Files
aiartstudio/server/utils/imageArchiveQueue.ts
T
2026-04-26 13:33:42 +08:00

293 lines
10 KiB
TypeScript

// server/utils/imageArchiveQueue.ts - 图床归档队列 worker:从数据库领取任务、限流上传并按退避策略重试。
import { randomUUID } from "node:crypto";
import { clearTimeout, setTimeout } from "node:timers";
import { consola } from "consola";
import {
claimRunnableImageArchiveTasks,
finishImageGenerationArchiveFailed,
finishImageGenerationArchiveSuccess,
getUserArchiveIdentity,
scheduleImageGenerationArchiveRetry,
type IImageArchiveTask
} from "~~/server/utils/imageGenerationRecords";
import { uploadImageFromUrl } from "~~/server/utils/lsky";
interface IImageArchiveWorkerConfig {
/** 同一 Node 进程内允许同时执行的归档任务数,控制下载/上传带来的内存和网络峰值 */
concurrency: number;
/** 单条归档任务最多尝试次数,超过后记录为最终失败 */
maxAttempts: number;
/** RUNNING 任务超过该时间未完成时视为锁失效,可被其他 worker 重新领取 */
lockTtlMs: number;
/** 没有可执行任务时的轮询间隔,避免空转打数据库 */
pollMs: number;
}
interface IImageArchiveWorkerState {
/** 当前进程正在执行的归档任务数量 */
activeCount: number;
/** 防止同一进程内多个扫描循环重叠领取任务 */
draining: boolean;
/** 标记插件是否已经启动过 worker,避免开发热更新重复初始化 */
started: boolean;
/** 下一次扫描的定时器句柄,用于新任务入队时提前唤醒 */
timer: ReturnType<typeof setTimeout> | null;
/** 写入数据库锁的 worker 标识,用于日志排查和多实例抢占 */
workerId: string;
}
/** 默认只开 2 个归档槽位,避免多用户同时归档时挤占生图主链路资源 */
const DEFAULT_CONCURRENCY = 2;
/** 默认最多重试 3 次,覆盖短暂网络抖动,同时避免坏任务长期循环 */
const DEFAULT_MAX_ATTEMPTS = 3;
/** 锁超时 10 分钟,进程崩溃或重启后任务可以自动恢复 */
const DEFAULT_LOCK_TTL_MS = 10 * 60 * 1000;
/** 空队列每 15 秒扫一次;生图成功会主动唤醒,不依赖纯轮询 */
const DEFAULT_POLL_MS = 15 * 1000;
/** 失败退避节奏:第一次 30 秒,第二次 2 分钟,第三次及以后 10 分钟 */
const RETRY_DELAYS_MS = [30 * 1000, 2 * 60 * 1000, 10 * 60 * 1000];
// Nuxt 开发热更新可能重复加载模块;把状态挂到 globalThis,避免重复 worker 抢同一批任务。
const globalForArchiveWorker = globalThis as unknown as {
imageArchiveWorkerState?: IImageArchiveWorkerState;
};
const workerState =
globalForArchiveWorker.imageArchiveWorkerState ??
(globalForArchiveWorker.imageArchiveWorkerState = {
activeCount: 0,
draining: false,
started: false,
timer: null,
workerId: `archive-${randomUUID()}`
});
/** 启动应用内图床归档 worker;重复调用只会唤醒同一个 worker */
export const startImageArchiveWorker = () => {
if (!workerState.started) {
workerState.started = true;
consola.info("[imageArchiveQueue] worker 启动", {
workerId: workerState.workerId,
concurrency: getImageArchiveWorkerConfig().concurrency
});
}
scheduleDrain(0);
};
/** 生图成功入队后调用,用于尽快扫描新任务 */
export const wakeImageArchiveWorker = () => {
if (!workerState.started) return;
scheduleDrain(0);
};
/** 扫描数据库中的可执行归档任务,按当前空闲槽位领取并启动异步执行 */
const drainImageArchiveQueue = async () => {
// 归档扫描只负责领取任务和启动执行器;真正上传在 runImageArchiveTask 中异步完成。
if (workerState.draining) return;
workerState.draining = true;
let nextDelayMs = DEFAULT_POLL_MS;
try {
const config = getImageArchiveWorkerConfig();
const availableSlots = config.concurrency - workerState.activeCount;
// 槽位满时不再访问数据库,等任务完成后由 finally 主动唤醒下一轮扫描。
if (availableSlots <= 0) {
nextDelayMs = config.pollMs;
return;
}
// 数据库层会用条件 update 抢锁,保证多实例部署时同一条记录只会被一个 worker 领取。
const tasks = await claimRunnableImageArchiveTasks({
limit: availableSlots,
workerId: workerState.workerId,
lockTtlMs: config.lockTtlMs
});
if (tasks.length === 0) {
nextDelayMs = config.pollMs;
return;
}
for (const task of tasks) {
runImageArchiveTask(task, config);
}
// 如果本轮没填满并发槽,立刻再扫一次,尽快把可执行任务填到上限。
nextDelayMs =
workerState.activeCount < config.concurrency ? 0 : config.pollMs;
} catch (error) {
consola.error("[imageArchiveQueue] 扫描归档任务失败", {
workerId: workerState.workerId,
error: toArchiveLogError(error)
});
} finally {
workerState.draining = false;
if (workerState.started) {
scheduleDrain(nextDelayMs);
}
}
};
/** 执行单条归档任务:下载上游图片、上传图床并回写归档结果 */
const runImageArchiveTask = (
task: IImageArchiveTask,
config: IImageArchiveWorkerConfig
) => {
workerState.activeCount += 1;
// 不 await 这里的 IIFE:扫描循环要尽快释放,归档任务本身受 activeCount 限流。
void (async () => {
