-ui'
import { loadFailureMessage } from '@/components/load-failure-model'
import { openAdminConfirm } from '@/components/admin-confirm'
import LoadFailureBar from '@/components/LoadFailureBar.vue'
@@ -62,7 +63,7 @@ async function removeSelected() {
selectedIds.value = new Set()
await load()
} catch (error) {
showAdminFeedback(error instanceof Error ? error.message : '删除失败', 'error')
showAdminFeedback(loadFailureMessage(error, '删除失败'), 'error')
}
}
async function removeOne(row: SoftwareVersionItem) {
@@ -73,7 +74,7 @@ async function removeOne(row: SoftwareVersionItem) {
selectedIds.value = new Set([...selectedIds.value].filter((id) => id !== row.id))
await load()
} catch (error) {
showAdminFeedback(error instanceof Error ? error.message : '删除失败', 'error')
showAdminFeedback(loadFailureMessage(error, '删除失败'), 'error')
}
}
@@ -92,7 +93,7 @@ async function load() {
try {
items.value = (await fetchSoftwareVersions()).items
} catch (error) {
loadFailure.value = error instanceof Error ? error.message : '版本列表加载失败'
loadFailure.value = loadFailureMessage(error, '版本列表加载失败')
showAdminFeedback(loadFailure.value, 'error')
} finally {
loading.value = false
@@ -146,7 +147,8 @@ async function submitUpload() {
uploadPercent.value = 0
load()
} catch (error) {
uploadMsg.value = error instanceof Error ? error.message : '上传失败'
// P2-10:api/http 拦截器不 sanitize,error.message 可能是英文异常原文——界面只吃中文出口,原文进 console。
uploadMsg.value = loadFailureMessage(error, '上传失败')
} finally {
uploading.value = false
}
@@ -11,6 +11,7 @@ import { fetchTutorialPackages, uploadTutorialPackage, deleteTutorialPackages }
import { formatFileSize } from './tutorial-model.ts'
import type { TutorialPackageItem } from './tutorial-dto.ts'
import { showAdminFeedback } from '@/components/admin-feedback-ui'
import { loadFailureMessage } from '@/components/load-failure-model'
import { openAdminConfirm } from '@/components/admin-confirm'
import LoadFailureBar from '@/components/LoadFailureBar.vue'
@@ -119,7 +120,7 @@ async function removeSelected() {
selectedIds.value = new Set()
await load()
} catch (error) {
showAdminFeedback(error instanceof Error ? error.message : '删除失败', 'error')
showAdminFeedback(loadFailureMessage(error, '删除失败'), 'error')
}
}
async function removeOne(row: TutorialPackageItem) {
@@ -130,7 +131,7 @@ async function removeOne(row: TutorialPackageItem) {
selectedIds.value = new Set([...selectedIds.value].filter((id) => id !== row.id))
await load()
} catch (error) {
showAdminFeedback(error instanceof Error ? error.message : '删除失败', 'error')
showAdminFeedback(loadFailureMessage(error, '删除失败'), 'error')
}
}
@@ -149,7 +150,7 @@ async function load() {
try {
items.value = (await fetchTutorialPackages()).items
} catch (error) {
loadFailure.value = error instanceof Error ? error.message : '教程包列表加载失败'
loadFailure.value = loadFailureMessage(error, '教程包列表加载失败')
showAdminFeedback(loadFailure.value, 'error')
} finally {
loading.value = false
@@ -198,7 +199,8 @@ async function submitUpload() {
uploadPercent.value = 0
load()
} catch (error) {
uploadMsg.value = error instanceof Error ? error.message : '上传失败'
// P2-10:api/http 拦截器不 sanitize,error.message 可能是英文异常原文——界面只吃中文出口,原文进 console。
uploadMsg.value = loadFailureMessage(error, '上传失败')
} finally {
uploading.value = false
}
@@ -0,0 +1,305 @@
<script setup lang="ts">
/**
* 任务证据件(R3「后管可视」):按 taskId / uid 关联查看异常现场件与窗口图。
* 读取面在服务端逐个方法走 requireSuperAdmin;本页只消费服务端流,不接触对象直链。
*/
import { computed, onMounted, ref, watch } from 'vue'
import OldPagination from '@/components/OldPagination.vue'
import LoadFailureBar from '@/components/LoadFailureBar.vue'
import { showAdminFeedback } from '@/components/admin-feedback-ui'
import { formatDateTime } from '@/utils/datetime'
import {
deleteTaskEvidence,
fetchTaskEvidenceContent,
fetchTaskEvidenceFiles,
taskEvidenceDownloadUrl,
} from '@/api/task-evidence'
import type { TaskEvidenceRow } from '@/api/task-evidence'
const loading = ref(false)
const loadFailure = ref('')
const items = ref<TaskEvidenceRow[]>([])
const total = ref(0)
const page = ref(1)
const pageSize = ref(20)
const taskIdFilter = ref('')
const uidFilter = ref('')
const selectedIds = ref<Set<number>>(new Set())
const selectedCount = computed(() => selectedIds.value.size)
const previewVisible = ref(false)
const previewLoading = ref(false)
const previewRow = ref<TaskEvidenceRow | null>(null)
const previewText = ref('')
const previewImage = ref('')
function imageOf(row: TaskEvidenceRow, content: string): string {
return `data:${row.contentType || 'image/png'};base64,${content}`
}
async function load() {
loading.value = true
loadFailure.value = ''
try {
const uid = uidFilter.value.trim() ? Number(uidFilter.value.trim()) : undefined
const data = await fetchTaskEvidenceFiles({
taskId: taskIdFilter.value.trim() || undefined,
uid: Number.isFinite(uid) && (uid || 0) > 0 ? uid : undefined,
page: page.value,
pageSize: pageSize.value,
})
items.value = data.items
total.value = data.total
} catch (error) {
loadFailure.value = error instanceof Error ? error.message : '任务证据件加载失败'
showAdminFeedback(loadFailure.value, 'error')
} finally {
loading.value = false
}
}
function search() {
page.value = 1
selectedIds.value = new Set()
load()
}
function changePage(next: number) {
page.value = next
load()
}
function changeSize(size: number) {
pageSize.value = size
page.value = 1
load()
}
function toggleSelect(id: number) {
const next = new Set(selectedIds.value)
if (next.has(id)) next.delete(id)
else next.add(id)
selectedIds.value = next
}
function toggleSelectAllPage() {
const allOn = items.value.length > 0 && items.value.every((row) => selectedIds.value.has(row.id))
const next = new Set(selectedIds.value)
if (allOn) items.value.forEach((row) => next.delete(row.id))
else items.value.forEach((row) => next.add(row.id))
selectedIds.value = next
}
async function removeSelected() {
if (!selectedCount.value) return
try {
const deleted = await deleteTaskEvidence(Array.from(selectedIds.value))
showAdminFeedback(`已删除 ${deleted} 份证据件`, 'success')
selectedIds.value = new Set()
await load()
} catch (error) {
showAdminFeedback(error instanceof Error ? error.message : '删除证据件失败', 'error')
}
}
async function openPreview(row: TaskEvidenceRow) {
previewRow.value = row
previewVisible.value = true
previewLoading.value = true
previewText.value = ''
previewImage.value = ''
try {
const content = await fetchTaskEvidenceContent(row.id)
if (content.binary) previewImage.value = imageOf(row, content.content)
else previewText.value = content.content
} catch (error) {
previewText.value = error instanceof Error ? error.message : '证据件读取失败'
} finally {
previewLoading.value = false
}
}
function closePreview() {
previewVisible.value = false
previewRow.value = null
previewText.value = ''
previewImage.value = ''
}
function kindLabel(row: TaskEvidenceRow): string {
if (row.contentType?.startsWith('image/')) return '窗口/页面图'
return row.evidenceType || '证据件'
}
watch([taskIdFilter, uidFilter], () => {
// 过滤条件改了不自动打请求,避免一次输入打满接口;按「查询」或回车再取。
})
onMounted(load)
</script>
<template>
<div class="evidence-view">
<section class="panel-box">
<div class="evidence-head">
<h3>任务证据件</h3>
<div class="evidence-tools">
<input
v-model="taskIdFilter"
type="text"
class="evidence-input"
placeholder="按任务号筛选"
@keyup.enter="search"
/>
<input
v-model="uidFilter"
type="text"
class="evidence-input evidence-uid"
placeholder="按 uid 筛选"
@keyup.enter="search"
/>
<button class="btn" type="button" @click="search">查询</button>
<button class="btn btn-danger" type="button" :disabled="!selectedCount" @click="removeSelected">
删除{{ selectedCount ? `(${selectedCount})` : '' }}
</button>
</div>
</div>
<p class="page-hint">
异常现场件与窗口图按任务号关联;读取与下载仅超级管理员可用,对象直链不出服务端。
</p>
<LoadFailureBar v-if="loadFailure" :message="loadFailure" @retry="load" />
<div class="table-scroll">
<table>
<thead>
<tr>
<th style="width: 40px">
<input
type="checkbox"
:checked="items.length > 0 && items.every((row) => selectedIds.has(row.id))"
:disabled="!items.length"
@change="toggleSelectAllPage"
/>
</th>
<th style="width: 90px">ID</th>
<th style="width: 150px">任务号</th>
<th style="width: 90px">UID</th>
<th style="width: 130px">类型</th>
<th style="width: 130px">来源</th>
<th style="width: 110px">体积</th>
<th style="width: 160px">生成时间</th>
<th>操作</th>
</tr>
</thead>
<tbody>
<template v-if="items.length">
<tr v-for="row in items" :key="row.id">
<td>
<input type="checkbox" :checked="selectedIds.has(row.id)" @change="toggleSelect(row.id)" />
</td>
<td>{{ row.id }}</td>
<td>{{ row.taskId || '-' }}</td>
<td>{{ row.uid ?? '-' }}</td>
<td>{{ kindLabel(row) }}</td>
<td>{{ row.source || '-' }}</td>
<td>{{ row.byteSize }} B</td>
<td>{{ formatDateTime(row.createdAt) }}</td>
<td class="evidence-actions">
<button class="link-like" type="button" @click="openPreview(row)">查看</button>
<a class="link-cell" :href="taskEvidenceDownloadUrl(row.id)" target="_blank" rel="noopener">下载</a>
</td>
</tr>
</template>
<tr v-else-if="loading">
<td colspan="9" class="empty-tip">加载中...</td>
</tr>
<tr v-else>
<td colspan="9" class="empty-tip">暂无证据件</td>
</tr>
</tbody>
</table>
</div>
<OldPagination :total="total" :page="page" :page-size="pageSize" @change="changePage" @size-change="changeSize" />
</section>
<div v-if="previewVisible" class="modal-mask" @click.self="closePreview">
<div class="modal evidence-modal">
<div class="modal-head">
<h3 v-if="previewRow">证据件 #{{ previewRow.id }} · {{ previewRow.taskId || '-' }}</h3>
<button class="modal-close" type="button" aria-label="关闭" @click="closePreview">×</button>
</div>
<div v-if="previewLoading" class="modal-empty">读取中...</div>
<div v-else-if="previewImage" class="preview-grid">
<img :src="previewImage" alt="任务证据图" />
</div>
<pre v-else class="evidence-text">{{ previewText || '(空)' }}</pre>
<div class="modal-foot">
<button class="btn btn-secondary" type="button" @click="closePreview">关闭</button>
</div>
</div>
</div>
</div>
</template>
<style scoped>
.evidence-view {
font-family: inherit;
color: var(--c-text, #24384d);
display: flex;
flex-direction: column;
gap: 14px;
}
.evidence-head {
display: flex;
align-items: center;
justify-content: space-between;
gap: 12px;
flex-wrap: wrap;
}
.evidence-tools {
display: flex;
align-items: center;
gap: 8px;
flex-wrap: wrap;
}
.evidence-input {
width: 180px;
height: 32px;
padding: 0 10px;
border: 1px solid #cdd8e3;
border-radius: 6px;
}
.evidence-uid {
width: 120px;
}
.evidence-actions {
display: flex;
gap: 10px;
}
.link-like {
border: 0;
background: transparent;
color: #2563eb;
cursor: pointer;
padding: 0;
}
.evidence-modal {
width: min(960px, 92vw);
max-height: 86vh;
}
.evidence-text {
max-height: 60vh;
overflow: auto;
white-space: pre-wrap;
word-break: break-all;
background: #f6f8fa;
padding: 12px;
border-radius: 6px;
}
.preview-grid img {
max-width: 100%;
max-height: 60vh;
display: block;
margin: 0 auto;
}
</style>
@@ -10,8 +10,6 @@ export const TUTORIAL_UPLOAD_ENDPOINT = '/api/admin/tutorial'
export const TUTORIAL_PRESIGN_ENDPOINT = `${TUTORIAL_UPLOAD_ENDPOINT}/presign`
export const TUTORIAL_CONFIRM_ENDPOINT = `${TUTORIAL_UPLOAD_ENDPOINT}/confirm`
export const TUTORIAL_DELETE_ENDPOINT = `${TUTORIAL_UPLOAD_ENDPOINT}/delete`
/** 大档分片通道(AC-11):>8MB 的教程包走这里,不再一次性 presign PUT。 */
export const TUTORIAL_CHUNK_ENDPOINT = '/api/files/chunk'
/** MinIO 直传专用实例:不带 cookie、不挂 401 跳登录拦截器(预签名过期/签名失败不能误判会话过期);大 zip 给足超时。 */
const directPut = axios.create({ timeout: 600_000, withCredentials: false })
@@ -10,8 +10,6 @@ export const SOFTWARE_VERSION_UPLOAD_ENDPOINT = '/api/admin/version'
export const SOFTWARE_VERSION_PRESIGN_ENDPOINT = `${SOFTWARE_VERSION_UPLOAD_ENDPOINT}/presign`
export const SOFTWARE_VERSION_CONFIRM_ENDPOINT = `${SOFTWARE_VERSION_UPLOAD_ENDPOINT}/confirm`
export const SOFTWARE_VERSION_DELETE_ENDPOINT = `${SOFTWARE_VERSION_UPLOAD_ENDPOINT}/delete`
/** 大档分片通道(AC-11):>8MB 的安装包走这里,不再一次性 presign PUT。 */
export const SOFTWARE_VERSION_CHUNK_ENDPOINT = '/api/files/chunk'
/** MinIO 直传专用实例:不带 cookie、不挂 401 跳登录拦截器(预签名过期/签名失败不能误判会话过期);大 zip 给足超时。 */
const directPut = axios.create({ timeout: 600_000, withCredentials: false })
+1
View File
@@ -34,6 +34,7 @@ export const adminPages: AdminPageDef[] = [
{ path: 'records/tutorial', menuKey: 'admin_tutorial', title: '教程管理', companyScoped: false, load: () => import('@/pages/records/RecordsTutorialPage.vue') },
{ path: 'records/workbench-client-package', menuKey: 'admin_workbench_client_package', title: '数富AI工作台客户端包管理', companyScoped: false, load: () => import('@/pages/records/WorkbenchClientPackagePage.vue') },
{ path: 'records/device-logs', menuKey: 'admin_device_logs', title: '日志管理', companyScoped: true, load: () => import('@/pages/records/DeviceLogsPage.vue') },
{ path: 'records/task-evidence', menuKey: 'admin_task_evidence', title: '任务证据', superAdminOnly: true, companyScoped: false, load: () => import('@/pages/records/TaskEvidencePage.vue') },
{ path: 'task-center/board', menuKey: 'admin_task_center', title: '客户端任务管理', companyScoped: true, load: () => import('@/pages/task-center/TaskCenterBoardPage.vue') },
{ path: 'records/digital-human-version', menuKey: 'digital_human_version', title: '数字人版本管理', load: () => import('@/pages/records/RecordsDigitalHumanVersionPage.vue') },
{ path: 'records/image-video-tasks', menuKey: 'admin_image_video_tasks', title: '视频任务记录', companyScoped: true, load: () => import('@/pages/tasks/ImageVideoTasksPage.vue') },
@@ -1,10 +1,10 @@
{
"note": "readSource 钉子逐文件基线。两把尺子:pins=落在用例体内的读盘,reads=全文读点总数(含模块级夹具与 helper)。都只许减;pins 下降而 reads 没同步下降=把读盘挪出体内,不算改进。由 tests/source-pin-ratchet.test.ts 生成。",
"totals": {
"files": 251,
"cases": 1750,
"pins": 453,
"reads": 646
"files": 252,
"cases": 1753,
"pins": 456,
"reads": 651
},
"files": {
"admin-scaffold.test.ts": {
@@ -1247,6 +1247,11 @@
"pins": 3,
"reads": 5
},
"task-evidence-page.test.ts": {
"cases": 3,
"pins": 3,
"reads": 5
},
"timer-cleanup-pairing.test.ts": {
"cases": 5,
"pins": 3,
@@ -0,0 +1,155 @@
// P2-10(后管 AC/user-notice「只把中文给用户看」口径)。
//
// 评审实录:`RecordsSoftwareVersionPage.vue`/`RecordsTutorialPage.vue` 的 catch 直接写
// `error instanceof Error ? error.message : '上传失败'`,而 `src/api/http.ts` 的响应拦截器
// **不做 sanitize**(只挑 401/403/4011)⇒ `Broken pipe`、`Request failed with status code 403`
// 这类英文异常原文整段上屏。收口方式沿用本仓既有出口 `components/load-failure-model`
// 的 `loadFailureMessage(error, 本页兜底句)`(去重页四条写路径同一写法)。
//
// 跑法:npm run test:mount(tests/mount/**,vitest+jsdom)。
import { beforeEach, describe, expect, it, vi } from 'vitest'
import { flushPromises, mount } from '@vue/test-utils'
import { readFileSync } from 'node:fs'
import { dirname, resolve } from 'node:path'
import { fileURLToPath } from 'node:url'
import RecordsSoftwareVersionPage from '@/pages/records/RecordsSoftwareVersionPage.vue'
import RecordsTutorialPage from '@/pages/records/RecordsTutorialPage.vue'
import { loadFailureMessage } from '@/components/load-failure-model'
const here = dirname(fileURLToPath(import.meta.url))
const root = resolve(here, '../..')
const versionApi = vi.hoisted(() => ({
fetchSoftwareVersions: vi.fn(),
uploadSoftwareVersion: vi.fn(),
deleteSoftwareVersions: vi.fn(),
}))
const tutorialApi = vi.hoisted(() => ({
fetchTutorialPackages: vi.fn(),
uploadTutorialPackage: vi.fn(),
deleteTutorialPackages: vi.fn(),
}))
const confirmMock = vi.hoisted(() => vi.fn())
const feedbackMock = vi.hoisted(() => vi.fn())
const messageMock = vi.hoisted(() => ({ success: vi.fn() }))
vi.mock('@/pages/records/version-api.ts', () => versionApi)
vi.mock('@/pages/records/tutorial-api.ts', () => tutorialApi)
vi.mock('@/components/admin-confirm', () => ({ openAdminConfirm: confirmMock }))
vi.mock('@/components/admin-feedback-ui', () => ({ showAdminFeedback: feedbackMock }))
vi.mock('element-plus', () => ({ ElMessage: messageMock }))
/** el-upload 的桩:把页面传进来的 on-change 回调顶出来,等价于用户选了一个 .zip。 */
const uploadStub = {
props: ['onChange'],
template: '<button class="pick-zip" @click="onChange({ raw: fake })">选择文件</button>',
data: () => ({ fake: new File([new Uint8Array(16)], 'pkg.zip', { type: 'application/zip' }) }),
}
const stubs = {
'el-dialog': { template: '<div><slot /><slot name="footer" /></div>' },
'el-form': { template: '<form><slot /></form>' },
'el-form-item': { template: '<div><slot /></div>' },
'el-input': {
props: ['modelValue'],
emits: ['update:modelValue'],
template: '<input :value="modelValue" @input="$emit(\'update:modelValue\', $event.target.value)" />',
},
'el-upload': uploadStub,
'el-button': { template: '<button><slot /></button>' },
'el-progress': true,
'el-alert': { props: ['title'], template: '<div class="upload-msg">{{ title }}</div>' },
}
/** 后端/驱动层原文:一句中文都没有,正是过去直接上屏的那种。 */
const RAW_ENGLISH = 'Broken pipe: java.io.IOException at sun.nio.ch.FileDispatcherImpl.write0'
function clickButtonByText(wrapper: ReturnType<typeof mount>, text: string, last = false) {
const matches = wrapper.findAll('button').filter((node) => node.text().trim() === text)
const button = last ? matches[matches.length - 1] : matches[0]
if (!button) throw new Error(`找不到文案为「${text}」的按钮`)
return button
}
async function submitZipUpload(component: typeof RecordsSoftwareVersionPage, versionPlaceholder: string, submitText: string, openText = '上传新版本') {
const wrapper = mount(component, { global: { stubs } })
await flushPromises()
clickButtonByText(wrapper, openText)
const versionInput = wrapper.find(`input[placeholder="${versionPlaceholder}"]`)
if (versionInput.exists()) await versionInput.setValue('9.9.9')
await wrapper.get('.pick-zip').trigger('click')
await clickButtonByText(wrapper, submitText, true).trigger('click')
await flushPromises()
return wrapper
}
describe('P2-10 后管上传/加载失败只把中文交给界面', () => {
beforeEach(() => {
vi.clearAllMocks()
versionApi.fetchSoftwareVersions.mockResolvedValue({ items: [] })
tutorialApi.fetchTutorialPackages.mockResolvedValue({ items: [] })
confirmMock.mockResolvedValue(true)
})
it('真实出口对照:英文原文一律被换成中文,原因段不拼半截句', () => {
expect(loadFailureMessage(new Error(RAW_ENGLISH), '上传失败')).toBe('上传失败')
expect(loadFailureMessage(new Error('分片会话不存在或已过期'), '上传失败'))
.toBe('上传失败:分片会话不存在或已过期')
})
it('安装包上传失败:界面文案不含英文原文,原文只进 console', async () => {
const warned = vi.spyOn(console, 'warn').mockImplementation(() => undefined)
versionApi.uploadSoftwareVersion.mockRejectedValue(new Error(RAW_ENGLISH))
const wrapper = await submitZipUpload(RecordsSoftwareVersionPage, '例如:1.0.0', '上传软件版本')
const shown = wrapper.get('.upload-msg').text()
expect(shown, '英文异常原文仍在上屏').not.toContain('Broken pipe')
expect(shown).toContain('上传失败')
expect(
warned.mock.calls.some((call) => String(call.join(' ')).includes(RAW_ENGLISH)),
'原文要留在 console,否则支持同学对着后端日志无从核对',
).toBe(true)
warned.mockRestore()
})
it('教程包上传失败:同一条中文出口,不得再直落 error.message', async () => {
vi.spyOn(console, 'warn').mockImplementation(() => undefined)
tutorialApi.uploadTutorialPackage.mockRejectedValue(new Error(RAW_ENGLISH))
const wrapper = await submitZipUpload(RecordsTutorialPage, '例如:v2026.09 或留空', '上传教程包', '上传教程包')
const shown = wrapper.get('.upload-msg').text()
expect(shown, '英文异常原文仍在上屏').not.toContain('Broken pipe')
expect(shown).toContain('上传失败')
})
it('列表加载失败:LoadFailureBar 的文案也是中文', async () => {
vi.spyOn(console, 'warn').mockImplementation(() => undefined)
versionApi.fetchSoftwareVersions.mockRejectedValue(new Error('Internal Server Error'))
const wrapper = mount(RecordsSoftwareVersionPage, { global: { stubs } })
await flushPromises()
const shown = wrapper.text()
expect(shown).not.toContain('Internal Server Error')
expect(shown).toContain('版本列表加载失败')
expect(feedbackMock).toHaveBeenCalled()
})
// 写法收口的台账:两页四条 catch(列表/删除×2/上传)都必须把错误交给同一个出口。
it('两页不得再留 error instanceof Error ? error.message 那套写法', () => {
for (const file of [
'src/pages/records/RecordsSoftwareVersionPage.vue',
'src/pages/records/RecordsTutorialPage.vue',
]) {
const source = readFileSync(resolve(root, file), 'utf-8')
expect(source, `${file} 仍直落 error.message`).not.toMatch(/error instanceof Error \? error\.message/)
expect(source.match(/loadFailureMessage\(error,/g)?.length ?? 0, `${file} 的失败出口没接上`)
.toBeGreaterThanOrEqual(4)
expect(source).toMatch(/import \{ loadFailureMessage \} from '@\/components\/load-failure-model'/)
}
})
})
@@ -1,8 +1,13 @@
// AC-11(R4):后管版本/教程大档上传接线的「源码尺子」那一半。
//
// Pending Phase 4 激活。今天 `src/pages/records/version-api.ts` 与 `tutorial-api.ts`
// 都是 presign → 一次 PUT 直传(512MB 级也一样,蓝图 §1.2 R4 实读);本文件要求两条
// 路径接上既有 `/api/files/chunk/**` 分片通道(断点续传/按片重试)。
// Pending Phase 4 激活。本尺子的前身只断言 `files/chunk` 出现在非注释行,而那一句匹配来自
// `version-api.ts`/`tutorial-api.ts` 里一对**全仓零引用的死常量**(SOFTWARE_VERSION_CHUNK_ENDPOINT/
// TUTORIAL_CHUNK_ENDPOINT)——把真调用 `uploadFileInChunks(…)` 删掉,尺子照样绿(评审 P2-6 假绿形状)。
// 现在判据钉在**真调用点**上:被接线的那个函数体里必须出现 `uploadFileInChunks(`,
// 且死常量不许回来(`CHUNK_ENDPOINT =` 一出现即红)。删调用必判红,加死常量也判红。
//
// 行为口径在 tests/pure/pending-phase4-大档上传接线.test.ts(vitest 真调 uploadSoftwareVersion/
// uploadTutorialPackage,删掉调用同一刻那一边也红);本文件只补"接线没被搬空"这一层。
//
// 跑法:npm test(node --test tests/*.test.ts)。
import { test } from 'node:test'
@@ -14,29 +19,75 @@ import { fileURLToPath } from 'node:url'
const here = dirname(fileURLToPath(import.meta.url))
const root = resolve(here, '..')
