同步异常兜底
This commit is contained in:
1 parent
9cb8552a99
commit
a1b4a63552
20 files changed
+754
-54
No files matched your search
@@ -1,4 +1,5 @@
|
||||
import { basename } from 'node:path'
|
||||
import type { Readable } from 'node:stream'
|
||||
import { contentTypeFromKey } from '../content-type'
|
||||
import {
|
||||
getImageIndexRow,
|
||||
@@ -22,7 +23,19 @@ import {
|
||||
markJobFailed,
|
||||
updateBackupJob
|
||||
} from '../backup/jobs'
|
||||
import {
|
||||
createJobCancelledError,
|
||||
getJobAbortSignal,
|
||||
getJobOperationSignal,
|
||||
isJobCancellationRequested,
|
||||
isJobCancelledError,
|
||||
throwIfJobCancellationRequested
|
||||
} from '../backup/cancel-state'
|
||||
import { applySiteSyncMedia } from './apply'
|
||||
import {
|
||||
completeSiteSyncMediaArtifacts,
|
||||
recordSiteSyncArtifact
|
||||
} from './artifacts'
|
||||
import { toSiteSyncMediaTransferWithTags } from './manifest'
|
||||
import type {
|
||||
SiteSyncDirection,
|
||||
@@ -49,9 +62,19 @@ export function registerSiteSyncToken(jobId: string, token: string): void {
|
||||
tokens.set(jobId, token)
|
||||
}
|
||||
|
||||
async function streamToBuffer(stream: NodeJS.ReadableStream): Promise<Buffer> {
|
||||
const SITE_SYNC_JSON_TIMEOUT_MS = 2 * 60 * 1000
|
||||
const SITE_SYNC_OBJECT_TIMEOUT_MS = 15 * 60 * 1000
|
||||
|
||||
async function streamToBuffer(
|
||||
stream: Readable,
|
||||
signal?: AbortSignal
|
||||
): Promise<Buffer> {
|
||||
const chunks: Buffer[] = []
|
||||
for await (const chunk of stream) {
|
||||
if (signal?.aborted) {
|
||||
stream.destroy(createJobCancelledError())
|
||||
throw createJobCancelledError()
|
||||
}
|
||||
chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk))
|
||||
}
|
||||
return Buffer.concat(chunks)
|
||||
@@ -71,21 +94,46 @@ async function responseError(response: Response): Promise<Error> {
|
||||
|
||||
const SITE_SYNC_MAX_ATTEMPTS = 3
|
||||
|
||||
function normalizeSiteSyncError(error: Error): Error {
|
||||
if (error.name === 'TimeoutError') {
|
||||
return new Error('目标站点连接超时或无响应')
|
||||
}
|
||||
if (error instanceof TypeError && /fetch failed|terminated/i.test(error.message)) {
|
||||
return new Error('目标站点无法连接')
|
||||
}
|
||||
return error
|
||||
}
|
||||
|
||||
export async function retrySiteSyncOperation<T>(
|
||||
operation: () => Promise<T>
|
||||
operation: () => Promise<T>,
|
||||
signal?: AbortSignal
|
||||
): Promise<T> {
|
||||
if (signal?.aborted) throw createJobCancelledError()
|
||||
let lastError: Error | null = null
|
||||
for (let attempt = 1; attempt <= SITE_SYNC_MAX_ATTEMPTS; attempt++) {
|
||||
try {
|
||||
return await operation()
|
||||
} catch (error) {
|
||||
if (signal?.aborted) throw createJobCancelledError()
|
||||
lastError = error instanceof Error ? error : new Error('两站点同步失败')
|
||||
if (attempt < SITE_SYNC_MAX_ATTEMPTS) {
|
||||
await new Promise(resolve => setTimeout(resolve, attempt * 500))
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