const startedAt = Date.now();
consola.info("[imageArchiveQueue] 归档开始", {
workerId: workerState.workerId,
recordId: task.id.toString(),
attempt: task.archiveAttempts
});
try {
// 只读取用户展示名用于文件命名,不把用户敏感信息写入日志或响应。
const identity = await getUserArchiveIdentity(task.userId);
const uploaded = await uploadImageFromUrl({
imageUrl: task.imageUrl,
userId: task.userId,
username: identity.username,
recordId: task.id,
createdAt: task.startedAt
});
// 成功后补写 hosted 字段;前端仍通过历史接口读取归档结果,不暴露图床完整响应。
await finishImageGenerationArchiveSuccess(task.id, {
hostedImageUrl: uploaded.publicUrl,
imageMimeType: uploaded.mimetype,
hostedResponse: uploaded.response
});
consola.info("[imageArchiveQueue] 归档成功", {
workerId: workerState.workerId,
recordId: task.id.toString(),
mimeType: uploaded.mimetype,
byteLength: uploaded.byteLength,
elapsedMs: Date.now() - startedAt
});
} catch (error) {
await handleImageArchiveFailure(task, config, error);
} finally {
// 无论成功失败都释放本进程槽位,并立刻唤醒队列处理后续待执行任务。
workerState.activeCount = Math.max(0, workerState.activeCount - 1);
wakeImageArchiveWorker();
}
})();
};
/** 处理单条归档任务失败:决定最终失败或按退避时间重新入队 */
const handleImageArchiveFailure = async (
task: IImageArchiveTask,
config: IImageArchiveWorkerConfig,
error: unknown
) => {
const safeError = toArchiveLogError(error);
const errorMessage = getArchiveErrorMessage(error);
const shouldStop = task.archiveAttempts >= config.maxAttempts;
// 归档日志只保留排查摘要,不记录 token、图片二进制、完整图床响应或完整上游响应。
consola.error("[imageArchiveQueue] 归档失败", {
workerId: workerState.workerId,
recordId: task.id.toString(),
attempt: task.archiveAttempts,
terminal: shouldStop,
error: safeError
});
if (shouldStop) {
// 最终失败只影响归档字段,生图状态保持 SUCCEEDED,前端看到的是本地化“图片归档失败”。
await finishImageGenerationArchiveFailed(task.id, errorMessage);
return;
}
// 可重试失败会清掉本次锁并设置下一次执行时间,期间历史记录仍表现为归档中。
await scheduleImageGenerationArchiveRetry({
recordId: task.id,
nextRunAt: new Date(Date.now() + getRetryDelayMs(task.archiveAttempts)),
errorMessage
});
};
/** 安排下一次队列扫描;新唤醒会覆盖旧定时器,确保扫描节奏可控 */
const scheduleDrain = (delayMs: number) => {
// 新任务入队时会用 0ms 重新调度,因此这里先清理旧 timer,避免重复扫描。
if (workerState.timer) {
clearTimeout(workerState.timer);
}
workerState.timer = setTimeout(() => {
workerState.timer = null;
void drainImageArchiveQueue();
}, delayMs);
// 不让空队列轮询 timer 阻止 Node 进程正常退出。
workerState.timer.unref?.();
};
/** 从环境变量读取 worker 配置,非法值统一回退默认值 */
const getImageArchiveWorkerConfig = (): IImageArchiveWorkerConfig => {
return {
concurrency: getPositiveIntegerEnv(
"IMAGE_ARCHIVE_CONCURRENCY",
DEFAULT_CONCURRENCY
),
maxAttempts: getPositiveIntegerEnv(
"IMAGE_ARCHIVE_MAX_ATTEMPTS",
DEFAULT_MAX_ATTEMPTS
),
lockTtlMs: DEFAULT_LOCK_TTL_MS,
pollMs: DEFAULT_POLL_MS
};
};
/** 根据当前尝试次数选择退避时间,超过表长时沿用最后一个退避值 */
const getRetryDelayMs = (attempt: number) => {
const lastDelayMs = RETRY_DELAYS_MS[RETRY_DELAYS_MS.length - 1] ?? 0;
return (
RETRY_DELAYS_MS[Math.min(attempt - 1, RETRY_DELAYS_MS.length - 1)] ??
lastDelayMs
);
};
/** 读取正整数环境变量,避免 0、负数、NaN 破坏限流和重试策略 */
const getPositiveIntegerEnv = (name: string, fallback: number) => {
const value = Number.parseInt(process.env[name] ?? "", 10);
return Number.isInteger(value) && value > 0 ? value : fallback;
};
/** 保存到数据库的失败摘要只取可读 message,不保存复杂错误对象 */
const getArchiveErrorMessage = (error: unknown) => {
if (error instanceof Error && error.message) return error.message;
if (typeof error === "string" && error) return error;
return "图片归档失败";
};
/** 归档日志错误摘要,不包含 stack、请求体、响应体或任何密钥字段 */
const toArchiveLogError = (error: unknown) => {
const maybeError = error as {
message?: string;
name?: string;
response?: {
status?: number;
};
status?: number;
statusCode?: number;
statusMessage?: string;
};
return {
name: maybeError.name,
message: maybeError.message ?? maybeError.statusMessage,
status:
maybeError.status ?? maybeError.statusCode ?? maybeError.response?.status
};
};