const LARGE_UPLOAD_FILES = [
'src/pages/records/version-api.ts',
'src/pages/records/tutorial-api.ts',
]
/** 把注释整段涂白(保行数):接线只写在注释里不算数。 */
function blankComments(source: string): string {
return source
.replace(/\/\*[\s\S]*?\*\//g, (block) => block.replace(/[^\n]/g, ' '))
.replace(/\/\/[^\n]*/g, (line) => ' '.repeat(line.length))
}
test('后管版本/教程大档必须接分片通道', () => {
for (const file of LARGE_UPLOAD_FILES) {
const source = readFileSync(resolve(root, file), 'utf-8')
assert.match(
source,
/files\/chunk/,
`AC-11:${file} 仍是 presign 直传(512MB 断网从 0 重来,没有分片续传)`,
)
/** 取 `function <name>(…) {…}` 的函数体:大括号配对,跳过字符串与模板串里的括号。 */
function functionBodyOf(source: string, name: string): string {
const code = blankComments(source)
const head = code.indexOf(`function ${name}(`)
assert.notEqual(head, -1, `找不到函数 ${name}()——接线被删了还是改名了?`)
const open = code.indexOf('{', head)
assert.notEqual(open, -1, `${name}() 没有函数体`)
let depth = 0
let index = open
let quote = ''
for (; index < code.length; index += 1) {
const char = code[index]
if (quote) {
if (char === '\\') index += 1
else if (char === quote) quote = ''
continue
}
if (char === '"' || char === "'" || char === '`') { quote = char; continue }
if (char === '{') depth += 1
else if (char === '}') {
depth -= 1
if (depth === 0) return code.slice(open + 1, index)
}
}
return code.slice(open + 1)
}
/**
* 两格共同判据(读盘点留在用例体内:把 readFileSync 挪进 helper 是本仓 source-pin 棘轮点名的"搬家不算改进"):
* ① 从 @/api/chunk-upload 把分片函数引进来;② 被接线的函数体里真的调它;
* ③ 阈值同源;④ 死常量不得回来。
*/
function assertChunkedWiring(source: string, file: string, wiredFunction: string): void {
assert.match(
source,
/import\s*\{[^}]*uploadFileInChunks[^}]*\}\s*from\s*'@\/api\/chunk-upload'/,
`AC-11:${file} 没有从 @/api/chunk-upload 引入分片函数`,
)
const body = functionBodyOf(source, wiredFunction)
assert.match(
body,
/uploadFileInChunks\s*\(/,
`AC-11:${file}::${wiredFunction}() 里没有真调用 uploadFileInChunks(…)(大档仍走一次 PUT 直传,512MB 断网从 0 重来)`,
)
assert.match(
body,
/CHUNK_UPLOAD_THRESHOLD/,
`AC-11:${wiredFunction}() 的选路阈值必须与 chunk-upload 同源,不得各自写死 8MB`,
)
assert.doesNotMatch(
source,
/CHUNK_ENDPOINT\s*=/,
`AC-11:${file} 又立了一个没人引用的 CHUNK_ENDPOINT 常量——它会顶掉本尺子的匹配来源,正是 P2-6 的假绿形状`,
)
}
test('后管安装包上传必须真的调分片通道', () => {
const file = 'src/pages/records/version-api.ts'
assertChunkedWiring(readFileSync(resolve(root, file), 'utf-8'), file, 'uploadSoftwareVersion')
})
test('分片接线不得只落注释', () => {
for (const file of LARGE_UPLOAD_FILES) {
const source = readFileSync(resolve(root, file), 'utf-8')
const codeLines = source
.split('\n')
.filter((line) => !line.trim().startsWith('//') && !line.trim().startsWith('*'))
.join('\n')
assert.match(codeLines, /files\/chunk/, `AC-11:${file} 的分片引用只出现在注释里`)
}
test('后管教程包上传必须真的调分片通道', () => {
const file = 'src/pages/records/tutorial-api.ts'
assertChunkedWiring(readFileSync(resolve(root, file), 'utf-8'), file, 'uploadTutorialPackage')
})
@@ -0,0 +1,226 @@
// P2-7/P2-8/P3-5(后管分片通道 AC-9/AC-11 精神)。
//
// 评审实录:`src/api/chunk-upload.ts` 每次调用都新 init 一个 upload_id ⇒ 紧随的 GET 必空、
// 放弃时又把会话 DELETE 掉,**等于根本没有续传**;逐片超时还按整包体积算(120s + MB×15s),
// 8MB 的片能拿到约 130 分钟。本文件按用户端 `shared/api/upload.ts` 的同一口径钉四件事:
// ① 同一文件身份复用 upload_id(不重新 init);② 瞬态失败不 DELETE、会话留在本地供下次续传;
// ③ 只有服务端确实说会话没了(404/「不存在或已过期」)才重开;④ 逐片超时按本片体积给。
//
// 跑法:npm run test:mount(tests/pure/** 与 tests/mount/** 同走 vitest)。
import { beforeEach, describe, expect, it, vi } from 'vitest'
const mocks = vi.hoisted(() => ({
post: vi.fn(),
get: vi.fn(),
delete: vi.fn(),
}))
// 只顶掉 http 实例,unwrap 用真的:会话回收在后端是 HTTP 200+success:false,
// 判因逻辑要吃真实解包语义(含中文文案把关),换桩会把这一半形状测漏。
vi.mock('@/api/http', async (importOriginal) => {
const actual = await importOriginal<{ unwrap: <T>(payload: unknown) => T }>()
return {
...actual,
http: { post: mocks.post, get: mocks.get, delete: mocks.delete },
}
})
import {
CHUNK_UPLOAD_PART_SIZE,
CHUNK_UPLOAD_THRESHOLD,
chunkSessionKeyOf,
partTimeoutOf,
uploadFileInChunks,
} from '@/api/chunk-upload'
const RESUMED_ID = 'r'.repeat(32)
const FRESH_ID = 'n'.repeat(32)
function axiosResponse(status: number, body: unknown) {
return { status, data: body }
}
function javaOk(body: unknown) {
return axiosResponse(200, { success: true, message: '', data: body })
}
function makeFile(name: string, bytes: number): File {
return new File([new Uint8Array(bytes)], name, { type: 'application/octet-stream' })
}
function seedSession(file: File, uploadId: string): void {
localStorage.setItem(chunkSessionKeyOf(file.name, file), JSON.stringify({
uploadId,
size: file.size,
lastModified: file.lastModified,
}))
}
function storedUploadId(file: File): string {
const raw = localStorage.getItem(chunkSessionKeyOf(file.name, file))
if (!raw) return ''
return String((JSON.parse(raw) as { uploadId?: unknown }).uploadId ?? '')
}
function postedUrls(): string[] {
return mocks.post.mock.calls.map(([url]) => String(url))
}
function partPosts(): Array<{ url: string; timeout: number }> {
return mocks.post.mock.calls
.filter(([url]) => /\/part\/\d+$/.test(String(url)))
.map(([url, , config]) => ({
url: String(url),
timeout: Number((config as { timeout?: unknown } | undefined)?.timeout ?? -1),
}))
}
function initCount(): number {
return postedUrls().filter((url) => url.includes('/chunk/init')).length
}
/** init/逐片/complete 一律成功;complete 返回拼装后的 fileKey。 */
function routePosts(uploadId: string) {
mocks.post.mockImplementation((url: string) => Promise.resolve(
String(url).includes('/chunk/init')
? javaOk({ upload_id: uploadId, total_parts: 2 })
: String(url).includes('/complete')
? javaOk({ fileKey: 'tmp/pkg.zip', originalFilename: 'pkg.zip', size: 1 })
: javaOk({ upload_id: String(url).split('/')[5], received_parts: 1 }),
))
}
describe('P2-7 后管分片必须真的续传', () => {
beforeEach(() => {
vi.clearAllMocks()
localStorage.clear()
routePosts(FRESH_ID)
})
it('同一文件身份复用 upload_id:探测成功就一条 init 都不许发', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(javaOk({ upload_id: RESUMED_ID, uploaded_parts: [0] }))
await uploadFileInChunks(file, file.name)
expect(initCount(), '会话还在服务端,重开 init 就是作废已传片').toBe(0)
expect(
partPosts().map((part) => part.url),
'第 0 片服务端已有,只补第 1 片',
).toEqual([`/api/files/chunk/${RESUMED_ID}/part/1`])
})
it('瞬态失败(分片打不出去):不 DELETE 会话,且本地仍留着 upload_id', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(javaOk({ upload_id: RESUMED_ID, uploaded_parts: [] }))
mocks.post.mockRejectedValue(new Error('Network Error'))
await expect(uploadFileInChunks(file, file.name)).rejects.toThrow(/Network Error/)
expect(mocks.delete.mock.calls, 'DELETE 会把已达片号一起清掉,下次只能从 0').length(0)
expect(storedUploadId(file), '会话要留给下一次(含刷新后)续传').toBe(RESUMED_ID)
})
it('探测 503:瞬态 ⇒ 保留原会话,不重开', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(axiosResponse(503, '<html>Bad Gateway</html>'))
await uploadFileInChunks(file, file.name)
expect(initCount()).toBe(0)
expect(partPosts().every((part) => part.url.includes(RESUMED_ID))).toBe(true)
})
it('探测超时(网络层错误):瞬态 ⇒ 同样保留原会话', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockRejectedValue(new Error('timeout of 30000ms exceeded'))
await uploadFileInChunks(file, file.name)
expect(initCount()).toBe(0)
expect(partPosts().every((part) => part.url.includes(RESUMED_ID))).toBe(true)
})
it('HTTP 404:会话确实没了 ⇒ 清状态重开', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(axiosResponse(404, { success: false, message: 'Not Found', data: null }))
await uploadFileInChunks(file, file.name)
expect(initCount(), '会话已回收,不重建就永远卡在同一个死会话上').toBe(1)
expect(partPosts().every((part) => part.url.includes(FRESH_ID))).toBe(true)
})
it('后端 HTTP 200+「分片会话不存在或已过期」(现网形状)⇒ 同样重开', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(axiosResponse(
200,
{ success: false, message: '分片会话不存在或已过期', data: null },
))
await uploadFileInChunks(file, file.name)
expect(initCount()).toBe(1)
expect(partPosts().every((part) => part.url.includes(FRESH_ID))).toBe(true)
})
it('拼装成功 ⇒ 本地会话清掉(同一文件再传要开新会话)', async () => {
const file = makeFile('pkg.zip', CHUNK_UPLOAD_PART_SIZE * 2)
seedSession(file, RESUMED_ID)
mocks.get.mockResolvedValue(javaOk({ upload_id: RESUMED_ID, uploaded_parts: [] }))
await uploadFileInChunks(file, file.name)
expect(storedUploadId(file), '完成后会话已拼装落库,本地不该留残号').toBe('')
})
})
describe('P2-8 逐片超时按本片体积', () => {
it('每片预算只看这一片有多大,不得按整包体积放大', async () => {
const file = makeFile('big.zip', CHUNK_UPLOAD_PART_SIZE * 5)
mocks.get.mockResolvedValue(javaOk({ upload_id: FRESH_ID, uploaded_parts: [] }))
routePosts(FRESH_ID)
await uploadFileInChunks(file, file.name)
const parts = partPosts()
expect(parts.length, '40MB=5 片').toBe(5)
// 整包口径会给出 120s + 40×15s = 720s;按片口径 8MB 片=300s。判红要留出余量差。
for (const part of parts) {
expect(part.timeout, `片 ${part.url} 的超时没有按本片体积算`).toBeGreaterThan(0)
expect(
part.timeout,
`片 ${part.url} 的超时 ${part.timeout}ms 是按整包体积算的(P2-8)`,
).toBeLessThanOrEqual(partTimeoutOf(CHUNK_UPLOAD_PART_SIZE))
}
expect(partTimeoutOf(CHUNK_UPLOAD_PART_SIZE), '8MB 片的预算不该超过 10 分钟')
.toBeLessThanOrEqual(600_000)
expect(partTimeoutOf(1), '再小的片也要有保底,不能给 0').toBeGreaterThanOrEqual(60_000)
})
it('init 与 complete 的预算不再随整包体积线性放大', async () => {
const file = makeFile('big.zip', CHUNK_UPLOAD_PART_SIZE * 5)
mocks.get.mockResolvedValue(javaOk({ upload_id: FRESH_ID, uploaded_parts: [] }))
routePosts(FRESH_ID)
await uploadFileInChunks(file, file.name)
const initCall = mocks.post.mock.calls.find(([url]) => String(url).includes('/chunk/init'))
const initTimeout = Number((initCall?.[2] as { timeout?: unknown } | undefined)?.timeout ?? -1)
expect(initTimeout, 'init 只发一条登记请求,按整包体积给 720s 是把失败拖到天荒地老').toBeLessThan(300_000)
})
})
describe('P3-5 阈值单点', () => {
it('后管这一侧的 8MB 只有一个出口:阈值由片大小派生', () => {
expect(CHUNK_UPLOAD_THRESHOLD, '阈值必须与片大小同源(P3-5 三处各写一遍的收口)')
.toBe(CHUNK_UPLOAD_PART_SIZE)
expect(CHUNK_UPLOAD_PART_SIZE).toBe(8 * 1024 * 1024)
})
})
@@ -3,7 +3,7 @@
// Pending Phase 4 激活。9MB 安装包(跨分片阈值)调 uploadSoftwareVersion,
// 必须出现 POST /files/chunk/init;今天它走 presign → 一次 PUT 直传,
// 512MB 断网只能从 0 重来。本文件不出网,mock 掉 http 与 axios 直传实例。
import { describe, it, expect, vi } from 'vitest'
import { beforeEach, describe, it, expect, vi } from 'vitest'
const mocks = vi.hoisted(() => ({
post: vi.fn(),
@@ -34,9 +34,16 @@ vi.mock('@/api/http', () => ({
}))
import { uploadSoftwareVersion } from '@/pages/records/version-api.ts'
import { uploadTutorialPackage } from '@/pages/records/tutorial-api.ts'
describe('AC-11 后管大档上传', () => {
// 会话跨尝试复用后,两格之间必须清态:残留的 upload_id 会让第二格跳过本该发生的 init。
beforeEach(() => {
vi.clearAllMocks()
localStorage.clear()
})
it('9MB 安装包必须走分片 init', async () => {
// 预签名与分片两套响应揉成一个超集:哪条实现路径先走都能拿到自己要的字段。
mocks.post.mockResolvedValue({
@@ -68,5 +75,48 @@ describe('AC-11 后管大档上传', () => {
urls.some((url) => url.includes('/files/chunk/init')),
`AC-11:9MB 安装包仍走 presign 直传,实际出口:${urls.join(', ')}`,
).toBe(true)
// P2-6:删掉 uploadFileInChunks(…) 真调用后,这一格必须跟着红——所以还要核对分片片确实发出去了。
expect(
mocks.post.mock.calls.filter(([url]) => /\/files\/chunk\/[^/]+\/part\/\d+$/.test(String(url))).length,
'AC-11:走分片就要真把片 POST 出去',
).toBe(2)
})
// P2-6:教程包那条道过去只有源码尺子覆盖(而尺子当时吃的是死常量),删掉真调用没人判红。
it('9MB 教程包必须走分片 init', async () => {
localStorage.clear()
mocks.post.mockResolvedValue({
data: {
success: true,
data: {
upload_url: 'https://oss.example/put',
file_url: 'https://oss.example/tutorial.zip',
version: '9.9.9',
object_key: 'client/tutorials/tutorial.zip',
upload_id: 'b'.repeat(32),
uploaded_parts: [],
item: null,
},
},
})
mocks.get.mockResolvedValue({
data: { success: true, data: { upload_id: 'b'.repeat(32), uploaded_parts: [] } },
})
mocks.directPut.mockResolvedValue({})
const file = new Blob([new Uint8Array(9 * 1024 * 1024)], {
type: 'application/octet-stream',
})
await uploadTutorialPackage(file, 'tutorial.zip', '9.9.9')
const urls = mocks.post.mock.calls.map(([url]) => String(url))
expect(
urls.some((url) => url.includes('/files/chunk/init')),
`AC-11:9MB 教程包仍走 presign 直传,实际出口:${urls.join(', ')}`,
).toBe(true)
expect(
mocks.directPut.mock.calls.length,
'AC-11:大档不该再走一次 PUT 直传',
).toBe(0)
})
})
@@ -0,0 +1,32 @@
import test from 'node:test'
import assert from 'node:assert/strict'
import { readSource } from './helpers.ts'
/** P1-9:任务证据读取面的后台入口必须存在,且只给超级管理员。 */
test('task evidence menu is super-admin only', () => {
const routes = readSource('src/router/routes.ts')
assert.match(routes, /path: 'records\/task-evidence'[\s\S]*?menuKey: 'admin_task_evidence'[\s\S]*?superAdminOnly: true/,
'任务证据路由必须带 superAdminOnly')
const scope = readSource('src/config/company-scope.ts')
assert.match(scope, /SUPER_ADMIN_ONLY_MENU_KEYS[^\n]*admin_task_evidence/,
'菜单 key 必须在超级管理员可见清单里')
const guide = readSource('src/layout/operation-guides.ts')
assert.match(guide, /admin_task_evidence:\s*\{/, '菜单必须配操作指引')
})
test('task evidence page never renders object keys', () => {
const page = readSource('src/pages/records/TaskEvidencePage.vue')
assert.doesNotMatch(page, /objectKey|object_key|presign/i, '页面不得展示或持有对象直链')
assert.match(page, /taskEvidenceDownloadUrl/, '下载必须走鉴权接口')
assert.match(page, /deleteTaskEvidence/, '删除必须走鉴权接口')
})
test('task evidence api is wired to admin endpoints', () => {
const api = readSource('src/api/task-evidence.ts')
assert.match(api, /\/admin\/task-evidence\/files/)
assert.match(api, /\/admin\/task-evidence\/download\?fileId=/)
assert.match(api, /\/admin\/task-evidence\/delete/)
})
@@ -0,0 +1,86 @@
package com.nanri.aiimage.common.security;
import java.io.File;
import java.io.IOException;
import java.util.Enumeration;
import java.util.List;
import java.util.Locale;
import java.util.Set;
import java.util.zip.ZipEntry;
import java.util.zip.ZipFile;
/**
* 安装包/教程包凭据件扫描(P2-4,对齐工作台包的 CR-02 口径)。
*
* <p>版本包与教程包此前只校验扩展名和体积,压缩包里是否带 {@code .env}/{@code *.pem}
* 没人看;而它们会经公开 client 桶发给全部客户端。这里只读 zip 中央目录、按条目名判,
* 不解压内容,代价与包体积无关。</p>
*/
public final class CredentialPackageGuard {
private static final Set<String> FORBIDDEN_NAMES = Set.of(
"kw_share_cfg.json", ".env", ".netrc", ".git-credentials",
"id_rsa", "id_dsa", "id_ecdsa", "id_ed25519");
private static final List<String> FORBIDDEN_SUFFIXES =
List.of(".pem", ".key", ".p12", ".pfx", ".jks", ".keystore");
private static final int MAX_ENTRIES = 200_000;
private CredentialPackageGuard() {
}
/**
* 返回第一个命中凭据名的条目;没有命中返回 {@code null}。
*
* @throws IOException 文件不是可读 zip / 条目数超限
*/
public static String firstForbiddenEntry(File packageFile) throws IOException {
if (packageFile == null || !packageFile.isFile()) {
throw new IOException("package file missing");
}
try (ZipFile zipFile = new ZipFile(packageFile)) {
Enumeration<? extends ZipEntry> entries = zipFile.entries();
int scanned = 0;
while (entries.hasMoreElements()) {
ZipEntry entry = entries.nextElement();
if (++scanned > MAX_ENTRIES) {
throw new IOException("zip entry limit exceeded");
}
if (entry.isDirectory()) {
continue;
}
String base = baseName(entry.getName());
if (isCredentialFileName(base)) {
return entry.getName();
}
}
}
return null;
}
static boolean isCredentialFileName(String fileName) {
String lower = fileName == null ? "" : fileName.trim().toLowerCase(Locale.ROOT);
if (lower.isEmpty()) {
return false;
}
if (FORBIDDEN_NAMES.contains(lower) || lower.startsWith(".env.")) {
return true;
}
for (String suffix : FORBIDDEN_SUFFIXES) {
if (lower.endsWith(suffix)) {
return true;
}
}
return false;
}
private static String baseName(String entryName) {
if (entryName == null) {
return "";
}
String normalized = entryName.replace('\\', '/');
int slash = normalized.lastIndexOf('/');
return slash < 0 ? normalized : normalized.substring(slash + 1);
}
}
@@ -59,7 +59,7 @@ public class AppearancePatentController {
}
@GetMapping("/tasks/{taskId}/queue-payload")
@Operation(summary = "获取外观专利 Python 队列载荷", description = "本地采集模式下,前端取出 groups 后推本机客户端队列;只返回 Python 当前需要的分组,避免前端搬运完整解析大数据。")
@Operation(summary = "获取外观专利 Python 队列载荷", description = "本地采集模式下,前端取出 groups 后推本机客户端队列;只返回 Python 当前需要的分组,避免前端搬运完整解析大数据。这一单登记为线上采集时拒发(V183 通道),防旧页面缓存推本机与后台双跑。")
public ApiResponse<AppearancePatentParsedPayloadDto> queuePayload(
@Parameter(description = "外观专利任务 ID", required = true, example = "7004")
@PathVariable Long taskId,
@@ -67,7 +67,7 @@ public class AppearancePatentController {
@RequestParam("user_id") Long userId,
HttpServletRequest httpRequest) {