const onAbort = () => {
|
||||
clearTimeout(timer)
|
||||
reject(createJobCancelledError())
|
||||
}
|
||||
const timer = setTimeout(() => {
|
||||
signal?.removeEventListener('abort', onAbort)
|
||||
resolve()
|
||||
}, attempt * 500)
|
||||
signal?.addEventListener('abort', onAbort, { once: true })
|
||||
})
|
||||
}
|
||||
}
|
||||
}
|
||||
throw lastError ?? new Error('两站点同步失败')
|
||||
throw lastError
|
||||
? normalizeSiteSyncError(lastError)
|
||||
: new Error('两站点同步失败')
|
||||
}
|
||||
|
||||
async function fetchSiteSyncJsonOnce<T>(
|
||||
@@ -107,14 +155,25 @@ async function fetchSiteSyncJsonOnce<T>(
|
||||
export async function fetchSiteSyncJson<T>(
|
||||
url: string,
|
||||
token: string,
|
||||
init?: RequestInit
|
||||
init?: RequestInit,
|
||||
signal?: AbortSignal
|
||||
): Promise<T> {
|
||||
return retrySiteSyncOperation(() => (
|
||||
fetchSiteSyncJsonOnce<T>(url, token, init)
|
||||
))
|
||||
return retrySiteSyncOperation(
|
||||
() => fetchSiteSyncJsonOnce<T>(url, token, {
|
||||
...init,
|
||||
signal: signal
|
||||
? AbortSignal.any([
|
||||
signal,
|
||||
AbortSignal.timeout(SITE_SYNC_JSON_TIMEOUT_MS)
|
||||
])
|
||||
: AbortSignal.timeout(SITE_SYNC_JSON_TIMEOUT_MS)
|
||||
}),
|
||||
signal
|
||||
)
|
||||
}
|
||||
|
||||
async function uploadObject(
|
||||
jobId: string,
|
||||
options: SiteSyncOptions,
|
||||
token: string,
|
||||
backend: StorageBackend,
|
||||
@@ -124,9 +183,15 @@ async function uploadObject(
|
||||
): Promise<void> {
|
||||
const stored = await backend.head(key)
|
||||
if (!stored) throw new Error(`源文件不存在:${key}`)
|
||||
const bytes = await streamToBuffer(await backend.createStream(key))
|
||||
const bytes = await streamToBuffer(
|
||||
await backend.createStream(key),
|
||||
getJobAbortSignal(jobId)
|
||||
)
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
const form = new FormData()
|
||||
const isPrimary = key === row.key
|
||||
form.append('jobId', jobId)
|
||||
form.append('mediaKey', row.key)
|
||||
form.append('key', key)
|
||||
form.append('conflictMode', options.conflictMode)
|
||||
form.append('ownerUsername', ownerUsername ?? '')
|
||||
@@ -143,11 +208,16 @@ async function uploadObject(
|
||||
await fetchSiteSyncJsonOnce<{ success: boolean }>(
|
||||
targetUrl(options.targetBaseUrl, '/api/site-sync/object'),
|
||||
token,
|
||||
{ method: 'POST', body: form }
|
||||
{
|
||||
method: 'POST',
|
||||
body: form,
|
||||
signal: getJobOperationSignal(jobId, SITE_SYNC_OBJECT_TIMEOUT_MS)
|
||||
}
|
||||
)
|
||||
}
|
||||
|
||||
async function downloadSiteSyncObject(
|
||||
jobId: string,
|
||||
options: SiteSyncOptions,
|
||||
token: string,
|
||||
key: string
|
||||
@@ -160,7 +230,8 @@ async function downloadSiteSyncObject(
|
||||
{
|
||||
headers: {
|
||||
'Auth-Token': token
|
||||
}
|
||||
},
|
||||
signal: getJobOperationSignal(jobId, SITE_SYNC_OBJECT_TIMEOUT_MS)
|
||||
}
|
||||
)
|
||||
if (!response.ok) throw await responseError(response)
|
||||