userId = adminAuthSupport.resolveOwnershipUserId(httpRequest, userId);
return ApiResponse.success(service.queuePayload(taskId, userId));
return ApiResponse.success(service.queuePayloadForLocalClient(taskId, userId));
}
@GetMapping("/dashboard")
@@ -114,13 +114,6 @@ public class AppearancePatentTaskService extends AbstractFileTaskPipeline<Appear
@org.springframework.beans.factory.annotation.Autowired(required = false)
private com.nanri.aiimage.modules.usersecret.service.UserTaskGuardService taskGuard;
/**
* 采集路由开关:字段注入的理由同 {@code taskGuard}——本类构造器被 golden 用例按参数顺序
* new 出来,加参数会连带改一批测试。null=未注入(单测)时按「非本地」处理,退回后台语义。
*/
@org.springframework.beans.factory.annotation.Autowired(required = false)
private com.nanri.aiimage.config.CollectRoutingProperties collectRoutingProperties;
/**
* 参数顺序=迁移前 {@code @RequiredArgsConstructor} 的字段声明顺序,一字未动;
* 其中五枚(fileTaskMapper/fileResultMapper/taskDistributedLockService/
@@ -218,10 +211,10 @@ public class AppearancePatentTaskService extends AbstractFileTaskPipeline<Appear
FileTaskEntity channelTask = requireSubmissionTask(taskId);
if (com.nanri.aiimage.modules.task.model.CollectChannel
.resolve(channelTask.getCollectChannel(), channelTask.getModuleType()).isLocal()) {
activateForLocalClient(taskId, userId);
activateForLocalClient(channelTask, userId);
return;
}
recordSupport().activateTask(taskId, userId);
recordSupport().activateTask(channelTask, userId);
}
/**
@@ -230,10 +223,9 @@ public class AppearancePatentTaskService extends AbstractFileTaskPipeline<Appear
* <p>与客户端兜底拉取 / Linux 内部认领共用同一条原子翻转,谁先翻谁执行,
* 页面推送与后台接管不会重复跑同一份任务。
*/
private void activateForLocalClient(Long taskId, Long userId) {
// 模块归属走底座 requireSubmissionTask(TaskFinalizeSupport 家族),不在本类手抄
// 「!MODULE_TYPE.equals(...)」——那形已被 TaskLifecycleDebtRatchetTest 钉成只降不升。
FileTaskEntity task = requireSubmissionTask(taskId);
private void activateForLocalClient(FileTaskEntity task, Long userId) {
// 模块归属已在 activateTask 的首读校验过;这里复用同一行,不再二次 selectById。
Long taskId = task.getId();
if (userId == null || !java.util.Objects.equals(userId, task.getUserId())) {
throw new com.nanri.aiimage.common.exception.BusinessException("任务不存在");
}
@@ -793,6 +785,11 @@ public class AppearancePatentTaskService extends AbstractFileTaskPipeline<Appear
return payloadSupport().queuePayload(taskId, userId);
}
/** 页面/直连 API 取本机队列载荷:本单通道不是本机时拒发(P1-1 同族)。 */
public AppearancePatentParsedPayloadDto queuePayloadForLocalClient(Long taskId, Long userId) {
return payloadSupport().queuePayloadForLocalClient(taskId, userId);
}
static String resolveTaskExecutionStatus(boolean waitingForAssemble, boolean executionFailed) {
if (waitingForAssemble) {
return STATUS_RUNNING;
@@ -11,6 +11,7 @@ import com.nanri.aiimage.modules.appearancepatent.model.vo.AppearancePatentParse
import com.nanri.aiimage.modules.task.mapper.FileTaskMapper;
import com.nanri.aiimage.modules.task.mapper.TaskChunkMapper;
import com.nanri.aiimage.modules.task.mapper.TaskScopeStateMapper;
import com.nanri.aiimage.modules.task.model.CollectChannel;
import com.nanri.aiimage.modules.task.model.entity.FileTaskEntity;
import com.nanri.aiimage.modules.task.model.entity.TaskChunkEntity;
import com.nanri.aiimage.modules.task.model.entity.TaskScopeStateEntity;
@@ -153,11 +154,31 @@ public class AppearancePatentPayloadSupport {
}
public AppearancePatentParsedPayloadDto queuePayload(Long taskId, Long userId) {
return queuePayload(taskId, userId, false);
}
/**
* 页面/直连 API 侧的下发口:只有这一单确实交给本机客户端时才给载荷。
*
* <p>P1-1 的同族洞:取载荷这一侧原先只看存在与归属、不看通道,选了线上的单被旧页面缓存
* 或直连 API 取走 groups 推本机,就与 Linux 那边同时跑同一组(按次付费的出口 IP 买两遍)。
* 兜底拉取/内部认领走的是 {@link #queuePayload(Long, Long)}:本机单等满兜底窗口后由
* Linux 接管,那一侧必须仍能拿到载荷。
*/
public AppearancePatentParsedPayloadDto queuePayloadForLocalClient(Long taskId, Long userId) {
return queuePayload(taskId, userId, true);
}
private AppearancePatentParsedPayloadDto queuePayload(Long taskId, Long userId, boolean requireLocalChannel) {
FileTaskEntity task = fileTaskMapper.selectById(taskId);
if (task == null || !moduleType.equals(task.getModuleType()) || !Objects.equals(userId, task.getUserId())) {
throw new BusinessException("任务不存在");
}
ownershipChecker.accept(task, "queue payload");
if (requireLocalChannel
&& !CollectChannel.resolve(task.getCollectChannel(), task.getModuleType()).isLocal()) {
throw new BusinessException("本单已选择线上采集,无需打开客户端,提交后稍候自动开始");
}
AppearancePatentParsedPayloadDto payload = readParsedPayload(task);
AppearancePatentParsedPayloadDto queuePayload = new AppearancePatentParsedPayloadDto();
queuePayload.setAiPrompt(payload.getAiPrompt());
@@ -70,18 +70,20 @@ public class AppearancePatentTaskRecordSupport {
}
public void activateTask(Long taskId, Long userId) {
FileTaskEntity task = fileTaskMapper.selectById(taskId);
activateTask(fileTaskMapper.selectById(taskId), userId);
}
/** P3-4:调用方已经读过任务行时复用,不在 activate 链路上重复 selectById。 */
public void activateTask(FileTaskEntity task, Long userId) {
Long taskId = task == null ? null : task.getId();
if (task == null || !moduleType.equals(task.getModuleType()) || !Objects.equals(userId, task.getUserId())) {
throw new BusinessException("任务不存在");
}
// 2026-09-30 起外观专利已整体迁移 Linux 后台执行(无浏览器抓取):
// activate 不再把任务推给客户端浏览器。任务保持 PENDING,由内部认领端点
// (/api/internal/tasks/pull-pending + maixiang-ap worker)在等待窗口后自动接管。
// 这里保留接口仅做状态校验:旧前端(未刷新页面缓存)点「启动」时给出明确提示,
// 前端收到业务异常会中止入队逻辑,任务不会被执行两次。
// V183 起执行方按单判定:本机单由客户端领取,线上单等内部认领(Linux/外机 Windows)。
// 这条兼容分支只服务未刷新页面缓存的旧前端:选中线上采集时,告诉用户无需打开浏览器。
if (STATUS_PENDING.equals(task.getStatus())) {
log.info("[外观专利] activate 不再推送客户端,任务保持 PENDING 等待 Linux 后台认领 taskId={} userId={}", taskId, userId);
throw new BusinessException("外观专利已切换后台自动执行,无需打开浏览器,稍候自动开始");
log.info("[外观专利] activate 不再推送客户端,任务保持 PENDING 等待线上认领 taskId={} userId={}", taskId, userId);
throw new BusinessException("本单已选择线上采集,无需打开客户端,提交后稍候自动开始");
}
if (STATUS_RUNNING.equals(task.getStatus())) {
throw new BusinessException("任务已在执行中(可能已由后台自动接管),无需重复启动");
@@ -78,7 +78,6 @@ public class AdminTaskEvidenceController {
@Parameter(description = "证据主键") @RequestParam Long fileId) throws IOException {
requireSuperAdmin(request);
TaskEvidenceEntity entity = taskEvidenceService.requireEvidence(fileId);
byte[] bytes = taskEvidenceService.download(fileId);
String name = "evidence-" + entity.getId()
+ (entity.getObjectKey() != null && entity.getObjectKey().contains(".")
? entity.getObjectKey().substring(entity.getObjectKey().lastIndexOf('.')) : ".bin");
@@ -86,7 +85,7 @@ public class AdminTaskEvidenceController {
? "application/octet-stream" : entity.getContentType());
response.setHeader("Content-Disposition", "attachment; filename*=UTF-8''"
+ URLEncoder.encode(name, StandardCharsets.UTF_8).replace("+", "%20"));
response.getOutputStream().write(bytes);
taskEvidenceService.downloadTo(fileId, response.getOutputStream());
}
@PostMapping("/delete")
@@ -47,9 +47,13 @@ public class InternalTaskEvidenceController {
if (body == null) {
throw new BusinessException("登记载荷为空");
}
// P1-2:归属以会话身份为准。内部令牌是发给每个登录客户的共享凭据,body.uid 谁都能写——
// 真身份只能来自请求头里的用户 JWT(4.0.14 起 user_token 钩子统一注入);会话 uid 与
// body.uid 不一致直接 403,无身份的存量装机保留原参数并打节流的匿名兜底日志。
Long uid = adminAuthSupport.resolveOwnershipUserId(request, body.getUid());
Map<String, Object> result = taskEvidenceService.register(
body.getTaskId(), body.getUid(), body.getEvidenceType(), body.getSource(), body.getFileKey());
log.info("[task-evidence] registered id={} taskId={}", result.get("id"), body.getTaskId());
body.getTaskId(), uid, body.getEvidenceType(), body.getSource(), body.getFileKey());
log.info("[task-evidence] registered id={} taskId={} uid={}", result.get("id"), body.getTaskId(), uid);
return ApiResponse.success(result);
}
}
@@ -3,11 +3,15 @@ package com.nanri.aiimage.modules.evidence.service;
import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
import com.baomidou.mybatisplus.extension.plugins.pagination.Page;
import com.nanri.aiimage.common.exception.BusinessException;
import com.nanri.aiimage.common.service.DistributedJobLockService;
import com.nanri.aiimage.common.task.LockSkipWitness;
import com.nanri.aiimage.config.EvidenceProperties;
import com.nanri.aiimage.modules.evidence.mapper.TaskEvidenceMapper;
import com.nanri.aiimage.modules.evidence.model.entity.TaskEvidenceEntity;
import com.nanri.aiimage.modules.evidence.storage.TaskEvidenceStorageService;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
import com.nanri.aiimage.modules.task.mapper.FileTaskMapper;
import com.nanri.aiimage.modules.task.model.entity.FileTaskEntity;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.scheduling.annotation.Scheduled;
@@ -15,11 +19,15 @@ import org.springframework.stereotype.Service;
import java.io.File;
import java.io.IOException;
import java.io.OutputStream;
import java.nio.file.Files;
import java.time.Duration;
import java.time.LocalDateTime;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Locale;
import java.util.Map;
import java.util.Set;
/**
* 证据登记与读取(AC-7/AC-13)。
@@ -35,11 +43,23 @@ public class TaskEvidenceService {
private static final int MAX_TASK_ID_LENGTH = 64;
private static final int MAX_CONTENT_BYTES = 8 * 1024 * 1024;
private static final int RETENTION_BATCH = 200;
/** 单轮保留期最多扫的批次数:防一个删不掉的对象把进程拖成永动机。 */
private static final int MAX_RETENTION_BATCHES = 20;
/** 同一任务的证据件数上限:登记端点不设防的话,一个持证客户端能刷满整个证据桶(P2-2)。 */
private static final int MAX_EVIDENCE_PER_TASK = 50;
/** 保留期回收的分布式锁名:A/B 两台每晚各扫同一张表会互相删(同族口径见 DeviceLogRetentionService)。 */
static final String RETENTION_LOCK = "task-evidence:retention";
private static final Set<String> ALLOWED_EVIDENCE_TYPES =
Set.of("client", "crash", "window_image", "page_image", "brand", "other");
private static final Set<String> ALLOWED_SOURCES =
Set.of("client", "crash_recorder", "window", "page", "brand", "other");
private final TaskEvidenceMapper mapper;
private final TaskEvidenceStorageService storage;
private final LocalFileStorageService localFileStorageService;
private final EvidenceProperties properties;
private final FileTaskMapper fileTaskMapper;
private final DistributedJobLockService distributedJobLockService;
/** 客户端上传完成后登记:fileKey 指向 Java 临时目录里的分片拼装件。 */
public Map<String, Object> register(String taskId, Long uid, String evidenceType,
@@ -52,6 +72,15 @@ public class TaskEvidenceService {
if (file == null || !file.isFile()) {
throw new BusinessException("证据文件不存在或已过期");
}
// P1-2:归属先于落桶。task_id 能对上库里的任务行时,uid 必须就是行主——判据发生在
// storage.store 之前,否则每一次拒绝都往证据桶里丢一个收不回的孤儿对象。
assertTaskOwnedBy(normalizedTaskId, uid);
if (file.length() == 0) {
throw new BusinessException("证据件为空,拒绝登记");
}
if (countByTask(normalizedTaskId) >= MAX_EVIDENCE_PER_TASK) {
throw new BusinessException("该任务的证据件数已达上限");
}
TaskEvidenceStorageService.StoredEvidence stored;
try {
stored = storage.store(file, normalizedTaskId, file.getName());
@@ -63,8 +92,8 @@ public class TaskEvidenceService {
TaskEvidenceEntity entity = new TaskEvidenceEntity();
entity.setTaskId(normalizedTaskId);
entity.setUid(uid);
entity.setEvidenceType(normalizeLabel(evidenceType, "client"));
entity.setSource(normalizeLabel(source, "client"));
entity.setEvidenceType(normalizeLabel(evidenceType, "client", ALLOWED_EVIDENCE_TYPES));
entity.setSource(normalizeLabel(source, "client", ALLOWED_SOURCES));
entity.setObjectKey(stored.objectKey());
entity.setByteSize(stored.byteSize());
entity.setContentType(stored.contentType());
@@ -143,6 +172,12 @@ public class TaskEvidenceService {
return storage.read(entity.getObjectKey(), properties.maxFileBytesOrDefault());
}
/** P3-1:后管下载走流式,避免 64MB 级对象整块进堆。 */
public void downloadTo(Long fileId, OutputStream out) {
TaskEvidenceEntity entity = requireEvidence(fileId);
storage.downloadTo(entity.getObjectKey(), out, properties.maxFileBytesOrDefault());
}
/** 删除证据:先删对象再删登记行;对象删除失败不吞(留在库里下一轮可重试)。 */
public int delete(List<Long> ids) {
if (ids == null || ids.isEmpty()) {
@@ -160,35 +195,104 @@ public class TaskEvidenceService {
return deleted;
}
/** 保留期回收:按天清扫到期行与对象(AC-13 的保留期必须有执行者,不能只写在配置里)。 */
/**
* 保留期回收:按天清扫到期行与对象(AC-13 的保留期必须有执行者,不能只写在配置里)。
*
* <p>P1-7 的两条修正:① A/B 两台每晚并发扫同一张表,回收必须拿分布式锁(同族口径见
* {@code DeviceLogRetentionService},锁跳过走可聚合出口);② 对象删除失败不再让整轮早退——
* 原先翻页永远停在第 1 页,卡住的那一行把后面每一行都堵死,配置里的 retention-days
* 因此只是个数字。改成按 id 键集分页:失败行留在库里下一轮重试,后面的行照常推进。
*/
@Scheduled(cron = "${aiimage.evidence.retention-cron:0 30 3 * * *}")
public void purgeExpired() {
LocalDateTime expireBefore = LocalDateTime.now().minusDays(properties.retentionDaysOrDefault());
while (true) {
Page<TaskEvidenceEntity> expiredPage = mapper.selectPage(
new Page<>(1, RETENTION_BATCH),
new LambdaQueryWrapper<TaskEvidenceEntity>()
.lt(TaskEvidenceEntity::getCreatedAt, expireBefore)
.orderByAsc(TaskEvidenceEntity::getId));
List<TaskEvidenceEntity> batch = expiredPage.getRecords();
DistributedJobLockService.LockHandle lockHandle =
distributedJobLockService.tryLock(RETENTION_LOCK, Duration.ofMinutes(15));
if (lockHandle == null) {
// 点位字面量必须写死(LockSkipSilentExitGuardrailTest 的判据),与 RETENTION_LOCK 同值。
LockSkipWitness.skipped("task-evidence:retention");
log.info("[task-evidence] 保留期清理跳过:另一实例持锁");
return;
}
try (lockHandle) {
purgeExpiredLocked(LocalDateTime.now().minusDays(properties.retentionDaysOrDefault()));
} catch (Exception ex) {
log.warn("[task-evidence] 保留期清理失败 reason={}", ex.getMessage(), ex);
}
}
private void purgeExpiredLocked(LocalDateTime expireBefore) {
long afterId = 0L;
int removed = 0;
int failed = 0;
int batches = 0;
while (batches < MAX_RETENTION_BATCHES) {
List<TaskEvidenceEntity> batch = mapper.selectList(new LambdaQueryWrapper<TaskEvidenceEntity>()
.gt(TaskEvidenceEntity::getId, afterId)
.lt(TaskEvidenceEntity::getCreatedAt, expireBefore)
.orderByAsc(TaskEvidenceEntity::getId)
.last("limit " + RETENTION_BATCH));
if (batch.isEmpty()) {
return;
break;
}
int removed = 0;
batches++;
for (TaskEvidenceEntity entity : batch) {
afterId = entity.getId();
try {
storage.delete(entity.getObjectKey());
removed += mapper.deleteById(entity.getId());
} catch (RuntimeException ex) {
// 只跳过这一行:登记行留着,下一轮按同一条键集再试;后面的行绝不陪葬。
failed++;
log.warn("[task-evidence] retention delete failed id={} reason={}",
entity.getId(), ex.getMessage());
}
}
if (removed == 0) {
return;
}
log.info("[task-evidence] retention purge removed={} before={}", removed, expireBefore);
}
if (failed > 0) {
log.warn("[task-evidence] retention purge finished with failures removed={} failed={} batches={}",
removed, failed, batches);
} else if (removed > 0 || batches > 0) {
log.info("[task-evidence] retention purge removed={} batches={}", removed, batches);
}
}
/**
* P1-2 的归属判定:task_id 对得上库里的任务行时 uid 必须就是行主;对不上行(客户端本地
* 任务号,如商品采集的自有作业 id)只要求带 uid 自证——否则任何一个持内部令牌的客户端
* 都能往别人的取证面里挂证据。
*/
private void assertTaskOwnedBy(String taskId, Long uid) {
Long ownerId = resolveTaskOwnerId(taskId);
if (uid == null) {
throw new BusinessException(403, "证据登记必须携带任务归属用户");
}
if (ownerId != null && !ownerId.equals(uid)) {
throw new BusinessException(403, "证据件只能登记到本人任务上");
}
}
/** 数字串按主键查(超长数字串会溢出,降级走 task_no),其余按 task_no 点查。 */
private Long resolveTaskOwnerId(String taskId) {
FileTaskEntity row = null;
if (taskId.chars().allMatch(Character::isDigit)) {
try {
row = fileTaskMapper.selectById(Long.parseLong(taskId));
} catch (NumberFormatException overflow) {
row = null;
}
}
if (row == null) {
row = fileTaskMapper.selectOne(new LambdaQueryWrapper<FileTaskEntity>()
.eq(FileTaskEntity::getTaskNo, taskId)
.last("limit 1"));
}
return row == null ? null : row.getUserId();
}
private long countByTask(String taskId) {
Long count = mapper.selectCount(new LambdaQueryWrapper<TaskEvidenceEntity>()
.eq(TaskEvidenceEntity::getTaskId, taskId));
return count == null ? 0L : count;
}
private String normalizeTaskId(String taskId) {
@@ -202,13 +306,15 @@ public class TaskEvidenceService {
return value;
}
private String normalizeLabel(String value, String fallback) {
String text = value == null ? "" : value.trim();
private String normalizeLabel(String value, String fallback, Set<String> allowed) {
String text = value == null ? "" : value.trim().toLowerCase(Locale.ROOT);
if (text.isEmpty()) {
return fallback;
}
// 列宽是 VARCHAR(32),超长标签在这里截断而不是让 insert 抛数据截断错。
return text.length() > 32 ? text.substring(0, 32) : text;
// 列宽是 VARCHAR(32),超长标签在这里截断而不是让 insert 抛数据截断错;
// 不在白名单内的标签一律落 other,不让客户端自造类型污染取证面。
String normalized = text.length() > 32 ? text.substring(0, 32) : text;
return allowed.contains(normalized) ? normalized : "other";
}
private void deleteTempQuietly(File file) {
@@ -51,6 +51,11 @@ public class TaskEvidenceStorageService {
properties.bucketOrDefault(), objectKey, maxBytes);
}
/** P3-1:下载走流式,不把整个证据件读进堆。 */
public void downloadTo(String objectKey, java.io.OutputStream out, long maxBytes) {
ossStorageService.readObjectToStream(properties.bucketOrDefault(), objectKey, out, maxBytes);
}
public void delete(String objectKey) {
if (objectKey == null || objectKey.isBlank()) {
return;
@@ -51,13 +51,13 @@ public class ChunkedUploadService {
private static final ObjectMapper JSON = new ObjectMapper();
private static final String META_NAME = "meta.json";
/** 单个分片上限:比前端 16MB 分片留足余量,防一次请求把临时盘打满。 */
/** 单个分片上限:两端与客户端实际片大小均为 8MB(各自独立构建根,无共享常量),这里留 8 倍余量。 */
private static final long MAX_PART_BYTES = 64L * 1024 * 1024;
private final StorageProperties storageProperties;
/** 同一会话的落片/拼装/放弃串行化;会话结束即回收,避免长期驻留。 */
private final ConcurrentHashMap<String, Object> sessionLocks = new ConcurrentHashMap<>();
/** 同一会话的落片/拼装/放弃串行化;static 让统一清理(LocalTempCleanupService)拿到同一把锁。 */
private static final ConcurrentHashMap<String, Object> SESSION_LOCKS = new ConcurrentHashMap<>();
@Value("${aiimage.file.chunk-upload.max-total-bytes:2147483648}")
private long maxTotalBytes = 2L * 1024 * 1024 * 1024;
@@ -194,7 +194,7 @@ public class ChunkedUploadService {
"拼装后大小与声明不一致:声明 " + meta.totalBytes + ",实际 " + total);
}
deleteQuietly(sessionDir(uploadId));
sessionLocks.remove(uploadId);
SESSION_LOCKS.remove(uploadId);
String sha256 = HexFormat.of().formatHex(digest.digest());
log.info("[chunk-upload] 拼装完成 uploadId={} fileName={} bytes={} sha256={}",
uploadId, meta.fileName, total, sha256);
@@ -208,7 +208,7 @@ public class ChunkedUploadService {
synchronized (lockOf(uploadId)) {
deleteQuietly(sessionDir(uploadId));
}
sessionLocks.remove(uploadId);
SESSION_LOCKS.remove(uploadId);
log.info("[chunk-upload] 已放弃分片会话 uploadId={} fileName={}", uploadId, meta.fileName);
}
@@ -273,7 +273,21 @@ public class ChunkedUploadService {
}
private Object lockOf(String uploadId) {
return sessionLocks.computeIfAbsent(uploadId, key -> new Object());
return sessionLockOf(uploadId);
}
/**
* P2-3:统一清理与 savePart/complete 必须是<b>同一把</b>锁,所以锁身份收在这里而不是
* 各自 new 一份 map(两份 map 等于没锁)。锁的是盘上目录名,进程内互斥即够——
* 每台实例只清自己的 local-temp 目录,跨实例不存在同一块盘。
*/
public static Object sessionLockOf(String uploadId) {
return SESSION_LOCKS.computeIfAbsent(uploadId, key -> new Object());
}
/** 会话目录名形状(32 位十六进制);统一清理按它决定要不要拿会话锁。 */
public static boolean isSessionDirectoryName(String name) {
return name != null && UPLOAD_ID_PATTERN.matcher(name).matches();
}
/**
@@ -88,7 +88,7 @@ public class LocalTempCleanupService {
continue;
}
if (child.isDirectory() && CHUNK_UPLOAD_DIR_NAME.equals(child.getName())) {
deletedSourceCount += deleteExpiredChildrenRecursively(child, sourceExpireBefore);
deletedSourceCount += deleteExpiredChunkUploads(child, sourceExpireBefore);
// 删片会刷新会话目录 mtime,按"过期"再判目录会留下空壳;这里按空目录自底向上收口。
deleteEmptyDirectoriesDeep(child, tempDir);
}
@@ -126,6 +126,33 @@ public class LocalTempCleanupService {
return 0;
}
/**
* P2-3:分片会话的过期件回收必须与 savePart/complete 共用同一把会话锁——原来裸删,
* 在飞上传的半截会话可能被整点回收腰斩,用户的大档从 0 重来(正是断点续传要消灭的事)。
* 非会话名的残留(历史上漏下的散件)保持原路径。
*/
private int deleteExpiredChunkUploads(File chunkRoot, Instant expireBefore) {
int deletedCount = 0;
File[] sessions = chunkRoot.listFiles();
if (sessions == null) {
return 0;
}
for (File session : sessions) {
if (!PathSafetyGuard.isInside(chunkRoot, session)) {
log.warn("local temp cleanup skipped unsafe path: {}", session.getAbsolutePath());
continue;
}
if (ChunkedUploadService.isSessionDirectoryName(session.getName())) {
synchronized (ChunkedUploadService.sessionLockOf(session.getName())) {
deletedCount += deleteExpiredChildrenRecursively(session, expireBefore);
}
} else {
deletedCount += deleteExpiredChildrenRecursively(session, expireBefore);
}
}
return deletedCount;
}
private void deleteEmptyDirectories(File directory, File stopAt) {
File current = directory;
while (current != null && !current.equals(stopAt) && isDirectoryEmpty(current)) {
@@ -21,6 +21,7 @@ import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.File;
import java.io.InputStream;
import java.io.OutputStream;
import java.net.URI;
import java.net.URLDecoder;
import java.nio.charset.StandardCharsets;
@@ -541,6 +542,35 @@ public class OssStorageService {
}
}
/**
* 流式读对象(P3-1):证据件下载不再把整对象读进堆;超上限时中止并抛 IllegalArgumentException。
*/
public void readObjectToStream(String bucket, String objectKey, OutputStream out, long maxBytes) {
String normalizedBucket = requireStorageName(bucket, "bucket");
String normalizedObjectKey = requireStorageName(objectKey, "objectKey");
try (var stream = buildClient().getObject(GetObjectArgs.builder()
.bucket(normalizedBucket)
.object(normalizedObjectKey)
.build())) {
byte[] buffer = new byte[64 * 1024];
long total = 0;
int read;
while ((read = stream.read(buffer)) != -1) {
total += read;
if (total > maxBytes) {
throw new IllegalArgumentException("对象过大,超过读取上限 " + maxBytes + " 字节: "
+ normalizedBucket + "/" + normalizedObjectKey);
}
out.write(buffer, 0, read);
}
out.flush();
} catch (IllegalArgumentException ex) {
throw ex;
} catch (Exception ex) {
throw storageFailure("read", normalizedBucket + "/" + normalizedObjectKey, ex);
}
}
/**
* Reads a managed result object from either its object key or stored public URL.
* This overload keeps callers from depending on the configured bucket name.
@@ -54,7 +54,8 @@ public class PermissionMenuService {
/** 角色内在闸门菜单:非超管看不见、谁也授不出去(设计文档 §8.1)。
* 与前端 src/config/company-scope.ts 的 SUPER_ADMIN_ONLY_MENU_KEYS 必须同值,由 Task 7 的跨端用例钉住。 */
private static final Set<String> SUPER_ADMIN_ONLY_KEYS = Set.of("admin_companies");
private static final Set<String> SUPER_ADMIN_ONLY_KEYS =
Set.of("admin_companies", "admin_task_evidence");
/**
* 管理/平台配置类菜单:出口仍是 {@code requireAdmin}(用户、菜单、分组、密钥、公司、版本、教程、
@@ -214,12 +214,6 @@ public class SimilarAsinTaskService implements SimilarAsinPipelineHost {
private final SimilarAsinImagePrefetchService imagePrefetchService;
private final UserApiSecretService userApiSecretService;
/**
* 采集路由开关:字段注入,避免给 {@code @RequiredArgsConstructor} 加参数而牵连一批
* 手工 new 本类的测试。null(单测未注入)按「本地」处理,保持 activate 的原有行为。
*/
@org.springframework.beans.factory.annotation.Autowired(required = false)
private com.nanri.aiimage.config.CollectRoutingProperties collectRoutingProperties;
/**
* Task 89:Excel 行解析器(表头/单元格读取、别名匹配、空行跳过、单字段截断)。
* 由 parseAndCreateTask 委托;解析语义与搬移前 parseWorkbook 完全一致。
@@ -1409,7 +1409,9 @@ public class SimilarAsinPipelineSupport {
}
LlmBatchContext context = SimilarAsinPayloadSupport.readLlmBatchContext(objectMapper, state);
if (context == null || context.resultId() == null) {
markLlmStateTerminal(state, LLM_STATUS_FAILED, "LLM 批次上下文缺失");
markLlmStateTerminal(state, LLM_STATUS_FAILED, "LLM 异步工作流丢失批次上下文");
// S-4:异步分支同样要给"欠费/坏密钥"留整单判死出口;非致命错误时该方法是 no-op。
host.terminalizeOnFatalLlmFailure(taskForPoll(taskId), state.getLlmError());
continue;
}
List<SimilarAsinResultRowDto> batchRows = readLlmBatchRows(state);
@@ -1423,8 +1425,10 @@ public class SimilarAsinPipelineSupport {
allRowsByBaseId);
}
markLlmStateTerminal(state, LLM_STATUS_FAILED, "LLM 异步工作流轮询超时");
// S-4:轮询超时本身不是欠费;但若 state 上已带 LLM 欠费/坏密钥归因,走同一判死出口。
host.terminalizeOnFatalLlmFailure(task, state.getLlmError());
maybeFinalizeLlmJob(taskId, context);
log.warn("[similar-asin] 文件任务超时兜底已将 LLM pending 批次置为失败 taskId={} stateId={} jobId={}",
log.warn("[similar-asin] 异步工作流轮询超时,本批按行失败收口 taskId={} stateId={} jobId={}",
taskId, state.getId(), context.jobId());
}
}
@@ -1,6 +1,7 @@
package com.nanri.aiimage.modules.softwareversion.service;
import cn.hutool.crypto.digest.DigestUtil;
import com.nanri.aiimage.common.security.CredentialPackageGuard;
import com.nanri.aiimage.common.text.UserMessages;
import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
import com.nanri.aiimage.common.exception.BusinessException;
@@ -281,6 +282,7 @@ public class SoftwareVersionService {
deleteTempQuietly(source);
return toItemMap(existing);
}
rejectCredentialFiles(source);
String objectKey = softwareVersionObjectKey(normalizedVersion);
String fileUrl = ossStorageService.uploadSoftwareVersionPackage(source, objectKey);
SoftwareVersionEntity entity = new SoftwareVersionEntity();
@@ -296,6 +298,28 @@ public class SoftwareVersionService {
return toItemMap(saved == null ? entity : saved);
}
/**
* P2-4:版本包会进公开 client 桶、发给所有客户端;压缩包里带 .env/*.pem 的一律拒绝。
* 只读 zip 条目名,不解压内容。
*/
private void rejectCredentialFiles(File source) {
String entry;
try {
entry = CredentialPackageGuard.firstForbiddenEntry(source);
} catch (IOException ex) {
log.warn("[software-version] 安装包无法读取,拒绝发布 name={} reason={}",
source.getName(), ex.getMessage());
deleteTempQuietly(source);
throw new BusinessException("安装包无法读取或不是有效 zip,拒绝发布");
}
if (entry != null) {
log.warn("[software-version] 安装包含凭据文件,拒绝发布 name={} entry={}",
source.getName(), entry);
deleteTempQuietly(source);
throw new BusinessException("安装包含凭据文件,拒绝发布");
}
}
private void deleteTempQuietly(File file) {
if (file != null && file.exists() && !file.delete()) {
log.warn("[software-version] 清理分片临时件失败 path={}", file.getAbsolutePath());
@@ -25,4 +25,35 @@ public interface FileTaskMapper extends BaseMapper<FileTaskEntity> {
List<Long> selectExpiredTerminalTaskIds(@Param("moduleType") String moduleType,
@Param("cutoff") LocalDateTime cutoff,
@Param("batchSize") int batchSize);
/**
* 客户端兜底拉取候选:本用户 + 指定状态 + 白名单模块 + 创建早于 cutoff,且<b>这一单走本机通道</b>。
*
* <p>通道判定必须在 SQL 里、在 {@code LIMIT} 之前。2026-10-31 的 P1-1 把它写在 Java 侧
* (先 {@code limit 5} 取页、再 {@code filter(isLocal())}),混通道时只要排队里较老的 5 条是
* ONLINE,本机通道的单本轮及以后都领不到——不是延后,是排到队尾之前永久饿死。
*
* <p>只投影认领与组装载荷要用的列(其余字段由调用方 {@code selectById} 再取),
* 与迁移前的 Wrapper {@code select(...)} 逐字同形。
*/
List<FileTaskEntity> selectLocalChannelPendingTasks(@Param("userId") Long userId,
@Param("status") String status,
@Param("moduleTypes") List<String> moduleTypes,
@Param("localDefaultModules") List<String> localDefaultModules,
@Param("cutoff") LocalDateTime cutoff,
@Param("limit") int limit);
/**
* 服务端内部认领候选(Linux 后台 worker):指定状态 + 白名单模块 + 创建早于 cutoff,
* 且「不是本机通道」或「本机通道已等满兜底窗口 {@code localCutoff}」。
*
* <p>同样是 SQL 侧谓词:原先先按 {@code cutoff}(2 分钟)取页、再在 Java 里按 30 分钟让位,
* 一页里全是让位中的本机单时本轮线上单零吞吐。
*/
List<FileTaskEntity> selectInternalClaimCandidates(@Param("status") String status,
@Param("moduleTypes") List<String> moduleTypes,
@Param("localDefaultModules") List<String> localDefaultModules,
@Param("cutoff") LocalDateTime cutoff,
@Param("localCutoff") LocalDateTime localCutoff,
@Param("limit") int limit);
}
@@ -2,6 +2,7 @@ package com.nanri.aiimage.modules.task.model;
import com.nanri.aiimage.config.CollectRoutingProperties;
import java.util.List;
import java.util.Locale;
/**
@@ -23,17 +24,28 @@ public enum CollectChannel {
/** 线上采集:Linux 后台 worker 或外机 Windows。 */
ONLINE;
private static final List<String> LOCAL_DEFAULT_MODULES = List.of(
CollectRoutingProperties.MODULE_APPEARANCE_PATENT,
CollectRoutingProperties.MODULE_SIMILAR_ASIN);
/**
* 未登记通道时读作 {@link #LOCAL} 的模块清单。
*
* <p>{@link #defaultFor(String)} 与 mapper 里那条通道谓词({@code FileTaskMapper.xml} 的
* {@code resolvedChannelCase})共用这一处真源:SQL 侧要的是同一份「哪些模块默认本机」,
* 两边各写一套就会出现「Java 认为该让位、SQL 却把它捞进候选」的分裂判据。
*/
public static List<String> localDefaultModuleTypes() {
return LOCAL_DEFAULT_MODULES;
}
/** 该模块在没有显式登记通道时的默认执行方。 */
public static CollectChannel defaultFor(String moduleType) {
if (moduleType == null || moduleType.isBlank()) {
return ONLINE;
}
String key = moduleType.trim().toUpperCase(Locale.ROOT);
if (CollectRoutingProperties.MODULE_APPEARANCE_PATENT.equals(key)
|| CollectRoutingProperties.MODULE_SIMILAR_ASIN.equals(key)) {
return LOCAL;
}
return ONLINE;
return LOCAL_DEFAULT_MODULES.contains(key) ? LOCAL : ONLINE;
}
/** 解析库里存的通道值;空或不可识别时回落到该模块的默认(不让一个脏值把单子永久饿死)。 */
@@ -1,6 +1,5 @@
package com.nanri.aiimage.modules.task.service;
import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.nanri.aiimage.config.CollectRoutingProperties;
import com.nanri.aiimage.config.InstanceMetadata;
@@ -139,22 +138,14 @@ public class TaskClientPullService {
log.warn("[client-task-pull] 模块白名单为空,本次不拉取 userId={}", userId);
return List.of();
}
// 通道由这一单自己说(V183 collect_channel):选了线上的单子客户端不能领,领了就是抢跑。
// 通道由这一单自己说(V183 collect_channel):选了线上的单子客户端不能领,领了就是抢跑,
// 所以判定必须发生在 SQL 里——先按 limit 取页、回到 Java 再 filter 的话,混通道时较老的
// N 条是 ONLINE 就把本机单永久挡在队尾之后(不是延后,是饿死)。
long startedAt = System.currentTimeMillis();
LocalDateTime cutoff = LocalDateTime.now(DB_ZONE).minusMinutes(Math.max(1L, minPendingMinutes));
int safeLimit = Math.max(1, Math.min(limit, MAX_LIMIT));
List<FileTaskEntity> candidates = fileTaskMapper.selectList(new LambdaQueryWrapper<FileTaskEntity>()
.select(FileTaskEntity::getId, FileTaskEntity::getModuleType,
FileTaskEntity::getCollectChannel, FileTaskEntity::getCreatedAt)
.eq(FileTaskEntity::getUserId, userId)
.eq(FileTaskEntity::getStatus, STATUS_PENDING)
.in(FileTaskEntity::getModuleType, types)
.lt(FileTaskEntity::getCreatedAt, cutoff)
.orderByAsc(FileTaskEntity::getCreatedAt)
.last("limit " + safeLimit))
.stream()
.filter(task -> CollectChannel.resolve(task.getCollectChannel(), task.getModuleType()).isLocal())
.toList();
List<FileTaskEntity> candidates = fileTaskMapper.selectLocalChannelPendingTasks(
userId, STATUS_PENDING, types, CollectChannel.localDefaultModuleTypes(), cutoff, safeLimit);
if (candidates.isEmpty() && !types.isEmpty()) {
log.info("[client-task-pull] 白名单内没有本机通道的排队任务,本次不拉取 userId={} moduleTypes={}",
userId, moduleTypes);
@@ -195,27 +186,10 @@ public class TaskClientPullService {
Math.max(1L, internalMinPendingMinutes));
LocalDateTime localCutoff = now.minusMinutes(localFallbackMinutes);
int safeLimit = Math.max(1, Math.min(limit, MAX_LIMIT));
List<FileTaskEntity> candidates = fileTaskMapper.selectList(new LambdaQueryWrapper<FileTaskEntity>()
.select(FileTaskEntity::getId, FileTaskEntity::getModuleType,
FileTaskEntity::getCollectChannel, FileTaskEntity::getCreatedAt)
.eq(FileTaskEntity::getStatus, STATUS_PENDING)
.in(FileTaskEntity::getModuleType, types)
.lt(FileTaskEntity::getCreatedAt, cutoff)
.orderByAsc(FileTaskEntity::getCreatedAt)
.last("limit " + safeLimit));
if (!candidates.isEmpty()) {
int before = candidates.size();
// 让位只看这一单的通道:本机通道的单子等满兜底窗口才接管,选了线上的立刻可领
candidates = candidates.stream()
.filter(task -> !CollectChannel.resolve(task.getCollectChannel(), task.getModuleType()).isLocal()
|| task.getCreatedAt() == null
|| !task.getCreatedAt().isAfter(localCutoff))
.toList();
if (candidates.size() < before) {
log.info("[client-task-pull] 内部认领让位本机:本轮跳过 {} 条 LOCAL 通道的排队任务(兜底窗口 {} 分钟)",
before - candidates.size(), localFallbackMinutes);
}
}
// 让位判定同样在 SQL 里(与客户端拉取同一个形状):原先按 cutoff 取一页再在 Java 里按
// localCutoff 让位,整页都是「本机通道且还没等满兜底窗口」时,本轮线上单一条也领不到。
List<FileTaskEntity> candidates = fileTaskMapper.selectInternalClaimCandidates(
STATUS_PENDING, types, CollectChannel.localDefaultModuleTypes(), cutoff, localCutoff, safeLimit);
if (candidates.isEmpty()) {
return List.of();
}
@@ -240,17 +240,20 @@ public class TaskFileJobService {
}
/**
* 终态父任务的 job 直接拉满重试:不执行、不再被下一轮扫到。
* 终态父任务的 job 也放行认领(P1-13 根治):这张表只有 ASSEMBLE_RESULT 一种作业,
* 它的唯一职责是把<b>已回传的数据</b>拼成表,不消耗任何外部资源——父任务已终态恰恰
* 意味着数据已定型,此时出表才是对的。原先在这里无条件判死,出表只剩「卡死扫描
* 捞 FAILED+retry 到顶」一条补偿路径:至少一个扫描周期的窗口里用户停在红字,
* 兜底再一失败那张本可以出的表就永远出不来(2026-10-31 现网任务 33062 取证)。
*
* <p>终态回调要留给卡死扫描的失败兜底({@code fallbackAssembleOnFailure}):外观专利等模块
* 靠它用已回传分片出部分结果文件,「起码失败了也可以下载」才不会被 B28 的终态守卫顺手掐掉。
* 这里若直接打 {@code terminal_callback_at},兜底扫描再也捞不到这行,用户只剩红字、连半张表都没有。
* <p>终态守卫的本意是「终态后别再继续检测、停心跳续期」,保留在
* {@code TaskResultFileJobWorker} 的心跳层(stage=TERMINAL_STOP_RENEW);
* 出表执行与任务结论脱钩(铁律 14):组装只写 biz_file_result,biz_file_task
* 的 status/error_message 一字不动。
*/
private void skipJobOfTerminalTask(TaskFileJobEntity job, String stage) {
String message = "父任务已终态,结果作业不再执行";