@@ -173,6 +244,7 @@ async function runSiteSyncPull(
|
||||
token: string
|
||||
): Promise<void> {
|
||||
const job = getBackupJob(jobId)
|
||||
const jobSignal = getJobAbortSignal(jobId)
|
||||
const checkpoint = job?.payload ?? {}
|
||||
let lastMediaKey = typeof checkpoint.lastMediaKey === 'string'
|
||||
? checkpoint.lastMediaKey
|
||||
@@ -189,7 +261,7 @@ async function runSiteSyncPull(
|
||||
: []
|
||||
)
|
||||
|
||||
const saveCheckpoint = (message: string) => {
|
||||
const saveCheckpoint = (message: string, cancelled = false) => {
|
||||
updateBackupJob(jobId, {
|
||||
progress: scanned,
|
||||
total: totalMedia,
|
||||
@@ -198,6 +270,7 @@ async function runSiteSyncPull(
|
||||
targetBaseUrl: options.targetBaseUrl,
|
||||
conflictMode: options.conflictMode,
|
||||
direction: 'pull',
|
||||
cancelled,
|
||||
lastMediaKey,
|
||||
scanned,
|
||||
added,
|
||||
@@ -215,11 +288,19 @@ async function runSiteSyncPull(
|
||||
`扫描 ${scanned}/${totalMedia || scanned} · 新增 ${added} · 跳过 ${skipped} · 失败 ${failedKeys.size}${suffix}`
|
||||
)
|
||||
const failAndCheckpoint = (error: unknown) => {
|
||||
const message = progressMessage(' · 已保存断点')
|
||||
saveCheckpoint(message)
|
||||
const cancelled = isJobCancellationRequested(jobId)
|
||||
|| isJobCancelledError(error)
|
||||
const message = cancelled
|
||||
? progressMessage(' · 已取消,已保存断点')
|
||||
: progressMessage(' · 已保存断点')
|
||||
saveCheckpoint(message, cancelled)
|
||||
markJobFailed(
|
||||
jobId,
|
||||
error instanceof Error ? error.message : '从目标站点拉取失败',
|
||||
cancelled
|
||||
? '任务已取消'
|
||||
: error instanceof Error
|
||||
? error.message
|
||||
: '从目标站点拉取失败',
|
||||
message
|
||||
)
|
||||
}
|
||||
@@ -228,21 +309,22 @@ async function runSiteSyncPull(
|
||||
const targetUsers = new Set(
|
||||
(await fetchSiteSyncJson<SiteSyncUsers>(
|
||||
targetUrl(options.targetBaseUrl, '/api/site-sync/users'),
|
||||
token
|
||||
token,
|
||||
undefined,
|
||||
jobSignal
|
||||
)).users
|
||||
)
|
||||
const localUserMap = listUserIdUsernameMap()
|
||||
const localUsers = new Set(localUserMap.values())
|
||||
const missingUsers = [...targetUsers].filter(name => !localUsers.has(name))
|
||||
if (missingUsers.length) {
|
||||
throw new Error(`当前站点缺少同名用户:${missingUsers.join('、')}`)
|
||||
}
|
||||
|
||||
const backend = await retrySiteSyncOperation(() => getActiveBackend())
|
||||
const backend = await retrySiteSyncOperation(
|
||||
() => getActiveBackend(),
|
||||
jobSignal
|
||||
)
|
||||
const actorUserId = job?.userId ?? getPrimaryAdminUserId() ?? 0
|
||||
const readMediaPage = (path: string) => fetchSiteSyncJson<SiteSyncMediaPage>(
|
||||
targetUrl(options.targetBaseUrl, path),
|
||||
token
|
||||
token,
|
||||
undefined,
|
||||
jobSignal
|
||||
)
|
||||
|
||||
const processBundleOnce = async (
|
||||
@@ -250,11 +332,16 @@ async function runSiteSyncPull(
|
||||
isRetry: boolean
|
||||
): Promise<'added' | 'skipped'> => {
|
||||
const { media, assets } = bundle
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