markFailedPermanentAwaitingFallback(job, message);
log.info("[task-file-job] skipped because parent task is terminal jobId={} taskId={} moduleType={} stage={}",
job.getId(), job.getTaskId(), job.getModuleType(), stage);
private void noteTerminalTaskStillDeliverable(TaskFileJobEntity job) {
log.info("[task-file-job] parent task terminal; ASSEMBLE_RESULT still delivers jobId={} taskId={} moduleType={} stage=TERMINAL_ASSEMBLE",
job.getId(), job.getTaskId(), job.getModuleType());
}
private List<TaskFileJobEntity> claimCandidates(List<TaskFileJobEntity> candidates) {
@@ -263,8 +266,7 @@ public class TaskFileJobService {
continue;
}
if (isOwningTaskTerminal(candidate.getTaskId())) {
skipJobOfTerminalTask(candidate, "CLAIM");
continue;
noteTerminalTaskStillDeliverable(candidate);
}
if (markRunning(candidate.getId())) {
claimedIds.add(candidate.getId());
@@ -629,8 +631,8 @@ public class TaskFileJobService {
}
TaskFileJobEntity existing = taskFileJobMapper.selectById(jobId);
if (existing != null && isOwningTaskTerminal(existing.getTaskId())) {
skipJobOfTerminalTask(existing, "CLAIM_RUNNING");
return null;
// P1-13:终态父任务也放行认领(见 noteTerminalTaskStillDeliverable),出表不因终态停摆。
noteTerminalTaskStillDeliverable(existing);
}
if (!markRunning(jobId)) {
return null;
@@ -260,12 +260,9 @@ public class TaskResultFileJobWorker {
}
private void processClaimedWithHeartbeat(TaskFileJobEntity job, TaskFileJobEntity claim) {
if (taskFileJobService.isOwningTaskTerminal(job.getTaskId())) {
taskFileJobService.markFailedPermanentAwaitingFallback(job, "父任务已终态,结果作业不再执行");
log.info("[task-file-job] skipped because parent task is terminal jobId={} taskId={} moduleType={} stage=TERMINAL_SKIP",
job.getId(), job.getTaskId(), job.getModuleType());
return;
}
// P1-13:不再在执行入口按父任务终态判死——这张表只有纯出表作业,终态恰恰说明数据
// 已定型,此时出表才是对的(2026-10-31 现网任务 33062:守卫判死 + 兜底 50 秒后才有表,
// 窗口里用户只剩红字)。终态的「别再检测」语义保留在心跳层(stage=TERMINAL_STOP_RENEW)。
if (!taskFileJobService.activateRunningClaim(claim)) {
log.info("[task-file-job] queued claim expired before execution jobId={} taskId={} moduleType={} claimUpdatedAt={}",
job.getId(), job.getTaskId(), job.getModuleType(), claim.getUpdatedAt());
@@ -295,7 +292,9 @@ public class TaskResultFileJobWorker {
job.getId(), job.getTaskId(), job.getModuleType());
return;
}
taskFileJobService.markFailedPermanentAwaitingFallback(job, "父任务已终态,结果作业不再执行");
// P1-13:这条只该在「执行真的没产出结果」时走到——出表已放行,没出成表才是遗骸,
// 留给兜底扫描出部分结果(幂等重入)。
taskFileJobService.markFailedPermanentAwaitingFallback(job, "父任务已终态且本次执行未产出结果文件");
log.info("[task-file-job] parent task terminal after execution; failed job for fallback jobId={} taskId={} moduleType={} stage=TERMINAL_AFTER_EXECUTION",
job.getId(), job.getTaskId(), job.getModuleType());
}
@@ -328,14 +327,6 @@ public class TaskResultFileJobWorker {
job.getId(), job.getTaskId(), job.getModuleType(), job.getResultId());
return;
}
if (taskFileJobService.isOwningTaskTerminal(job.getTaskId())) {
try (lockHandle) {
taskFileJobService.markFailedPermanentAwaitingFallback(job, "父任务已终态,结果作业不再执行");
}
log.info("[task-file-job] skipped after acquiring lock because parent task is terminal jobId={} taskId={} moduleType={} stage=TERMINAL_AFTER_LOCK",
job.getId(), job.getTaskId(), job.getModuleType());
return;
}
activeLockHandles.put(job.getId(), lockHandle);
try (lockHandle) {
latest = taskFileJobService.findById(job.getId());
@@ -2,6 +2,7 @@ package com.nanri.aiimage.modules.tutorial.service;
import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
import com.nanri.aiimage.common.exception.BusinessException;
import com.nanri.aiimage.common.security.CredentialPackageGuard;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
import com.nanri.aiimage.modules.file.service.oss.OssStorageService;
import com.nanri.aiimage.modules.tutorial.mapper.TutorialPackageMapper;
@@ -11,6 +12,7 @@ import lombok.extern.slf4j.Slf4j;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.stereotype.Service;
import java.io.IOException;
import java.time.LocalDateTime;
import java.time.format.DateTimeFormatter;
import java.util.ArrayList;
@@ -160,6 +162,7 @@ public class TutorialPackageService {
}
return toItemMap(existing);
}
rejectCredentialFiles(source);
String generatedKey = OBJECT_KEY_PREFIX + OBJECT_NAME_FORMATTER.format(LocalDateTime.now())
+ "-" + safeFileName(normalizedName);
ossStorageService.uploadTutorialPackage(source, generatedKey);
@@ -262,6 +265,29 @@ public class TutorialPackageService {
}
}
/** P2-4:教程包同样进公开 client 桶,压缩包里带 .env/*.pem 的一律拒绝。 */
private void rejectCredentialFiles(java.io.File source) {
String entry;
try {
entry = CredentialPackageGuard.firstForbiddenEntry(source);
} catch (IOException ex) {
log.warn("[tutorial] 教程包无法读取,拒绝发布 name={} reason={}",
source.getName(), ex.getMessage());
if (source.exists() && !source.delete()) {
log.warn("[tutorial] 清理拒绝的临时件失败 path={}", source.getAbsolutePath());
}
throw new BusinessException("教程包无法读取或不是有效 zip,拒绝发布");
}
if (entry != null) {
log.warn("[tutorial] 教程包含凭据文件,拒绝发布 name={} entry={}",
source.getName(), entry);
if (source.exists() && !source.delete()) {
log.warn("[tutorial] 清理拒绝的临时件失败 path={}", source.getAbsolutePath());
}
throw new BusinessException("教程包含凭据文件,拒绝发布");
}
}
private String requireZipFileName(String fileName) {
String normalized = fileName == null ? "" : fileName.trim();
if (normalized.isEmpty()) {
@@ -37,6 +37,8 @@ AIIMAGE_OSS_ACCESS_KEY_SECRET=change-me
# 任务异常现场证据件(v1.2 Phase 3):独立桶与保留期必须显式配置;桶不得复用 nanri-ai-images。
AIIMAGE_EVIDENCE_OSS_BUCKET=task-evidence
AIIMAGE_EVIDENCE_RETENTION_DAYS=7
AIIMAGE_EVIDENCE_RETENTION_CRON=0 30 3 * * *
AIIMAGE_EVIDENCE_MAX_FILE_BYTES=67108864
AIIMAGE_EVIDENCE_KEY_PREFIX=task-evidence/
# 工作台客户端包直传预签名有效期(秒);90MB 级包在慢公网入口需要更长窗口。
@@ -0,0 +1,50 @@
-- V184: 兜底拉取与证据保留期两条取页路径的缺失索引(CR 第二轮 P1-1 同批 / P2-1)
--
-- 1) biz_file_task(status, module_type, created_at)
-- TaskClientPullService 的通道谓词从 Java 侧下推进 SQL 之后,内部认领(Linux worker)这一支
-- 的真实谓词就是 status='PENDING' AND module_type IN (...) AND created_at < ? ORDER BY created_at。
-- 现有索引里没有一条以 status 打头再跟 module_type(V120 是 (status, updated_at)、
-- V34 是 (module_type, user_id, created_at)、V92 是 (owner_instance_id, status, updated_at)),
-- 于是每轮按 PENDING 全集排序。
-- collect_channel 有意不进索引:通道判定写成 CASE(NULL 值在 SQL 三值逻辑下必须兜成模块默认,
-- 直接比值会让「未登记通道的非本机模块」被 NOT 判成 UNKNOWN 而永久领不到),表达式谓词不吃索引;
-- PENDING 本身是被本索引筛小的那一层,量级 = 在排队的单子。
-- 2) biz_task_evidence(created_at)
-- V182 只建了 (task_id, created_at):保留期回收的谓词是单列 created_at < ?,前导列不符 ⇒
-- purgeExpired 每一批全表扫,而这张表按「异常即产件」只增不减。
--
-- 风险:ALTER TABLE ADD INDEX 需申请元数据锁(MySQL 8 在线 DDL,INPLACE),建议低峰执行;
-- 本迁移按 information_schema 判存,重跑不会因索引已存在而失败。
-- 回滚(手工):
-- ALTER TABLE biz_file_task DROP INDEX idx_biz_file_task_status_module_created;
-- ALTER TABLE biz_task_evidence DROP INDEX idx_task_evidence_created_at;
SET @db_name = DATABASE();
-- 1) biz_file_task(status, module_type, created_at)
SET @idx_exists := (
SELECT COUNT(*) FROM information_schema.STATISTICS
WHERE TABLE_SCHEMA = @db_name AND TABLE_NAME = 'biz_file_task'
AND INDEX_NAME = 'idx_biz_file_task_status_module_created'
);
SET @sql := IF(@idx_exists = 0,
'ALTER TABLE biz_file_task ADD INDEX idx_biz_file_task_status_module_created (status, module_type, created_at)',
'SELECT 1'
);
PREPARE stmt FROM @sql;
EXECUTE stmt;
DEALLOCATE PREPARE stmt;
-- 2) biz_task_evidence(created_at)
SET @idx_exists := (
SELECT COUNT(*) FROM information_schema.STATISTICS
WHERE TABLE_SCHEMA = @db_name AND TABLE_NAME = 'biz_task_evidence'
AND INDEX_NAME = 'idx_task_evidence_created_at'
);
SET @sql := IF(@idx_exists = 0,
'ALTER TABLE biz_task_evidence ADD INDEX idx_task_evidence_created_at (created_at)',
'SELECT 1'
);
PREPARE stmt FROM @sql;
EXECUTE stmt;
DEALLOCATE PREPARE stmt;
@@ -0,0 +1,15 @@
-- 迁移 V185:后台菜单「任务证据」(v1.2 P1-9 / R3 后管可视)
--
-- 证据件读取面 /api/admin/task-evidence/** 已按 requireSuperAdmin 收口,但后台没有入口,
-- 等于「能上传、能登记、没人看得见」。这里只新增一条菜单行,挂在「记录与版本」分组下,
-- 幂等插入:column_key 已存在时不再写第二行。
--
-- 回滚:DELETE FROM columns WHERE column_key='admin_task_evidence';
INSERT INTO columns (name, column_key, menu_type, route_path, sort_order, parent_id)
SELECT '任务证据', 'admin_task_evidence', 'admin', 'records/task-evidence', 84, parent.id
FROM columns parent
WHERE parent.column_key = 'admin_group_record'
AND NOT EXISTS (
SELECT 1 FROM columns WHERE column_key = 'admin_task_evidence'
);
@@ -11,4 +11,48 @@
ORDER BY id ASC
LIMIT #{batchSize}
</select>
<!-- 这一单的执行通道,SQL 侧的 CollectChannel.resolve:登记过就按登记的算,
没登记(NULL)或存了不可识别值的按模块默认兜(localDefaultModules 由
CollectChannel.localDefaultModuleTypes() 传进来,两边共用一处真源)。
写成 CASE 而不是 collect_channel = 'LOCAL' OR …:后者在列为 NULL 时整个谓词吃三值逻辑
变成 UNKNOWN,NOT 之后仍不成立,会被默认兜成 ONLINE 的模块(COLLECT_DATA 等)就此领不到。 -->
<sql id="resolvedChannelCase">
CASE
WHEN collect_channel IN ('LOCAL', 'ONLINE') THEN collect_channel
WHEN module_type IN
<foreach collection="localDefaultModules" item="localDefaultModule" open="(" separator="," close=")">#{localDefaultModule}</foreach>
THEN 'LOCAL'
ELSE 'ONLINE'
END
</sql>
<!-- 客户端兜底拉取候选(P1-1):通道谓词与 status/模块/时间窗口同在 WHERE 里,LIMIT 之后不再有 Java 过滤。 -->
<select id="selectLocalChannelPendingTasks"
resultType="com.nanri.aiimage.modules.task.model.entity.FileTaskEntity">
SELECT id, module_type, collect_channel, created_at
FROM biz_file_task
WHERE user_id = #{userId}
AND status = #{status}
AND module_type IN
<foreach collection="moduleTypes" item="module" open="(" separator="," close=")">#{module}</foreach>
AND created_at &lt; #{cutoff}
AND <include refid="resolvedChannelCase"/> = 'LOCAL'
ORDER BY created_at ASC
LIMIT #{limit}
</select>
<!-- 服务端内部认领候选(P1-1 同形状):本机通道的单要让位给本地客户端,等满兜底窗口才接管。 -->
<select id="selectInternalClaimCandidates"
resultType="com.nanri.aiimage.modules.task.model.entity.FileTaskEntity">
SELECT id, module_type, collect_channel, created_at
FROM biz_file_task
WHERE status = #{status}
AND module_type IN
<foreach collection="moduleTypes" item="module" open="(" separator="," close=")">#{module}</foreach>
AND created_at &lt; #{cutoff}
AND (NOT (<include refid="resolvedChannelCase"/> = 'LOCAL') OR created_at &lt;= #{localCutoff})
ORDER BY created_at ASC
LIMIT #{limit}
</select>
</mapper>
@@ -74,7 +74,9 @@ class MigratedSqlEquivalenceTest {
// 99 → 100:Phase 5(AC-19)为任务中心深翻页新增窄锚点语句 {@code TaskCenterQueryMapper#selectTaskPageAnchor}
// (两列投影算页首游标),同轮点名进 NOT_MIGRATED;同轮的 selectTaskPage 改写让它从账本搬进
// NOT_MIGRATED(select 总数不变,只是记账位置换了)。
private static final int XML_SELECT_TOTAL = 100;
// 100 → 102:P1-1 把两条认领查询的通道谓词下推进 SQL,FileTaskMapper 新写 2 条 select,
// 逐条点名进 NOT_MIGRATED(SqlLayeringConventionTest.XML_STATEMENT_TOTAL 同轮 144 → 146)。
private static final int XML_SELECT_TOTAL = 102;
/**
* 不是「从注解搬过来」的语句:注解时代没有原文可对撞,等价性这支只能跳过它们。
@@ -135,7 +137,13 @@ class MigratedSqlEquivalenceTest {
// 所以整行从 migrated-statements.tsv 摘掉、点名到这里(LEDGER_ROWS 85→84)。
// selectTaskPageAnchor —— 本轮新写的两列窄投影,从来没有注解原文可对撞。
"TaskCenterQueryMapper#selectTaskPage",
"TaskCenterQueryMapper#selectTaskPageAnchor");
"TaskCenterQueryMapper#selectTaskPageAnchor",
// P1-1(2026-10-31):客户端兜底拉取与服务端内部认领的通道谓词从 Java 层下推进 SQL
// ——先 limit 取页再 filter(isLocal()) 会让本机单被前面的线上单永久挡死。两条都是
// 本轮新写的 select,从来没有注解原文可对撞(XML_SELECT_TOTAL 同轮 100 → 102,
// SqlLayeringConventionTest.XML_STATEMENT_TOTAL 同轮 144 → 146)。
"FileTaskMapper#selectLocalChannelPendingTasks",
"FileTaskMapper#selectInternalClaimCandidates");
/** 账本 5 列:接口 / 方法 / @Select 种类 / 是否含常量插值 / 注解原文(Java 反转义后那条串)。 */
private static final int LEDGER_COLUMNS = 5;
@@ -91,7 +91,11 @@ class SqlLayeringConventionTest {
// 143 → 144:Phase 5(AC-19/P1-B1)为任务中心深翻页新增窄锚点语句
// {@code TaskCenterQueryMapper#selectTaskPageAnchor}(两列投影算页首游标),
// 同轮登记进 MigratedSqlEquivalenceTest.NOT_MIGRATED(该文件的 XML_SELECT_TOTAL 同轮 99 → 100)。
private static final int XML_STATEMENT_TOTAL = 144;
// 144 → 146:P1-1(2026-10-31)把「这一单交本机还是交线上」的判定从 Java 层下推进 SQL——
// 原形状是先 limit 取页再 filter(isLocal()),混通道时较老的 N 条是 ONLINE 就把本机单永久饿死。
// FileTaskMapper 新写 2 条 select(客户端兜底拉取 / 服务端内部认领),同轮登记进
// MigratedSqlEquivalenceTest.NOT_MIGRATED(该文件 XML_SELECT_TOTAL 同轮 100 → 102)。
private static final int XML_STATEMENT_TOTAL = 146;
/**
* wrapper 里夹带函数级 SQL 的现状("这不是简单单表"的证据),文件名 → 条数:
@@ -0,0 +1,68 @@
package com.nanri.aiimage.common.security;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import java.io.File;
import java.io.IOException;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.zip.ZipEntry;
import java.util.zip.ZipOutputStream;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertNull;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
/** P2-4:版本包/教程包发布前的压缩包凭据件扫描。 */
class CredentialPackageGuardTest {
@TempDir
Path tempDir;
private File zipWith(String... entries) throws IOException {
File zip = tempDir.resolve("pkg-" + System.nanoTime() + ".zip").toFile();
try (ZipOutputStream out = new ZipOutputStream(Files.newOutputStream(zip.toPath()))) {
for (String entry : entries) {
out.putNextEntry(new ZipEntry(entry));
out.write("x".getBytes(StandardCharsets.UTF_8));
out.closeEntry();
}
}
return zip;
}
@Test
void nestedDotEnvIsRejected() throws IOException {
File pkg = zipWith("app/readme.txt", "app/conf/.env");
assertEquals("app/conf/.env", CredentialPackageGuard.firstForbiddenEntry(pkg));
}
@Test
void envVariantAndKeyMaterialAreRejected() throws IOException {
assertEquals("docs/.env.production",
CredentialPackageGuard.firstForbiddenEntry(zipWith("docs/.env.production")));
assertEquals("certs/server.PEM",
CredentialPackageGuard.firstForbiddenEntry(zipWith("certs/server.PEM")));
assertEquals("tools/kw_share_cfg.json",
CredentialPackageGuard.firstForbiddenEntry(zipWith("tools/kw_share_cfg.json")));
}
@Test
void ordinaryPackagePasses() throws IOException {
assertNull(CredentialPackageGuard.firstForbiddenEntry(
zipWith("app/readme.txt", "app/config/app.yml", "app/update/update.exe")));
}
@Test
void missingOrNonZipFileFailsClosed() throws IOException {
File notZip = tempDir.resolve("not-a-zip.zip").toFile();
Files.writeString(notZip.toPath(), "plain text");
IOException missing = assertThrows(IOException.class,
() -> CredentialPackageGuard.firstForbiddenEntry(tempDir.resolve("absent.zip").toFile()));
assertTrue(missing.getMessage().contains("missing"));
assertThrows(IOException.class, () -> CredentialPackageGuard.firstForbiddenEntry(notZip));
}
}
@@ -62,7 +62,8 @@ class CatchSilenceRatchetTest {
// 10-30 用户端代理白名单(极客/青果)抬 591 -> 598(7 处逐一记名见 STRICT_TOTAL 上方注释;
// 实测量到 599,多出的 1 格来自已提交的 79b04afd(AdminApiGuardFilter 编码绕过 catch),见同处说明)。
// 10-31 Phase 5.1:证据登记失败回收对象一枚(内层回收失败只 WARN),599 -> 600。
static final int GIVEN_NOT_RETHROW = 600;
// 10-31 CR 消项:证据保留期清理失败 WARN 兜底 + 任务号溢出回落 task_no,新增 2 处非抛分支,600 -> 602。
static final int GIVEN_NOT_RETHROW = 602;
static final int GIVEN_LOG_ONLY = 249;
static final int GIVEN_EMPTY = 56;
@@ -280,7 +281,8 @@ class CatchSilenceRatchetTest {
* 后转业务错误(RETHROW+rethrow 401);内层 catch 是对象回收本身失败只 WARN
* (LOG_ONLY+logOnly 190)。两处误差方向都是「不让孤儿对象永久占桶」。
*/
static final int STRICT_TOTAL = 1057;
// 10-31 CR 消项:版本包/教程包凭据扫描新增 6 处 catch(均转业务异常或带 WARN),1057 -> 1063。
static final int STRICT_TOTAL = 1063;
/**
* 重抛型 catch 的**下限**(不是上限):本轮把两处吞异常改成重抛之后实测 388。
* 这一格只能涨不能跌——它是"异常会往外走"的总量,把它钉成"只许减"就等于拦着别人修吞异常。
@@ -404,7 +406,7 @@ class CatchSilenceRatchetTest {
* (畸形编码保持原样交给容器 400),非本批新增,说明见 STRICT_TOTAL 上方。
*/
// 10-31 Phase 5.1:同 GIVEN_NOT_RETHROW,证据登记补偿回收失败只 WARN,599 -> 600。
static final int STRICT_NOT_RETHROW = 600;
static final int STRICT_NOT_RETHROW = 602;
/**
* 10-17 抬 172 -> 173,出处就是上面点名过的 {@code TaskHeartbeatLivenessService.stamp()}:
* 块体只有「取到 key 名 + log.warn」一句,写心跳证据失败时既不重试也不抛出——
@@ -9,7 +9,6 @@ import com.nanri.aiimage.common.exception.BusinessException;
import com.nanri.aiimage.common.service.DistributedJobLockService;
import com.nanri.aiimage.common.service.ResultDownloadResolver;
import com.nanri.aiimage.config.AppearancePatentProperties;
import com.nanri.aiimage.config.CollectRoutingProperties;
import com.nanri.aiimage.config.InstanceMetadata;
import com.nanri.aiimage.config.StorageProperties;
import com.nanri.aiimage.modules.appearancepatent.client.AppearancePatentLlmClient;
@@ -68,7 +67,7 @@ class AppearancePatentActivateRoutingTest {
FileTaskEntity task = pendingTask(com.nanri.aiimage.modules.task.model.CollectChannel.LOCAL);
when(fileTaskMapper.selectById(TASK_ID)).thenReturn(task);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
AppearancePatentTaskService service = service(new CollectRoutingProperties());
AppearancePatentTaskService service = service();
service.activateTask(TASK_ID, USER_ID);
@@ -85,12 +84,12 @@ class AppearancePatentActivateRoutingTest {
void 线上通道的单_activate_保持后台语义并明确拒绝() {
when(fileTaskMapper.selectById(TASK_ID))
.thenReturn(pendingTask(com.nanri.aiimage.modules.task.model.CollectChannel.ONLINE));
AppearancePatentTaskService service = service(new CollectRoutingProperties());
AppearancePatentTaskService service = service();
BusinessException error = assertThrows(BusinessException.class,
() -> service.activateTask(TASK_ID, USER_ID));
assertTrue(String.valueOf(error.getMessage()).contains("后台自动执行"), error.getMessage());
assertTrue(String.valueOf(error.getMessage()).contains("线上采集"), error.getMessage());
verify(fileTaskMapper, never()).update(isNull(), any());
}
@@ -100,7 +99,7 @@ class AppearancePatentActivateRoutingTest {
when(fileTaskMapper.selectById(TASK_ID)).thenReturn(task);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
service(new CollectRoutingProperties()).activateTask(TASK_ID, USER_ID);
service().activateTask(TASK_ID, USER_ID);
verify(taskCacheService).touchTaskHeartbeat(TASK_ID);
}
@@ -115,7 +114,7 @@ class AppearancePatentActivateRoutingTest {
return task;
}
private AppearancePatentTaskService service(CollectRoutingProperties routing) {
private AppearancePatentTaskService service() {
AppearancePatentTaskService service = new AppearancePatentTaskService(
mock(com.nanri.aiimage.modules.file.service.LocalFileStorageService.class),
null,
@@ -139,7 +138,6 @@ class AppearancePatentActivateRoutingTest {
mock(InstanceMetadata.class),
mock(TaskProgressLightAssembler.class),
mock(UserApiSecretService.class));
ReflectionTestUtils.setField(service, "collectRoutingProperties", routing);
return service;
}
@@ -0,0 +1,189 @@
package com.nanri.aiimage.modules.evidence.service;
import com.nanri.aiimage.common.exception.BusinessException;
import com.nanri.aiimage.common.service.DistributedJobLockService;
import com.nanri.aiimage.common.task.LockSkipWitness;
import com.nanri.aiimage.config.EvidenceProperties;
import com.nanri.aiimage.modules.evidence.mapper.TaskEvidenceMapper;
import com.nanri.aiimage.modules.evidence.model.entity.TaskEvidenceEntity;
import com.nanri.aiimage.modules.evidence.storage.TaskEvidenceStorageService;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
import com.nanri.aiimage.modules.task.mapper.FileTaskMapper;
import com.nanri.aiimage.modules.task.model.entity.FileTaskEntity;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.List;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.ArgumentMatchers.anyLong;
import static org.mockito.ArgumentMatchers.eq;
import static org.mockito.Mockito.doThrow;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.never;
import static org.mockito.Mockito.verify;
import static org.mockito.Mockito.verifyNoInteractions;
import static org.mockito.Mockito.when;
/**
* 证据登记的归属与保留期的执行保证(CR 第二轮 P1-2/P1-7/P2-2)。
*
* <p>P1-2 的判据是「归属先于落桶」:task_id 对得上任务行而 uid 不是行主 ⇒ 403 且
* {@code storage.store} 一次都不许被调——校验晚于落桶的话,每一次拒绝都在证据桶里
* 留一个收不回的孤儿对象。P1-7 的判据是「保留期必须有执行者且不堵车」:拿不到分布式锁
* 要走可聚合出口;单个对象删除失败只牺牲那一行,键集继续往前走。
*/
class TaskEvidenceOwnershipAndRetentionTest {
private static final String TASK_ID = "33062";
private final TaskEvidenceMapper mapper = mock(TaskEvidenceMapper.class);
private final TaskEvidenceStorageService storage = mock(TaskEvidenceStorageService.class);
private final LocalFileStorageService localFiles = mock(LocalFileStorageService.class);
private final EvidenceProperties properties = new EvidenceProperties();
private final FileTaskMapper fileTaskMapper = mock(FileTaskMapper.class);
private final DistributedJobLockService locks = mock(DistributedJobLockService.class);
@TempDir
Path tempDir;
@Test
void 他人任务的证据必须403且一次都不落桶() throws Exception {
stubTaskOwner(7L);
stubUploadableFile();
BusinessException ex = assertThrows(BusinessException.class,
() -> service().register(TASK_ID, 8L, "client", "goods-spider", "evidence.json"));
assertEquals(403, ex.getCode().intValue(), "冒名登记必须 403,不是 500 也不是静默成功");
verify(storage, never()).store(any(), any(), any());
verify(mapper, never()).insert(any(TaskEvidenceEntity.class));
}
@Test
void 无归属uid的登记必须拒绝() throws Exception {
stubTaskOwner(7L);
stubUploadableFile();
assertThrows(BusinessException.class,
() -> service().register(TASK_ID, null, "client", "goods-spider", "evidence.json"));
verify(storage, never()).store(any(), any(), any());
}
@Test
void 本人任务照常登记且登记行带上uid() throws Exception {
stubTaskOwner(7L);
Path file = stubUploadableFile();
when(storage.store(any(), eq(TASK_ID), eq("evidence.json"))).thenReturn(
new TaskEvidenceStorageService.StoredEvidence("task-evidence/k", 3L, "sha", "application/json"));
when(mapper.insert(any(TaskEvidenceEntity.class))).thenReturn(1);
service().register(TASK_ID, 7L, "client", "goods-spider", "evidence.json");
org.mockito.ArgumentCaptor<TaskEvidenceEntity> captor =
org.mockito.ArgumentCaptor.forClass(TaskEvidenceEntity.class);
verify(mapper).insert(captor.capture());
assertEquals(7L, captor.getValue().getUid(), "登记行的 uid 必须是行主本人");
assertEquals(TASK_ID, captor.getValue().getTaskId());
assertTrue(Files.notExists(file) || Files.exists(file), "临时件清理失败只留 warn,不影响登记");
}
@Test
void 客户端本地任务号带uid自证即可登记() throws Exception {
when(fileTaskMapper.selectById(anyLong())).thenReturn(null);
when(fileTaskMapper.selectOne(any())).thenReturn(null);
stubUploadableFile();
when(storage.store(any(), any(), any())).thenReturn(
new TaskEvidenceStorageService.StoredEvidence("task-evidence/k", 3L, "sha", "application/json"));
when(mapper.insert(any(TaskEvidenceEntity.class))).thenReturn(1);
service().register("local-job-abc", 5L, "client", "goods-spider", "evidence.json");
verify(storage).store(any(), eq("local-job-abc"), eq("evidence.json"));
}
@Test
void 零字节证据件拒绝登记() throws Exception {
when(fileTaskMapper.selectById(anyLong())).thenReturn(null);
when(fileTaskMapper.selectOne(any())).thenReturn(null);
Path empty = Files.createFile(tempDir.resolve("empty.png"));
when(localFiles.findLocalSourceFile("empty.png")).thenReturn(empty.toFile());
assertThrows(BusinessException.class,
() -> service().register(TASK_ID, 7L, "client", "goods-spider", "empty.png"));
verify(storage, never()).store(any(), any(), any());
}
@Test
void 同任务证据件数到上限必须拒绝() throws Exception {
when(fileTaskMapper.selectById(anyLong())).thenReturn(null);
when(fileTaskMapper.selectOne(any())).thenReturn(null);
when(mapper.selectCount(any())).thenReturn(50L);
stubUploadableFile();
assertThrows(BusinessException.class,
() -> service().register(TASK_ID, 7L, "client", "goods-spider", "evidence.json"));
verify(storage, never()).store(any(), any(), any());
}
@Test
void 清理拿不到锁必须走可聚合出口且不碰库() {
long before = LockSkipWitness.countOf(TaskEvidenceService.RETENTION_LOCK);
when(locks.tryLock(eq(TaskEvidenceService.RETENTION_LOCK), any())).thenReturn(null);
service().purgeExpired();
assertEquals(before + 1, LockSkipWitness.countOf(TaskEvidenceService.RETENTION_LOCK),
"锁跳过必须有 [lock-skip] 可聚合出口,静默 return 就是 P1-7 的另一半");
verifyNoInteractions(mapper, storage);
}
@Test
void 单个对象删除失败不堵后面的行() {
when(locks.tryLock(eq(TaskEvidenceService.RETENTION_LOCK), any()))
.thenReturn(mock(DistributedJobLockService.LockHandle.class));
TaskEvidenceEntity first = evidenceRow(1L, "k1");
TaskEvidenceEntity stuck = evidenceRow(2L, "k2");
TaskEvidenceEntity third = evidenceRow(3L, "k3");
when(mapper.selectList(any())).thenReturn(List.of(first, stuck, third), List.of());
doThrow(new RuntimeException("oss unavailable")).when(storage).delete("k2");
when(mapper.deleteById(anyLong())).thenReturn(1);
service().purgeExpired();
verify(mapper).deleteById(1L);
verify(mapper, never()).deleteById(2L);
verify(mapper).deleteById(3L);
}
private TaskEvidenceService service() {
return new TaskEvidenceService(mapper, storage, localFiles, properties, fileTaskMapper, locks);
}
private void stubTaskOwner(long uid) {
FileTaskEntity row = new FileTaskEntity();
row.setId(Long.parseLong(TASK_ID));
row.setUserId(uid);
when(fileTaskMapper.selectById(Long.parseLong(TASK_ID))).thenReturn(row);
}
private Path stubUploadableFile() throws Exception {
Path file = tempDir.resolve("evidence.json");
Files.writeString(file, "{\"traceback\": \"boom\"}");
when(localFiles.findLocalSourceFile("evidence.json")).thenReturn(file.toFile());
return file;
}
private static TaskEvidenceEntity evidenceRow(long id, String objectKey) {
TaskEvidenceEntity row = new TaskEvidenceEntity();
row.setId(id);
row.setTaskId(TASK_ID);
row.setObjectKey(objectKey);
return row;
}
}
@@ -0,0 +1,145 @@
package com.nanri.aiimage.modules.softwareversion.service;
import com.baomidou.mybatisplus.core.MybatisConfiguration;
import com.baomidou.mybatisplus.core.metadata.TableInfoHelper;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
import com.nanri.aiimage.modules.file.service.oss.OssStorageService;
import com.nanri.aiimage.modules.softwareversion.mapper.SoftwareVersionMapper;
import com.nanri.aiimage.modules.softwareversion.model.entity.SoftwareVersionEntity;
import org.apache.ibatis.builder.MapperBuilderAssistant;
import org.junit.jupiter.api.BeforeAll;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import org.mockito.ArgumentCaptor;
import org.springframework.test.util.ReflectionTestUtils;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.zip.ZipEntry;
import java.util.zip.ZipOutputStream;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.junit.jupiter.api.Assertions.assertNotEquals;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.never;
import static org.mockito.Mockito.verify;
import static org.mockito.Mockito.when;
/**
* 版本包「分片拼装件 → confirm 搬桶落库」这条通道的契约(CR 第二轮 P1-8)。
*
* <p>新重载 {@code confirmSoftwareVersion(version, fileKey)} 写好之后长期没有任何执行证据
* (后端整量绿也覆盖不到没被调用的分支)。幂等命中、临时件清理、缺件拒绝这三条是发版
* 判据:任何一条破掉,要么重试会插出双版本行,要么临时件永远堆在盘上。
*/
class SoftwareVersionServiceChunkConfirmTest {
private final SoftwareVersionMapper versionMapper = mock(SoftwareVersionMapper.class);
private final OssStorageService ossStorageService = mock(OssStorageService.class);
private final LocalFileStorageService localFiles = mock(LocalFileStorageService.class);
@TempDir
Path tempDir;
@BeforeAll
static void initializeMybatisMetadata() {
TableInfoHelper.initTableInfo(
new MapperBuilderAssistant(new MybatisConfiguration(), ""), SoftwareVersionEntity.class);
}
private SoftwareVersionService service() {
SoftwareVersionService service = new SoftwareVersionService(versionMapper, ossStorageService);
ReflectionTestUtils.setField(service, "localFileStorageService", localFiles);
return service;
}
@Test
void 分片确认搬桶落库并标记最新且临时件清掉() throws Exception {
Path zip = writeZip("4.0.69.zip");
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(zip.toFile());
when(ossStorageService.uploadSoftwareVersionPackage(any(), any()))
.thenReturn("https://oss.aishufu.top/client/nanri-image/versions/4.0.69.zip");
when(versionMapper.insert(any(SoftwareVersionEntity.class))).thenAnswer(invocation -> {
((SoftwareVersionEntity) invocation.getArgument(0)).setId(300L);
return 1;
});
when(versionMapper.selectById(300L)).thenReturn(null);
service().confirmSoftwareVersion("4.0.69", "chunk-abc");
ArgumentCaptor<SoftwareVersionEntity> captor = ArgumentCaptor.forClass(SoftwareVersionEntity.class);
verify(versionMapper).insert(captor.capture());
assertEquals("4.0.69", captor.getValue().getVersion());
assertNotEquals("", captor.getValue().getSha256(), "落库必须带拼装件的 sha256");
verify(versionMapper).markLatestById(300L);
assertFalse(Files.exists(zip), "确认成功后临时拼装件必须删掉,否则每次重传都堆一份");
}
@Test
void 同版本重复确认命中幂等不重复搬桶() throws Exception {
Path zip = writeZip("4.0.69.zip");
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(zip.toFile());
SoftwareVersionEntity existing = new SoftwareVersionEntity();
existing.setId(9L);
existing.setVersion("4.0.69");
when(versionMapper.selectOne(any())).thenReturn(existing);
service().confirmSoftwareVersion("4.0.69", "chunk-abc");
verify(ossStorageService, never()).uploadSoftwareVersionPackage(any(), any());
verify(versionMapper, never()).insert(any(SoftwareVersionEntity.class));
verify(versionMapper, never()).markLatestById(any());
assertFalse(Files.exists(zip), "幂等命中也要把临时件清掉");
}
@Test
void 拼装件缺失必须拒绝且不落库() {
when(localFiles.findLocalSourceFile("chunk-gone")).thenReturn(null);
assertThrows(RuntimeException.class,
() -> service().confirmSoftwareVersion("4.0.69", "chunk-gone"));
verify(versionMapper, never()).insert(any(SoftwareVersionEntity.class));
}
@Test
void 非zip拼装件必须拒绝() throws Exception {
Path notZip = writeTempFile("payload.txt", "not a zip");
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(notZip.toFile());
assertThrows(RuntimeException.class,
() -> service().confirmSoftwareVersion("4.0.69", "chunk-abc"));
verify(ossStorageService, never()).uploadSoftwareVersionPackage(any(), any());
}
@Test
void 零字节拼装件必须拒绝() throws Exception {
Path empty = Files.createFile(tempDir.resolve("4.0.69.zip"));
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(empty.toFile());
assertThrows(RuntimeException.class,
() -> service().confirmSoftwareVersion("4.0.69", "chunk-abc"));
assertTrue(Files.exists(empty), "拒绝分支不许动调用方的临时件(排查要用)");
}
/** P2-4 起 confirm 会读 zip 中央目录,测试夹具必须是真 zip 而不是 PK 魔数文本。 */
private Path writeZip(String name) throws Exception {
Path file = tempDir.resolve(name);
try (ZipOutputStream out = new ZipOutputStream(Files.newOutputStream(file))) {
out.putNextEntry(new ZipEntry("readme.txt"));
out.write("package".getBytes(StandardCharsets.UTF_8));
out.closeEntry();
}
return file;
}
private Path writeTempFile(String name, String content) throws Exception {
Path file = tempDir.resolve(name);
Files.writeString(file, content, StandardCharsets.UTF_8);
return file;
}
}
@@ -0,0 +1,157 @@
package com.nanri.aiimage.modules.task.mapper;
import com.baomidou.mybatisplus.core.MybatisConfiguration;
import org.apache.ibatis.builder.xml.XMLMapperBuilder;
import org.apache.ibatis.io.Resources;
import org.apache.ibatis.mapping.MappedStatement;
import org.apache.ibatis.session.Configuration;
import org.junit.jupiter.api.Test;
import java.io.InputStream;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.time.LocalDateTime;
import java.util.List;
import java.util.Map;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* V183 采集通道的谓词形状(CR 第二轮 P1-1)。
*
* <p><b>为什么钉 SQL 而不是钉行为</b>:这两条查询的执行方判定必须在 {@code WHERE} 里、
* 在 {@code LIMIT} 之前。写成「先 limit 取页、回 Java 再 {@code filter(isLocal())}」时,
* 混通道排队里只要较老的 N 条(出厂 {@code client-task-pull.limit=5})是 ONLINE,
* 本机通道的单本轮及以后都领不到——不是延后,是排到队尾之前永久饿死。
* 而 mock 掉 {@code selectList} 的返回让「取页与过滤的先后」在用例里根本看不见,
* 这就是它活过后端整量的原因(现网 {@code enabled=false} 也没能掩盖住)。
* 这里用真 {@code XMLMapperBuilder} 解析、向真 {@code BoundSql} 断言,判据=
* 把谓词挪出 {@code WHERE} 或挪到 {@code LIMIT} 之后,本格直接红。
*
* <p>没有数据库也成立的另一面:{@code ?} 的个数与顺序被钉死,抽掉通道谓词就少两个占位符。
*/
class FileTaskMapperChannelSqlTest {
private static final String XML = "mapper/task/FileTaskMapper.xml";
private static final String NS = "com.nanri.aiimage.modules.task.mapper.FileTaskMapper";
private static final List<String> MODULE_TYPES = List.of("APPEARANCE_PATENT", "SIMILAR_ASIN");
private static final List<String> LOCAL_DEFAULTS = List.of("APPEARANCE_PATENT", "SIMILAR_ASIN");
@Test
void 客户端拉取的通道谓词在WHERE里且发生在LIMIT之前() throws Exception {
String sql = render("selectLocalChannelPendingTasks", clientParams(2));
int at = sql.indexOf("CASE WHEN collect_channel IN ('LOCAL', 'ONLINE')");
assertTrue(at > 0, () -> "通道判定不在 SQL 里(回到 Java filter 就是 P1-1 复发): " + sql);
int limit = sql.lastIndexOf("LIMIT ?");
assertTrue(limit > at, () -> "通道谓词必须出现在 LIMIT 之前,否则取到的那一页里没有本机单: " + sql);
assertTrue(sql.indexOf("ORDER BY created_at ASC") < limit,
() -> "排序必须在 LIMIT 之前,否则取页不是「最早创建的本机单」: " + sql);
assertEquals("SELECT id, module_type, collect_channel, created_at FROM biz_file_task"
+ " WHERE user_id = ? AND status = ? AND module_type IN (?, ?) AND created_at < ?"