if (media.ownerUsername && !targetUsers.has(media.ownerUsername)) {
|
||||
throw new Error(`当前站点缺少同名用户:${media.ownerUsername}`)
|
||||
}
|
||||
if (
|
||||
!isRetry
|
||||
&& options.conflictMode === 'skip'
|
||||
&& getImageIndexRow(media.key)
|
||||
) {
|
||||
completeSiteSyncMediaArtifacts(jobId, media.key)
|
||||
return 'skipped'
|
||||
}
|
||||
|
||||
@@ -262,23 +349,31 @@ async function runSiteSyncPull(
|
||||
? findUserByUsername(media.ownerUsername)
|
||||
: null
|
||||
for (const asset of assets) {
|
||||
if (
|
||||
options.conflictMode === 'skip'
|
||||
&& await backend.head(asset.key)
|
||||
) {
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
const existedBefore = Boolean(await backend.head(asset.key))
|
||||
if (options.conflictMode === 'skip' && existedBefore) {
|
||||
continue
|
||||
}
|
||||
const bytes = await downloadSiteSyncObject(
|
||||
jobId,
|
||||
options,
|
||||
token,
|
||||
asset.key
|
||||
)
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
if (bytes.length !== asset.size) {
|
||||
throw new Error(
|
||||
`同步对象大小不一致:${asset.key},预期 ${asset.size},实际 ${bytes.length}`
|
||||
)
|
||||
}
|
||||
const isPrimary = asset.key === media.key
|
||||
recordSiteSyncArtifact({
|
||||
jobId,
|
||||
mediaKey: media.key,
|
||||
key: asset.key,
|
||||
backendId: backend.id,
|
||||
created: !existedBefore
|
||||
})
|
||||
await backend.put(asset.key, new Uint8Array(bytes), {
|
||||
originalName: isPrimary ? media.originalName : basename(asset.key),
|
||||
uploadedAt: media.uploadedAt,
|
||||
@@ -295,13 +390,18 @@ async function runSiteSyncPull(
|
||||
? 'file'
|
||||
: 'image'
|
||||
})
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
}
|
||||
|
||||
await applySiteSyncMedia(media, 'overwrite', actorUserId)
|
||||
completeSiteSyncMediaArtifacts(jobId, media.key)
|
||||
return 'added'
|
||||
}
|
||||
const processBundle = (bundle: SiteSyncMediaBundle, isRetry: boolean) => (
|
||||
retrySiteSyncOperation(() => processBundleOnce(bundle, isRetry))
|
||||
retrySiteSyncOperation(
|
||||
() => processBundleOnce(bundle, isRetry),
|
||||
jobSignal
|
||||
)
|
||||
)
|
||||
const recordOutcome = (outcome: 'added' | 'skipped', fileCount: number) => {
|
||||
if (outcome === 'added') added += 1
|
||||
@@ -382,6 +482,7 @@ async function runSiteSyncPull(
|
||||
return
|
||||
}
|
||||
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
totalFiles = filesSynced
|
||||
markJobDone(jobId, {
|
||||
message: `扫描 ${scanned}/${totalMedia || scanned} · 新增 ${added} · 跳过 ${skipped} · 失败 0`,
|
||||
@@ -412,6 +513,7 @@ export async function runSiteSyncJob(
|
||||
): Promise<void> {
|
||||
const token = tokens.get(jobId)
|
||||
if (!token) throw new Error('同步 Token 不可用,请重新发起任务')
|
||||
const jobSignal = getJobAbortSignal(jobId)
|
||||
|
||||
try {
|
||||
if (options.direction === 'pull') {
|
||||
@@ -421,7 +523,9 @@ export async function runSiteSyncJob(
|
||||
|
||||
const inventory = await fetchSiteSyncJson<SiteSyncInventory>(