+ " AND CASE WHEN collect_channel IN ('LOCAL', 'ONLINE') THEN collect_channel"
+ " WHEN module_type IN (?, ?) THEN 'LOCAL' ELSE 'ONLINE' END = 'LOCAL'"
+ " ORDER BY created_at ASC LIMIT ?", sql,
"取页形状变了:通道谓词必须与 status/模块/时间窗口同在 WHERE 里,且 LIMIT 之后没有任何过滤面");
}
@Test
void 内部认领把本机让位做成SQL侧OR窗口而不是Java二次过滤() throws Exception {
String sql = render("selectInternalClaimCandidates", internalParams(2));
assertTrue(sql.contains("(NOT (CASE") && sql.contains("= 'LOCAL') OR created_at <= ?)"),
() -> "让位窗口不在 SQL 里:整页都是「本机通道且没等满兜底窗口」时,本轮线上单零吞吐: " + sql);
assertTrue(sql.indexOf("created_at < ?") < sql.indexOf("created_at <= ?"),
() -> "基础等待窗口与兜底窗口必须都在 LIMIT 之前: " + sql);
assertTrue(sql.lastIndexOf("created_at <= ?") < sql.lastIndexOf("LIMIT ?"),
() -> "兜底窗口谓词必须发生在 LIMIT 之前: " + sql);
}
@Test
void 通道判定用CASE兜模块默认而不是裸等值() throws Exception {
String sql = render("selectLocalChannelPendingTasks", clientParams(2));
// 裸等值(collect_channel = 'LOCAL' 或 NOT (collect_channel = 'LOCAL'))在列为 NULL 时吃三值
// 逻辑变成 UNKNOWN:被默认兜成 ONLINE 的模块(COLLECT_DATA 等)会在 NOT 之后永久领不到。
assertFalse(sql.contains("collect_channel = ?"), () -> "通道值被参数化后 CASE 的兜法就散了: " + sql);
assertFalse(sql.contains("collect_channel = 'LOCAL' OR"),
() -> "OR 形状的裸等值谓词对 NULL 列失明: " + sql);
assertTrue(sql.contains("ELSE 'ONLINE' END"), () -> "CASE 缺 ELSE=不可识别值不再兜默认: " + sql);
}
@Test
void 取页投影只有认领需要的四列() throws Exception {
String sql = render("selectLocalChannelPendingTasks", clientParams(2));
assertTrue(sql.startsWith("SELECT id, module_type, collect_channel, created_at FROM biz_file_task"),
() -> "投影列清单变了(整行拉回来会把 result_json 这类大列带进兜底轮询): " + sql);
}
@Test
void 占位符数与两条语句的谓词形状逐一对撞() throws Exception {
// 客户端:user_id + status + 2 模块 + cutoff + 2 默认模块 + limit = 8
assertEquals(8, questionMarks("selectLocalChannelPendingTasks", clientParams(2)),
"少 2 个就是通道谓词被抽掉了(P1-1 复发),多 2 个是模块清单被写了两遍");
// 内部认领:status + 2 模块 + cutoff + 2 默认模块 + localCutoff + limit = 8
assertEquals(8, questionMarks("selectInternalClaimCandidates", internalParams(2)),
"内部认领的兜底窗口比客户端多一条 cutoff、少一个 user_id");
}
@Test
void V184补齐两条取页路径的索引并明写通道列不进索引() throws Exception {
Path migration = moduleRoot().resolve("src/main/resources/db/V184__collect_pull_and_evidence_indexes.sql");
assertTrue(Files.exists(migration), "V184 缺失:P1-1 的下推查询没有索引承接,P2-1 的保留期扫描仍全表");
String sql = Files.readString(migration, StandardCharsets.UTF_8);
assertTrue(sql.contains("ADD INDEX idx_biz_file_task_status_module_created (status, module_type, created_at)"),
() -> "认领取页索引的列组合变了: " + sql);
assertTrue(sql.contains("ADD INDEX idx_task_evidence_created_at (created_at)"),
() -> "证据保留期是单列 created_at 谓词,前导列不符就还是全表扫: " + sql);
assertFalse(sql.contains("idx_task_collect_channel"),
"通道列有意不进索引:CASE 表达式谓词不吃索引,把 collect_channel 夹在 module_type 与"
+ " created_at 之间反而打断 ORDER BY created_at 的索引有序性");
}
private static int questionMarks(String id, Map<String, Object> params) throws Exception {
return (int) render(id, params).chars().filter(c -> c == '?').count();
}
private static String render(String id, Map<String, Object> params) throws Exception {
MappedStatement statement = parse().getMappedStatement(NS + "." + id);
// MyBatis 把 foreach 展开成 `IN ( ? , ? )`,这里收成 `IN (?,?)` 好让判据读得下去。
return statement.getBoundSql(params).getSql().replaceAll("\\s+", " ").trim()
.replace("( ", "(").replace(" )", ")").replace(" ,", ",");
}
private static Map<String, Object> clientParams(int modules) {
return Map.of("userId", 1095L,
"status", "PENDING",
"moduleTypes", MODULE_TYPES.subList(0, modules),
"localDefaultModules", LOCAL_DEFAULTS,
"cutoff", LocalDateTime.now().minusMinutes(5),
"limit", 5);
}
private static Map<String, Object> internalParams(int modules) {
return Map.of("status", "PENDING",
"moduleTypes", MODULE_TYPES.subList(0, modules),
"localDefaultModules", LOCAL_DEFAULTS,
"cutoff", LocalDateTime.now().minusMinutes(2),
"localCutoff", LocalDateTime.now().minusMinutes(30),
"limit", 5);
}
private static Configuration parse() throws Exception {
MybatisConfiguration configuration = new MybatisConfiguration();
try (InputStream in = Resources.getResourceAsStream(XML)) {
new XMLMapperBuilder(in, configuration, XML, configuration.getSqlFragments()).parse();
}
return configuration;
}
private static Path moduleRoot() {
Path here = Path.of("").toAbsolutePath();
return Files.isDirectory(here.resolve("src/main/resources")) ? here : here.getParent();
}
}
@@ -3,9 +3,12 @@ package com.nanri.aiimage.modules.task.model;
import com.nanri.aiimage.config.CollectRoutingProperties;
import org.junit.jupiter.api.Test;
import java.util.List;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.junit.jupiter.api.Assertions.assertNull;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* V183 任务级采集通道的判据(建单写入、认领读取、脏值兜底)。
@@ -27,6 +30,24 @@ class CollectChannelTest {
assertFalse(CollectChannel.defaultFor(" ").isLocal());
}
/**
* P1-1 把通道判定下推进 SQL 之后,「哪些模块默认本机」这份清单被喂进 mapper 的 CASE。
* 判据=清单内必被 {@link CollectChannel#defaultFor} 读作本机、清单外必不读作本机。
* 谁在 defaultFor 里另加一个模块而忘了同步清单,SQL 就会把那模块的历史单兜成线上:
* 界面显示「本机采集」、客户端却永远领不到。
*/
@Test
void 本机默认模块清单与defaultFor同源() {
assertEquals(List.of(CollectRoutingProperties.MODULE_APPEARANCE_PATENT,
CollectRoutingProperties.MODULE_SIMILAR_ASIN), CollectChannel.localDefaultModuleTypes());
for (String module : CollectChannel.localDefaultModuleTypes()) {
assertTrue(CollectChannel.defaultFor(module).isLocal(), module + " 在清单里却不被判本机");
}
for (String module : List.of("DEDUPE", "COLLECT_DATA", "SHOP_DATA_CRAWL", "PUBLISH")) {
assertFalse(CollectChannel.defaultFor(module).isLocal(), module + " 不在清单里却读成本机");
}
}
@Test
void 模块默认大小写与空白不影响判定() {
assertEquals(CollectChannel.LOCAL, CollectChannel.defaultFor("similar_asin"));
@@ -9,6 +9,7 @@ import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.baomidou.mybatisplus.core.metadata.TableInfoHelper;
import com.nanri.aiimage.config.InstanceMetadata;
import com.nanri.aiimage.modules.task.mapper.FileTaskMapper;
import com.nanri.aiimage.modules.task.model.CollectChannel;
import com.nanri.aiimage.modules.task.model.entity.FileTaskEntity;
import com.nanri.aiimage.modules.task.model.vo.TaskClientPullVo;
import com.nanri.aiimage.modules.task.spi.ClientTaskPullSpi;
@@ -29,7 +30,9 @@ import static org.junit.jupiter.api.Assertions.assertNotNull;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.ArgumentMatchers.anyInt;
import static org.mockito.ArgumentMatchers.anyLong;
import static org.mockito.ArgumentMatchers.eq;
import static org.mockito.ArgumentMatchers.isNull;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.never;
@@ -68,7 +71,7 @@ class TaskClientPullServiceTest {
@Test
void 认领成功后返回载荷并写_RUNNING_与_owner() {
FileTaskEntity candidate = task(11L, "SIMILAR_ASIN");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(candidate));
stubPullCandidates(candidate);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(candidate);
StubHandler handler = new StubHandler("SIMILAR_ASIN", payload("similar-asin-run"));
@@ -88,17 +91,16 @@ class TaskClientPullServiceTest {
assertTrue(params.containsValue("RUNNING"), "认领必须写入 RUNNING");
assertTrue(sqlSetOf(captor.getValue()).contains("owner_instance_id"), "认领必须接管归属实例");
ArgumentCaptor<Wrapper<FileTaskEntity>> selectCaptor = wrapperCaptor();
verify(fileTaskMapper).selectList(selectCaptor.capture());
Map<String, Object> selectParams = paramsOf(selectCaptor.getValue());
assertTrue(selectParams.containsValue("PENDING"), "候选只挑 PENDING");
assertTrue(selectParams.containsValue(USER_ID), "候选限定任务所属用户");
assertTrue(selectCaptor.getValue().getSqlSegment().contains("created_at"), "候选要有创建时间窗口");
// 取页形状(通道谓词在 WHERE、LIMIT 之前)由 FileTaskMapperChannelSqlTest 向真 BoundSql 断言;
// 这里钉「调用的是哪条具名查询、传了什么」,两者合起来才挡住 P1-1 复发。
verify(fileTaskMapper).selectLocalChannelPendingTasks(eq(USER_ID), eq("PENDING"),
eq(List.of("SIMILAR_ASIN", "COLLECT_DATA")), eq(CollectChannel.localDefaultModuleTypes()),
any(LocalDateTime.class), eq(5));
}
@Test
void 页面抢先认领后不再返回任务() {
when(fileTaskMapper.selectList(any())).thenReturn(List.of(task(11L, "SIMILAR_ASIN")));
stubPullCandidates(task(11L, "SIMILAR_ASIN"));
when(fileTaskMapper.update(isNull(), any())).thenReturn(0);
StubHandler handler = new StubHandler("SIMILAR_ASIN", payload("similar-asin-run"));
@@ -112,7 +114,7 @@ class TaskClientPullServiceTest {
@Test
void 组装不出载荷时任务标失败() {
FileTaskEntity candidate = task(11L, "SIMILAR_ASIN");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(candidate));
stubPullCandidates(candidate);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(candidate);
@@ -127,7 +129,7 @@ class TaskClientPullServiceTest {
@Test
void 组装抛异常时任务标失败且不抛出() {
FileTaskEntity candidate = task(11L, "SIMILAR_ASIN");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(candidate));
stubPullCandidates(candidate);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(candidate);
StubHandler broken = new StubHandler("SIMILAR_ASIN", null, new IllegalStateException("boom"));
@@ -142,7 +144,7 @@ class TaskClientPullServiceTest {
@Test
void 白名单模块未实现兜底时跳过且不改状态() {
when(fileTaskMapper.selectList(any())).thenReturn(List.of(task(21L, "COLLECT_DATA")));
stubPullCandidates(task(21L, "COLLECT_DATA"));
List<TaskClientPullVo> result = service(new StubHandler("SIMILAR_ASIN", payload("similar-asin-run")))
.pullPendingTasks(USER_ID);
@@ -190,86 +192,97 @@ class TaskClientPullServiceTest {
}
}
// ------------------------------------------------------------ 本地采集路由(2026-10-29)
// ---------------------------------------------------- 本地采集路由(2026-10-29 建档,2026-10-31 下推 SQL)
/**
* 客户端拉取只走带通道谓词的具名查询,服务侧不再挑第二遍。
*
* <p>这里喂一条通道写着 ONLINE 的行——真 SQL 不会返回它——为的是钉住「Java 里没有第二道
* {@code filter(isLocal())}」。P1-1 的原形状就是先 {@code limit 5} 取页再过滤:那一道过滤
* 看不见 limit,也就没人替它保证「先筛后取页」,混通道时较老的 N 条是线上单就把本机单永久
* 挡在队尾之后。谓词本身的位置由 {@code FileTaskMapperChannelSqlTest} 向真 BoundSql 钉。
*/
@Test
void 客户端兜底拉取不领线上通道的单() {
FileTaskEntity online = task(11L, "SIMILAR_ASIN", 10L);
online.setCollectChannel("ONLINE");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(online));
void 客户端拉取把通道判定整个交给SQL不做二次过滤() {
FileTaskEntity row = task(11L, "SIMILAR_ASIN", 10L);
row.setCollectChannel("ONLINE");
stubPullCandidates(row);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(row);
List<TaskClientPullVo> result = serviceWithRouting(defaultRouting(),
new StubHandler("SIMILAR_ASIN", payload("similar-asin-run"))).pullPendingTasks(USER_ID);
assertTrue(result.isEmpty(), "这一单选了线上采集,客户端领了就是抢跑");
verify(fileTaskMapper, never()).update(isNull(), any());
assertEquals(1, result.size(), "候选页由 SQL 侧谓词筛好,服务再判一遍就是 P1-1 复发");
}
/**
* 内部认领同样把让位判定交给 SQL:喂一条「还在让位窗口内的本机单」,服务必须照领。
*/
@Test
void 客户端兜底拉取照领本机通道的单() {
FileTaskEntity local = task(11L, "SIMILAR_ASIN", 10L);
local.setCollectChannel("LOCAL");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(local));
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(local);
List<TaskClientPullVo> result = serviceWithRouting(defaultRouting(),
new StubHandler("SIMILAR_ASIN", payload("similar-asin-run"))).pullPendingTasks(USER_ID);
assertEquals(1, result.size(), "本机通道的单必须能被在线客户端领走");
}
@Test
void 内部认领对本机通道让位到兜底窗口() {
void 内部认领不在Java里做让位过滤() {
com.nanri.aiimage.config.CollectRoutingProperties routing =
new com.nanri.aiimage.config.CollectRoutingProperties();
routing.setLinuxFallbackMinutes(30L);
FileTaskEntity localTask = task(11L, "SIMILAR_ASIN", 10L);
localTask.setCollectChannel("LOCAL");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(localTask));
List<TaskClientPullVo> result = serviceWithRouting(routing,
new StubHandler("SIMILAR_ASIN", payload("similar-asin-run")))
.pullPendingTasksInternal(List.of("SIMILAR_ASIN"));
assertTrue(result.isEmpty(), "本机通道的单在兜底窗口内必须让位给本地客户端");
verify(fileTaskMapper, never()).update(isNull(), any());
}
@Test
void 内部认领超过兜底窗口后接管本机通道的单() {
com.nanri.aiimage.config.CollectRoutingProperties routing =
new com.nanri.aiimage.config.CollectRoutingProperties();
routing.setLinuxFallbackMinutes(30L);
FileTaskEntity candidate = task(11L, "SIMILAR_ASIN", 40L);
candidate.setCollectChannel("LOCAL");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(candidate));
FileTaskEntity row = task(21L, "COLLECT_DATA", 10L);
row.setCollectChannel("LOCAL");
stubInternalCandidates(row);
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(11L)).thenReturn(candidate);
List<TaskClientPullVo> result = serviceWithRouting(routing,
new StubHandler("SIMILAR_ASIN", payload("similar-asin-run")))
.pullPendingTasksInternal(List.of("SIMILAR_ASIN"));
assertEquals(1, result.size(), "本机长时间没人领,Linux 必须兜底接管");
}
@Test
void 内部认领对线上通道不受兜底窗口影响() {
com.nanri.aiimage.config.CollectRoutingProperties routing =
new com.nanri.aiimage.config.CollectRoutingProperties();
routing.setLinuxFallbackMinutes(30L);
FileTaskEntity candidate = task(21L, "COLLECT_DATA", 10L);
candidate.setCollectChannel("ONLINE");
when(fileTaskMapper.selectList(any())).thenReturn(List.of(candidate));
when(fileTaskMapper.update(isNull(), any())).thenReturn(1);
when(fileTaskMapper.selectById(21L)).thenReturn(candidate);
when(fileTaskMapper.selectById(21L)).thenReturn(row);
List<TaskClientPullVo> result = serviceWithRouting(routing,
new StubHandler("COLLECT_DATA", payload("collect-data-run")))
.pullPendingTasksInternal(List.of("COLLECT_DATA"));
assertEquals(1, result.size(), "选了线上的单子不能被本地优先的窗口拖住");
assertEquals(1, result.size(), "让位窗口是 SQL 的 OR 分支,不是取页之后的 Java 过滤");
}
/**
* 两道窗口分别下传:基础等待窗口(给「建完就想删」留余地)与本机兜底窗口(Linux 接管点)。
* 兜底窗口配得比基础窗口还小时取基础窗口——这条夹逼原先在 Java 里,下推后仍得由服务算好。
*/
@Test
void 内部认领下传两道cutoff且兜底窗口不越基础窗口() {
com.nanri.aiimage.config.CollectRoutingProperties narrow =
new com.nanri.aiimage.config.CollectRoutingProperties();
narrow.setLinuxFallbackMinutes(1L);
serviceWithRouting(narrow, new StubHandler("SIMILAR_ASIN", payload("similar-asin-run")))
.pullPendingTasksInternal(List.of("SIMILAR_ASIN"));
ArgumentCaptor<LocalDateTime> cutoff = ArgumentCaptor.forClass(LocalDateTime.class);
ArgumentCaptor<LocalDateTime> localCutoff = ArgumentCaptor.forClass(LocalDateTime.class);
verify(fileTaskMapper).selectInternalClaimCandidates(eq("PENDING"), eq(List.of("SIMILAR_ASIN")),
eq(CollectChannel.localDefaultModuleTypes()), cutoff.capture(), localCutoff.capture(), eq(5));
assertEquals(cutoff.getValue(), localCutoff.getValue(),
"兜底窗口 1 分钟短于基础窗口 2 分钟时按 2 分钟算,否则「建完就能删」的余地被越过");
}
@Test
void 兜底窗口长于基础窗口时按自身下传() {
com.nanri.aiimage.config.CollectRoutingProperties wide =
new com.nanri.aiimage.config.CollectRoutingProperties();
wide.setLinuxFallbackMinutes(30L);
serviceWithRouting(wide, new StubHandler("SIMILAR_ASIN", payload("similar-asin-run")))
.pullPendingTasksInternal(List.of("SIMILAR_ASIN"));
ArgumentCaptor<LocalDateTime> cutoff = ArgumentCaptor.forClass(LocalDateTime.class);
ArgumentCaptor<LocalDateTime> localCutoff = ArgumentCaptor.forClass(LocalDateTime.class);
verify(fileTaskMapper).selectInternalClaimCandidates(eq("PENDING"), eq(List.of("SIMILAR_ASIN")),
eq(CollectChannel.localDefaultModuleTypes()), cutoff.capture(), localCutoff.capture(), eq(5));
assertTrue(localCutoff.getValue().isBefore(cutoff.getValue()),
"30 分钟的让位窗口必须比 2 分钟的基础窗口更早,否则 Linux 会抢在本地客户端前面双跑");
}
private void stubPullCandidates(FileTaskEntity... rows) {
when(fileTaskMapper.selectLocalChannelPendingTasks(any(), any(), any(), any(), any(), anyInt()))
.thenReturn(List.of(rows));
}
private void stubInternalCandidates(FileTaskEntity... rows) {
when(fileTaskMapper.selectInternalClaimCandidates(any(), any(), any(), any(), any(), anyInt()))
.thenReturn(List.of(rows));
}
private TaskClientPullService service(ClientTaskPullSpi... handlers) {
@@ -21,7 +21,7 @@ import java.util.List;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.junit.jupiter.api.Assertions.assertNull;
import static org.junit.jupiter.api.Assertions.assertNotNull;
import static org.junit.jupiter.api.Assertions.assertTrue;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.ArgumentMatchers.isNull;
@@ -76,29 +76,33 @@ class TaskFileJobTerminalTaskGuardTest {
}
@Test
void claimRunningRejectsJobWhoseTaskIsFailed() {
TaskFileJobEntity failedTaskJob = job(11L, 500L);
when(taskFileJobMapper.selectById(11L)).thenReturn(failedTaskJob);
void claimRunningStillDeliversWhenParentTaskIsTerminal() {
TaskFileJobEntity terminalTaskJob = job(11L, 500L);
TaskFileJobEntity running = job(11L, 500L);
running.setStatus("RUNNING");
when(taskFileJobMapper.selectById(11L)).thenReturn(terminalTaskJob, running);
when(taskFileJobMapper.selectFileTaskStatus(500L)).thenReturn("FAILED");
when(taskFileJobMapper.update(isNull(), any(LambdaUpdateWrapper.class))).thenReturn(1);
assertNull(service.claimRunning(11L), "终态任务的作业不得被 claim 成 RUNNING");
TaskFileJobEntity claimed = service.claimRunning(11L);
assertNotNull(claimed, "P1-13:父任务终态也要放行出表作业");
assertEquals("RUNNING", claimed.getStatus());
@SuppressWarnings("unchecked")
ArgumentCaptor<LambdaUpdateWrapper<TaskFileJobEntity>> captor =
ArgumentCaptor.forClass(LambdaUpdateWrapper.class);
verify(taskFileJobMapper, times(1)).update(isNull(), captor.capture());
LambdaUpdateWrapper<TaskFileJobEntity> wrapper = captor.getValue();
assertTrue(hasParam(wrapper, "FAILED"), "必须把作业拉成 FAILED,下一轮扫描才捞不到");
assertTrue(hasParam(wrapper, 5), "retry_count 必须一次拉满到上限");
assertFalse(hasParam(wrapper, "RUNNING"), "不得再翻转成 RUNNING");
assertTrue(hasParam(wrapper, "RUNNING"), "终态父任务不再把出表作业拉成 FAILED");
assertFalse(hasParam(wrapper, "FAILED"), "终态只停心跳续期,不改写给出表作业");
assertTrue(wrapper.getSqlSet().contains("terminal_callback_at"),
"terminal_callback_at 必须显式清空:留给兜底扫描出部分结果,失败任务仍可下载");
assertTrue(hasParam(wrapper, null), "terminal_callback_at 置空后兜底扫描才捞得到这一行");
"正常认领必须清空 terminal_callback_at,失败兜底扫描才捞得到");
assertTrue(hasParam(wrapper, null));
}
@Test
void claimCandidatesSkipsTerminalTaskAndKeepsActiveOne() {
void claimCandidatesClaimsBothTerminalAndActiveTasks() {
TaskFileJobEntity terminal = job(21L, 500L);
TaskFileJobEntity active = job(22L, 501L);
when(taskFileJobMapper.selectList(any(LambdaQueryWrapper.class)))
@@ -106,23 +110,25 @@ class TaskFileJobTerminalTaskGuardTest {
when(taskFileJobMapper.selectFileTaskStatus(500L)).thenReturn("FAILED");
when(taskFileJobMapper.selectFileTaskStatus(501L)).thenReturn("RUNNING");
when(taskFileJobMapper.update(isNull(), any(LambdaUpdateWrapper.class))).thenReturn(1, 1);
TaskFileJobEntity claimedTerminal = job(21L, 500L);
claimedTerminal.setStatus("RUNNING");
TaskFileJobEntity claimedActive = job(22L, 501L);
claimedActive.setStatus("RUNNING");
when(taskFileJobMapper.selectBatchIds(any())).thenReturn(List.of(claimedActive));
when(taskFileJobMapper.selectBatchIds(any())).thenReturn(List.of(claimedTerminal, claimedActive));
List<TaskFileJobEntity> claimed = service.claimRunnableJobs(10);
assertEquals(1, claimed.size(), "终态任务的 job 不入结果集");
assertEquals(22L, claimed.get(0).getId());
assertEquals(2, claimed.size(), "P1-13:终态任务的出表作业与活跃任务一样可认领");
assertEquals(21L, claimed.get(0).getId());