|
||||
targetUrl(options.targetBaseUrl, '/api/site-sync/inventory'),
|
||||
token
|
||||
token,
|
||||
undefined,
|
||||
jobSignal
|
||||
)
|
||||
const targetUsers = new Set(inventory.users)
|
||||
const targetMedia = new Set(inventory.mediaKeys)
|
||||
@@ -466,6 +570,7 @@ export async function runSiteSyncJob(
|
||||
})
|
||||
|
||||
for (const row of rows) {
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
const backend = await getBackendForKey(row.key)
|
||||
let assetKeys: string[]
|
||||
try {
|
||||
@@ -487,7 +592,7 @@ export async function runSiteSyncJob(
|
||||
})
|
||||
}
|
||||
|
||||
const saveCheckpoint = (message: string) => {
|
||||
const saveCheckpoint = (message: string, cancelled = false) => {
|
||||
updateBackupJob(jobId, {
|
||||
progress: scanned,
|
||||
total: rows.length,
|
||||
@@ -496,6 +601,7 @@ export async function runSiteSyncJob(
|
||||
targetBaseUrl: options.targetBaseUrl,
|
||||
conflictMode: options.conflictMode,
|
||||
direction: 'push',
|
||||
cancelled,
|
||||
lastMediaKey,
|
||||
scanned,
|
||||
added,
|
||||
@@ -520,11 +626,22 @@ export async function runSiteSyncJob(
|
||||
let lastCheckpointAt = 0
|
||||
const processPlanOnce = async (plan: typeof plans[number], isRetry: boolean) => {
|
||||
const { row, ownerUsername, backend, keys } = plan
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
if (
|
||||
!isRetry
|
||||
&& options.conflictMode === 'skip'
|
||||
&& targetMedia.has(row.key)
|
||||
) {
|
||||
await fetchSiteSyncJsonOnce<{ success: boolean }>(
|
||||
targetUrl(options.targetBaseUrl, '/api/site-sync/complete'),
|
||||
token,
|
||||
{
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({ jobId, mediaKey: row.key }),
|
||||
signal: getJobOperationSignal(jobId, SITE_SYNC_JSON_TIMEOUT_MS)
|
||||
}
|
||||
)
|
||||
skipped += 1
|
||||
filesSynced = Math.min(totalFiles, filesSynced + keys.length)
|
||||
failedKeys.delete(row.key)
|
||||
@@ -532,6 +649,7 @@ export async function runSiteSyncJob(
|
||||
}
|
||||
|
||||
for (const key of keys) {
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
const stored = await backend.head(key)
|
||||
if (!stored) throw new Error(`源文件不存在:${key}`)
|
||||
if (
|
||||
@@ -540,9 +658,18 @@ export async function runSiteSyncJob(
|
||||
) {
|
||||
continue
|
||||
}
|
||||
await uploadObject(options, token, backend, row, ownerUsername, key)
|
||||
await uploadObject(
|
||||
jobId,
|
||||
options,
|
||||
token,
|
||||
backend,
|
||||
row,
|
||||
ownerUsername,
|
||||
key
|
||||
)
|
||||
}
|
||||
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
await fetchSiteSyncJsonOnce<{ success: boolean }>(
|
||||
targetUrl(options.targetBaseUrl, '/api/site-sync/media'),
|
||||
token,
|
||||
@@ -550,9 +677,11 @@ export async function runSiteSyncJob(
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
jobId,
|
||||
media: toSiteSyncMediaTransferWithTags(row, ownerUsername),
|
||||
conflictMode: options.conflictMode
|
||||
})
|
||||
}),
|
||||
signal: getJobOperationSignal(jobId, SITE_SYNC_JSON_TIMEOUT_MS)
|
||||
}
|
||||
)
|
||||
|
||||
@@ -563,18 +692,29 @@ export async function runSiteSyncJob(
|
||||
const processPlan = (
|
||||
plan: typeof plans[number],
|
||||
isRetry: boolean
|
||||
) => retrySiteSyncOperation(() => processPlanOnce(plan, isRetry))
|
||||
) => retrySiteSyncOperation(
|
||||
() => processPlanOnce(plan, isRetry),
|
||||
jobSignal
|
||||
)
|
||||
|
||||
const retryPlans = plans.filter(plan => failedKeys.has(plan.row.key))
|
||||
for (const plan of retryPlans) {
|
||||
try {
|
||||
await processPlan(plan, true)
|
||||
lastMediaKey = plan.row.key
|
||||
} catch {
|
||||
} catch (error) {
|
||||
failedKeys.add(plan.row.key)
|
||||
const message = `扫描 ${scanned}/${rows.length} · 新增 ${added} · 跳过 ${skipped} · 失败 ${failedKeys.size} · 已保存断点`
|
||||
saveCheckpoint(message)
|
||||
markJobFailed(jobId, '连续 3 次重试失败,已保存断点', message)
|
||||
const cancelled = isJobCancellationRequested(jobId)
|
||||
|| isJobCancelledError(error)
|
||||
const message = `扫描 ${scanned}/${rows.length} · 新增 ${added} · 跳过 ${skipped} · 失败 ${failedKeys.size} · ${cancelled ? '已取消,' : ''}已保存断点`
|
||||
saveCheckpoint(message, cancelled)
|
||||
markJobFailed(
|
||||
jobId,
|
||||
cancelled
|
||||
? '任务已取消'
|
||||
: '连续 3 次重试失败,已保存断点',
|
||||
message
|
||||
)
|
||||
return
|
||||
}
|
||||
}
|
||||
@@ -584,12 +724,20 @@ export async function runSiteSyncJob(
|
||||
scanned += 1
|
||||
try {
|
||||
await processPlan(plan, false)
|
||||
} catch {
|
||||
} catch (error) {
|
||||
failedKeys.add(plan.row.key)
|
||||
lastMediaKey = plan.row.key
|
||||
const message = `扫描 ${scanned}/${rows.length} · 新增 ${added} · 跳过 ${skipped} · 失败 ${failedKeys.size} · 已保存断点`
|
||||
saveCheckpoint(message)
|
||||
markJobFailed(jobId, '连续 3 次重试失败,已保存断点', message)
|
||||
const cancelled = isJobCancellationRequested(jobId)
|
||||
|| isJobCancelledError(error)
|
||||
const message = `扫描 ${scanned}/${rows.length} · 新增 ${added} · 跳过 ${skipped} · 失败 ${failedKeys.size} · ${cancelled ? '已取消,' : ''}已保存断点`
|
||||
saveCheckpoint(message, cancelled)
|
||||
markJobFailed(
|
||||
jobId,
|
||||
cancelled
|
||||
? '任务已取消'
|
||||
: '连续 3 次重试失败,已保存断点',
|
||||
message
|
||||
)
|
||||
return
|
||||
}
|
||||
lastMediaKey = plan.row.key
|
||||
@@ -614,6 +762,7 @@ export async function runSiteSyncJob(
|
||||
return
|
||||
}
|
||||
|
||||
throwIfJobCancellationRequested(jobId)
|
||||
markJobDone(jobId, {
|
||||
message: `扫描 ${rows.length}/${rows.length} · 新增 ${added} · 跳过 ${skipped} · 失败 0`,
|
||||
payload: {
|
||||
@@ -632,6 +781,10 @@ export async function runSiteSyncJob(
|
||||
}
|
||||
})
|
||||
} catch (error) {
|
||||
if (isJobCancellationRequested(jobId) || isJobCancelledError(error)) {
|
||||
markJobFailed(jobId, '任务已取消', '任务已取消')
|
||||
return
|
||||
}
|
||||
const message = error instanceof Error ? error.message : '两站点同步失败'
|
||||
markJobFailed(jobId, message)
|
||||
throw error
|
||||
|
||||
Reference in new issue
Block a user