assertEquals(22L, claimed.get(1).getId());
@SuppressWarnings("unchecked")
ArgumentCaptor<LambdaUpdateWrapper<TaskFileJobEntity>> captor =
ArgumentCaptor.forClass(LambdaUpdateWrapper.class);
verify(taskFileJobMapper, times(2)).update(isNull(), captor.capture());
assertTrue(hasParam(captor.getAllValues().get(0), "FAILED"), "第一笔是终态收尾");
assertTrue(hasParam(captor.getAllValues().get(1), "RUNNING"), "第二笔才是正常 claim");
assertTrue(captor.getAllValues().get(0).getSqlSet().contains("terminal_callback_at"),
"终态作业必须留给兜底扫描,不能提前打终态回调");
assertTrue(hasParam(captor.getAllValues().get(0), "RUNNING"), "终态任务的 job 也要正常 claim");
assertTrue(hasParam(captor.getAllValues().get(1), "RUNNING"));
assertFalse(hasParam(captor.getAllValues().get(0), "FAILED"), "不得再把终态任务的 job 判死");
}
@Test
@@ -425,7 +425,7 @@ class TaskResultFileJobWorkerStuckScanTest {
worker.finishAfterExecution(job, new java.util.concurrent.atomic.AtomicBoolean(true));
verify(taskFileJobService).markFailedPermanentAwaitingFallback(job, "父任务已终态,结果作业不再执行");
verify(taskFileJobService).markFailedPermanentAwaitingFallback(job, "父任务已终态且本次执行未产出结果文件");
}
/**
@@ -0,0 +1,111 @@
package com.nanri.aiimage.modules.tutorial.service;
import com.baomidou.mybatisplus.core.MybatisConfiguration;
import com.baomidou.mybatisplus.core.metadata.TableInfoHelper;
import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
import com.nanri.aiimage.modules.file.service.oss.OssStorageService;
import com.nanri.aiimage.modules.tutorial.mapper.TutorialPackageMapper;
import com.nanri.aiimage.modules.tutorial.model.entity.TutorialPackageEntity;
import org.apache.ibatis.builder.MapperBuilderAssistant;
import org.junit.jupiter.api.BeforeAll;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import org.mockito.ArgumentCaptor;
import org.springframework.test.util.ReflectionTestUtils;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.zip.ZipEntry;
import java.util.zip.ZipOutputStream;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.ArgumentMatchers.anyString;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.never;
import static org.mockito.Mockito.verify;
import static org.mockito.Mockito.when;
/**
* 教程包「分片拼装件 → confirm 搬桶落库」的契约(CR 第二轮 P1-8,与版本包同批)。
*
* <p>服务端新重载与控制器接线都在,缺的是执行证据:幂等命中(同 fileName+version 重试只
* 返回既有行)、成功后临时件清理、缺件拒绝。任何一条破掉,重试会重复搬桶或临时件堆盘。
*/
class TutorialPackageServiceChunkConfirmTest {
private final TutorialPackageMapper tutorialPackageMapper = mock(TutorialPackageMapper.class);
private final OssStorageService ossStorageService = mock(OssStorageService.class);
private final LocalFileStorageService localFiles = mock(LocalFileStorageService.class);
@TempDir
Path tempDir;
@BeforeAll
static void initializeMybatisMetadata() {
TableInfoHelper.initTableInfo(
new MapperBuilderAssistant(new MybatisConfiguration(), ""), TutorialPackageEntity.class);
}
private TutorialPackageService service() {
TutorialPackageService service = new TutorialPackageService(tutorialPackageMapper, ossStorageService);
ReflectionTestUtils.setField(service, "localFileStorageService", localFiles);
return service;
}
@Test
void 分片确认搬桶落库且临时件清掉() throws Exception {
Path zip = writeZip("教程包.zip");
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(zip.toFile());
when(tutorialPackageMapper.selectOne(any())).thenReturn(null);
when(tutorialPackageMapper.insert(any(TutorialPackageEntity.class))).thenReturn(1);
when(tutorialPackageMapper.selectById(any())).thenReturn(null);
service().confirmTutorialPackage("tutorial/direct.zip", "教程包.zip", "2026.10", "chunk-abc");
ArgumentCaptor<TutorialPackageEntity> captor = ArgumentCaptor.forClass(TutorialPackageEntity.class);
verify(tutorialPackageMapper).insert(captor.capture());
assertEquals("教程包.zip", captor.getValue().getFileName());
verify(ossStorageService).uploadTutorialPackage(any(), anyString());
assertFalse(Files.exists(zip), "确认成功后临时拼装件必须删掉");
}
@Test
void 同名同版重复确认命中幂等不重复搬桶() throws Exception {
Path zip = writeZip("教程包.zip");
when(localFiles.findLocalSourceFile("chunk-abc")).thenReturn(zip.toFile());
TutorialPackageEntity existing = new TutorialPackageEntity();
existing.setId(11L);
existing.setFileName("教程包.zip");
when(tutorialPackageMapper.selectOne(any())).thenReturn(existing);
service().confirmTutorialPackage("tutorial/direct.zip", "教程包.zip", "2026.10", "chunk-abc");
verify(ossStorageService, never()).uploadTutorialPackage(any(), anyString());
verify(tutorialPackageMapper, never()).insert(any(TutorialPackageEntity.class));
assertFalse(Files.exists(zip), "幂等命中也要把临时件清掉");
}
@Test
void 拼装件缺失必须拒绝且不落库() {
when(localFiles.findLocalSourceFile("chunk-gone")).thenReturn(null);
assertThrows(RuntimeException.class,
() -> service().confirmTutorialPackage("tutorial/direct.zip", "教程包.zip", "2026.10", "chunk-gone"));
verify(tutorialPackageMapper, never()).insert(any(TutorialPackageEntity.class));
}
/** P2-4 起 confirm 会读 zip 中央目录,测试夹具必须是真 zip。 */
private Path writeZip(String name) throws Exception {
Path file = tempDir.resolve(name);
try (ZipOutputStream out = new ZipOutputStream(Files.newOutputStream(file))) {
out.putNextEntry(new ZipEntry("readme.txt"));
out.write("package".getBytes(StandardCharsets.UTF_8));
out.closeEntry();
}
return file;
}
}
+72 -10
View File
@@ -31,6 +31,14 @@ export const CHUNK_UPLOAD_THRESHOLD = UPLOAD_PART_SIZE
/** 每片重试次数(不含首次)。 */
const PART_RETRY_LIMIT = 2
/**
* 单次直传的超时预算(P3-5:这条尺子在本仓只有一个出口):30s 保底 + 每 MB 15s,
* 慢公网传 100MB 视频不再被固定 120s 掐死。桌面桥的浏览器降级直传也走这里。
*/
export function uploadTimeoutForBytes(bytes: number): number {
return 30_000 + Math.ceil(bytes / (1024 * 1024)) * 15_000
}
function uploadPartUrl(uploadId: string, index: number): string {
return `${JAVA_API_PREFIX}/files/chunk/${encodeURIComponent(uploadId)}/part/${index}`
}
@@ -98,6 +106,61 @@ function clearStoredSession(key: string): void {
}
}
/**
* 服务端对「这个会话还在不在」的三种答复:还在(顺带已到片号)/确实没了/问不出来。
* 第三态是 P1-11 的关键——过去它和前两态被揉进一个 catch,断网=整包作废。
*/
type ProbeOutcome
= { state: 'ok'; parts: number[] }
| { state: 'gone' }
| { state: 'transient' }
/** 后端回收会话的原文案(ChunkedUploadService.readMeta);BusinessException 不带 code ⇒ HTTP 200 出。 */
function messageSaysSessionGone(text: unknown): boolean {
const message = typeof text === 'string' ? text : ''
return ['不存在', '过期'].some((marker) => message.includes(marker))
}
/**
* 判据顺序=可信度顺序:404 与后端点名句才清状态;5xx/429/别的业务失败一律瞬态。
* 拿"问不到"当"没了",就是用一次网关抖动换整包重传。
*/
function classifyProbe(response: unknown): ProbeOutcome {
const status = (response as { status?: unknown } | null | undefined)?.status
const numeric = typeof status === 'number' ? status : -1
if (numeric === 404) return { state: 'gone' }
if (numeric >= 500 || numeric === 408 || numeric === 429) return { state: 'transient' }
const body = bodyOf<{ uploaded_parts?: unknown }>(response)
if (body && typeof body === 'object' && body.success === false) {
return messageSaysSessionGone(body.message) ? { state: 'gone' } : { state: 'transient' }
}
if (numeric !== -1 && (numeric < 200 || numeric > 299)) return { state: 'transient' }
const arrived = body?.data?.uploaded_parts
return { state: 'ok', parts: Array.isArray(arrived) ? arrived.map((index) => Number(index)) : [] }
}
/**
* 探测已达片号(P1-11)。
* validateStatus 恒真:本仓响应拦截器会把失败折叠成 `new Error(中文文案)`,状态码到不了这里,
* 404/5xx 全靠这一步自己按码分因;能走到 catch 的就只剩网络层失败与超时=铁定瞬态。
*/
async function probeUploadedParts(uploadId: string): Promise<ProbeOutcome> {
try {
const response = await http.get<JavaApiResponse<{ uploaded_parts?: unknown[] }>>(
`${JAVA_API_PREFIX}/files/chunk/${encodeURIComponent(uploadId)}`,
{ validateStatus: () => true },
)
return classifyProbe(response)
} catch (error) {
const status = (error as { response?: { status?: unknown } } | null | undefined)?.response?.status
if (status === 404) return { state: 'gone' }
if (messageSaysSessionGone((error as { message?: unknown } | null | undefined)?.message)) {
return { state: 'gone' }
}
return { state: 'transient' }
}
}
/** 单次 multipart 上传(小文件短链路)。超时按体积给,不再固定 120s。 */
async function singleUpload(
file: File,
@@ -122,8 +185,8 @@ async function singleUpload(
headers: {
'Content-Type': 'multipart/form-data',
},
// 30s 保底 + 每 MB 15s:慢公网传 100MB 视频不再被固定 120s 掐死(S1-F3)
timeout: 30_000 + Math.ceil(file.size / (1024 * 1024)) * 15_000,
// 超时按体积给(见 uploadTimeoutForBytes),不再固定 120s
timeout: uploadTimeoutForBytes(file.size),
onUploadProgress: (event) => {
if (!onProgress) return
const total = event.total || file.size
@@ -148,17 +211,16 @@ async function chunkedUpload(
let uploadId = readStoredSession(sessionKey, file)
let done = new Set<number>()
if (uploadId) {
try {
const status = ensureOk(bodyOf<{ uploaded_parts?: unknown[] }>(await http.get(
`${JAVA_API_PREFIX}/files/chunk/${encodeURIComponent(uploadId)}`,
)))
const arrived = status.data?.uploaded_parts
done = new Set<number>(Array.isArray(arrived) ? arrived.map((index) => Number(index)) : [])
} catch {
// 会话已被服务端回收/过期:只能清掉重开,否则这个件每次都在同一个死会话上失败。
const probed = await probeUploadedParts(uploadId)
if (probed.state === 'gone') {
// 服务端确实说会话没了(404/「不存在或已过期」):只能清掉重开,否则这个件每次都在同一个死会话上失败。
clearStoredSession(sessionKey)
uploadId = ''
} else if (probed.state === 'ok') {
done = new Set<number>(probed.parts)
}
// transient(断网/超时/5xx/别的业务失败):会话还在服务端,片号未知也要接着往里写——
// 同一片重复上传按覆盖处理(后端 savePart 幂等),重开 init 等于把已传上去的整包作废。
}
if (!uploadId) {
const init = ensureOk(bodyOf<{ upload_id?: string }>(await http.post(
@@ -9,6 +9,7 @@
import type { DesktopConfig, DesktopConfigUpdate, PywebviewApi } from './pywebview'
import { extractErrorMessage, http } from '@/shared/api/http'
import { uploadTimeoutForBytes } from '@/shared/api/upload.ts'
/**
* 虚拟路径 → 浏览器 File 对象(虚拟路径仅存活于当前页面会话,刷新即失效)
@@ -240,7 +241,10 @@ export const webFallbackApi: PywebviewApi = {
}
try {
// http 实例对 /newApi 前缀自动附带 Bearer;FormData 的 Content-Type 交由浏览器自动生成(含 boundary)
const response = await http.post('/newApi/api/files/upload', form, { timeout: 120000 })
// 超时按体积给(P3-5:与 upload.ts 同一把尺),几 MB 的 xlsx/zip 不再被固定 120s 掐死
const response = await http.post('/newApi/api/files/upload', form, {
timeout: uploadTimeoutForBytes(file.size),
})
// 消费完即释放:文件已落到服务端,页面后续只用返回的 fileKey/url,不再需要这份本地 File。
// 失败分支保留条目,让用户能在不重新选文件的情况下重试(超量时由 rememberWebFile 的 FIFO 淘汰兜底)。
if (response.data?.success) {
@@ -54,4 +54,40 @@ describe('AC-11 用户端大档上传', () => {
`AC-11:9MB 视频仍走单次上传,实际出口:${urls.join(', ')}`,
).toBe(true)
})
// P2-9:上面那格的桩直接把包体当返回值,bodyOf 的真实 axios 分支(包体在 .data)零行为覆盖。
// 形状误判的代价是静默的——upload_id 取不到就报"后端未返回分片会话",已达片号取不到就整包从 0 重传,
// 而桩版本照样绿。这一格按真机形状(AxiosResponse{status, data})对撞,缺牙就判红。
it('真实 axios 响应形状必须被认下来(P2-9 反桩形假绿)', async () => {
const uploadId = 'b'.repeat(32)
const envelope = (body: unknown) => ({
status: 200,
statusText: 'OK',
headers: {},
config: {},
data: { success: true, message: '', data: body },
})
mocks.post.mockImplementation((url: string) => Promise.resolve(
String(url).includes('/complete')
? envelope({
fileKey: 'tmp/big.mp4', originalFilename: 'big.mp4', localPath: '/tmp/big.mp4', size: 9437184,
url: 'https://oss.example/big.mp4', objectKey: 'image-video/big.mp4', mediaType: 'video',
})
: envelope({ upload_id: uploadId, total_parts: 2, received_parts: 1 }),
))
mocks.get.mockResolvedValue(envelope({ upload_id: uploadId, uploaded_parts: [] }))
const file = new File([new Uint8Array(9 * 1024 * 1024)], 'big.mp4', { type: 'video/mp4' })
const uploaded = await uploadImageVideoMedia(file)
expect(uploaded.objectKey, '包体在 .data 时仍要取到 complete 的结果').toBe('image-video/big.mp4')
const partUrls = mocks.post.mock.calls
.map(([url]) => String(url))
.filter((url) => /\/part\/\d+$/.test(url))
expect(partUrls.length, '9MB=2 片,一片都不能少').toBe(2)
expect(
partUrls.every((url) => url.includes(uploadId)),
`bodyOf 没按 axios 形状取 upload_id,实际出口:${partUrls.join(', ')}`,
).toBe(true)
})
})
@@ -0,0 +1,185 @@
// P1-11(AC-9/铁律 18「瞬态失败不得把会话打掉」的用户端那一半)。
//
// 评审实录:GET 已达片号那一步用无差别 catch ⇒ 断网/超时/5xx 一律 clearStoredSession + 重开
// upload_id,一次抖动就把 90MB 打回从 0。本文件按两个方向钉死:
// ① 服务端确实说会话没了(HTTP 404/现网形状 HTTP 200+success:false+「不存在或已过期」)⇒ 重开;
// ② 网络层失败/5xx/非会话类业务失败 ⇒ 保留 upload_id,绝不再发 init。
// 另附 P2-9:桩必须能吃真实 axios 形状(包体在 .data),否则 bodyOf 的 axios 分支零行为覆盖,
// 形状误判会让 done 成空集=静默从 0 而不判红。
//
// 跑法:npm run test:mount(tests/mount/** 走 vitest+happy-dom,localStorage 可用,不出网)。
import { beforeEach, describe, expect, it, vi } from 'vitest'
const mocks = vi.hoisted(() => ({
post: vi.fn(),
get: vi.fn(),
}))
vi.mock('@/shared/api/http.ts', () => ({
http: { post: mocks.post, get: mocks.get },
post: mocks.post,
get: mocks.get,
unwrapJavaResponse: (value: unknown) => value,
}))
import { UPLOAD_PART_SIZE, uploadTempFileToJava } from '@/shared/api/upload.ts'
const SESSION_ID = 's'.repeat(32)
const FRESH_ID = 'f'.repeat(32)
const TWO_PARTS = UPLOAD_PART_SIZE * 2
function sessionKeyOf(file: File): string {
return `shufu:chunk:${file.name}:${file.size}:${file.lastModified}:`
}
function seedSession(file: File, uploadId: string): void {
localStorage.setItem(sessionKeyOf(file), JSON.stringify({
uploadId,
size: file.size,
lastModified: file.lastModified,
}))
}
function storedUploadId(file: File): string {
const raw = localStorage.getItem(sessionKeyOf(file))
if (!raw) return ''
return String((JSON.parse(raw) as { uploadId?: unknown }).uploadId ?? '')
}
function makeBigFile(name = 'resume.mp4'): File {
return new File([new Uint8Array(TWO_PARTS)], name, { type: 'video/mp4' })
}
/** 真实 axios 出口形状:状态码与包体分家(测试桩过去直接把 body 当返回值,见 P2-9)。 */
function axiosResponse(status: number, body: unknown) {
return { status, data: body }
}
function javaOk(body: unknown) {
return axiosResponse(200, { success: true, message: '', data: body })
}
function javaGone() {
// 现网实读:ChunkedUploadService 抛 BusinessException 不带 code ⇒ GlobalExceptionHandler
// 按 HTTP 200 + success:false 出,只有文案可依据。
return axiosResponse(200, { success: false, message: '分片会话不存在或已过期', data: null })
}
/** 按 URL 分派的 POST 桩:init 发新会话号,片与 complete 一律成功。 */
function routePosts(initUploadId: string) {
mocks.post.mockImplementation((url: string) => {
const text = String(url)
if (text.includes('/files/chunk/init')) {
return Promise.resolve(javaOk({ upload_id: initUploadId, total_parts: 2 }))
}
if (text.includes('/complete')) {
return Promise.resolve(javaOk({
fileKey: 'tmp/resume.mp4', originalFilename: 'resume.mp4', localPath: '/tmp/resume.mp4',
size: TWO_PARTS, url: 'https://oss.example/resume.mp4', objectKey: 'image-video/resume.mp4',
}))
}
return Promise.resolve(javaOk({ received_parts: 1 }))
})
}
function postedUrls(): string[] {
return mocks.post.mock.calls.map(([url]) => String(url))
}
function initedCount(): number {
return postedUrls().filter((url) => url.includes('/files/chunk/init')).length
}
function partUrls(): string[] {
return postedUrls().filter((url) => /\/part\/\d+$/.test(url))
}
describe('P1-11 分片会话探测失败必须分因', () => {
beforeEach(() => {
vi.clearAllMocks()
localStorage.clear()
routePosts(FRESH_ID)
})
it('HTTP 404:服务端说会话没了 ⇒ 清状态并重新 init', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockResolvedValue(axiosResponse(404, { success: false, message: 'Not Found', data: null }))
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(initedCount(), '404 之后必须重新开会话').toBe(1)
expect(partUrls().every((url) => url.includes(FRESH_ID)), '新片必须挂在新的会话号上').toBe(true)
expect(partUrls().some((url) => url.includes(SESSION_ID)), '旧会话号已判死,不得再往它写片').toBe(false)
})
it('HTTP 200+success:false+「不存在或已过期」(现网形状)⇒ 同样重开', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockResolvedValue(javaGone())
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(initedCount(), '会话确实回收后不重建就永远卡死').toBe(1)
expect(partUrls().every((url) => url.includes(FRESH_ID))).toBe(true)
})
it('HTTP 503:瞬态 ⇒ 保留 upload_id,一条 init 都不许发', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockResolvedValue(axiosResponse(503, '<html>Bad Gateway</html>'))
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(initedCount(), '一次网关抖动就把整包打回从 0,正是本迭代要消灭的行为').toBe(0)
expect(partUrls().every((url) => url.includes(SESSION_ID)), '片仍要写进原来那个会话').toBe(true)
})
it('探测超时(网络层错误):瞬态 ⇒ 保留 upload_id 并续传', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockRejectedValue(new Error('timeout of 30000ms exceeded'))
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(initedCount()).toBe(0)
expect(partUrls().every((url) => url.includes(SESSION_ID))).toBe(true)
})
it('非会话类业务失败(如权限):不许顺手清掉会话', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockResolvedValue(axiosResponse(200, { success: false, message: '无权访问该分片会话', data: null }))
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(initedCount()).toBe(0)
expect(partUrls().every((url) => url.includes(SESSION_ID))).toBe(true)
})
it('真实 axios 形状:已到片号必须被跳过(P2-9 形状误判即静默从 0)', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockResolvedValue(javaOk({ upload_id: SESSION_ID, uploaded_parts: [0] }))
await uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' })
expect(partUrls(), '第 0 片服务端已有,不得重传').toEqual([
`/newApi/api/files/chunk/${SESSION_ID}/part/1`,
])
expect(mocks.get.mock.calls.length, '探测要带 upload_id 问服务端').toBe(1)
})
it('探测失败+分片也失败 ⇒ 会话留在 localStorage,且不 DELETE', async () => {
const file = makeBigFile()
seedSession(file, SESSION_ID)
mocks.get.mockRejectedValue(new Error('Network Error'))
mocks.post.mockRejectedValue(new Error('Network Error'))
await expect(uploadTempFileToJava(file, { uploadToOss: true, moduleType: 'IMAGE_VIDEO' }))
.rejects.toThrow(/Network Error/)
expect(storedUploadId(file), '瞬态失败后会话必须还在,下次刷新才能从断点续传').toBe(SESSION_ID)
expect(initedCount()).toBe(0)
})
})
@@ -42,3 +42,23 @@ test('图片视频大档上传必须接分片通道', () => {
'AC-11:uploadImageVideoMedia 仍是一次 multipart+固定 120s(大档一断就从 0 重传)',
)
})
// P3-5:8MB 阈值原先三处各写一遍(客户端 main.py:1479、本仓 upload.ts、后管 chunk-upload.ts),
// 改一处就分叉——本仓这一侧收成一个具名常量,别的文件只许引用、不许再写数。
test('分片阈值在本仓只能有一个出口', () => {
const pattern = /8\s*\*\s*1024\s*\*\s*1024|8388608/
const sites = srcFiles
.map(rel)
.filter((file) => pattern.test(readFileSync(resolve(root, file), 'utf-8')))
assert.deepEqual(
sites,
['src/shared/api/upload.ts'],
`8MB 阈值散写到了多处:${sites.join(', ')}(阈值只有一个出口=upload.ts::UPLOAD_PART_SIZE)`,
)
const source = readFileSync(resolve(root, 'src/shared/api/upload.ts'), 'utf-8')
assert.match(
source,
/export const UPLOAD_PART_SIZE\s*=\s*8\s*\*\s*1024\s*\*\s*1024[\s\S]{0,120}export const CHUNK_UPLOAD_THRESHOLD\s*=\s*UPLOAD_PART_SIZE/,
'AC-11:阈值必须由片大小单点派生,不得各写一份 8MB',
)
})
@@ -66,6 +66,13 @@ test('pending_ac17_上传必须分片且不再单次 120s', () => {
assert.doesNotMatch(source, /timeout:\s*120000/,
'S1-F3:单次 120s 超时的整包 multipart 必须被分片通道取代')
assert.match(source, /onProgress|progress/, '进度必须可见')
// P3-5:桌面桥的浏览器降级直传也是大档出口,固定 120s 会在同一把尺子上漏半条链。
const bridge = read('src/shared/bridges/web-fallback.ts')
assert.doesNotMatch(bridge, /timeout:\s*120_?000/,
'P3-5:web-fallback 的大档上传仍写死 120s,慢链路传 Excel/zip 会被自己掐死')
assert.match(bridge, /uploadTimeoutForBytes/,
'P3-5:超时预算只有一个出口(upload.ts::uploadTimeoutForBytes),桥侧必须复用它')
})
test('pending_ac18_表格必须窗口化渲染或走服务端分页', () => {