mirror of
https://github.com/whyour/qinglong.git
synced 2026-07-17 09:34:31 +08:00
修复添加任务有可能失败
This commit is contained in:
+19
-19
@@ -60,11 +60,11 @@ class Application {
|
||||
// Wait for gRPC worker to signal it's ready before starting HTTP worker
|
||||
this.waitForWorkerReady(grpcWorker, 30000)
|
||||
.then(() => {
|
||||
Logger.info('✌️ gRPC worker is ready, starting HTTP worker');
|
||||
Logger.info('[boot] gRPC worker is ready, starting HTTP worker');
|
||||
this.httpWorker = this.forkWorker('http');
|
||||
})
|
||||
.catch((error) => {
|
||||
Logger.error('✌️ Failed to wait for gRPC worker:', error);
|
||||
Logger.error('[boot] Failed to wait for gRPC worker:', error);
|
||||
process.exit(1);
|
||||
});
|
||||
|
||||
@@ -73,7 +73,7 @@ class Application {
|
||||
if (metadata) {
|
||||
if (!this.isShuttingDown) {
|
||||
Logger.error(
|
||||
`✌️ ${metadata.serviceType} worker ${worker.process.pid} died (${signal || code
|
||||
`${metadata.serviceType} worker ${worker.process.pid} died (${signal || code
|
||||
}). Restarting...`,
|
||||
);
|
||||
// If gRPC worker died, restart it and wait for it to be ready
|
||||
@@ -81,26 +81,26 @@ class Application {
|
||||
const newGrpcWorker = this.forkWorker('grpc');
|
||||
this.waitForWorkerReady(newGrpcWorker, 30000)
|
||||
.then(() => {
|
||||
Logger.info('✌️ gRPC worker restarted and ready');
|
||||
Logger.info('gRPC worker restarted and ready');
|
||||
// Re-register cron jobs by notifying the HTTP worker
|
||||
if (this.httpWorker) {
|
||||
try {
|
||||
this.httpWorker.send('reregister-crons');
|
||||
Logger.info('✌️ Sent reregister-crons message to HTTP worker');
|
||||
Logger.info('Sent reregister-crons message to HTTP worker');
|
||||
} catch (error) {
|
||||
Logger.error('✌️ Failed to send reregister-crons message:', error);
|
||||
Logger.error('Failed to send reregister-crons message:', error);
|
||||
}
|
||||
}
|
||||
})
|
||||
.catch((error) => {
|
||||
Logger.error('✌️ Failed to restart gRPC worker:', error);
|
||||
Logger.error('Failed to restart gRPC worker:', error);
|
||||
process.exit(1);
|
||||
});
|
||||
} else {
|
||||
// For HTTP worker, just restart it
|
||||
const newWorker = this.forkWorker(metadata.serviceType);
|
||||
this.httpWorker = newWorker;
|
||||
Logger.info(`✌️ Restarted ${metadata.serviceType} worker (PID: ${newWorker.process.pid})`);
|
||||
Logger.info(`Restarted ${metadata.serviceType} worker (PID: ${newWorker.process.pid})`);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -169,7 +169,7 @@ class Application {
|
||||
if (worker) {
|
||||
const exitPromise = new Promise<void>((resolve) => {
|
||||
worker.once('exit', () => {
|
||||
Logger.info(`✌️ Worker ${worker.process.pid} exited`);
|
||||
Logger.info(`Worker ${worker.process.pid} exited`);
|
||||
resolve();
|
||||
});
|
||||
|
||||
@@ -177,7 +177,7 @@ class Application {
|
||||
worker.send('shutdown');
|
||||
} catch (error) {
|
||||
Logger.warn(
|
||||
`✌️ Failed to send shutdown to worker ${worker.process.pid}:`,
|
||||
`Failed to send shutdown to worker ${worker.process.pid}:`,
|
||||
error,
|
||||
);
|
||||
}
|
||||
@@ -192,14 +192,14 @@ class Application {
|
||||
Promise.all(workerPromises),
|
||||
new Promise<void>((resolve) => {
|
||||
setTimeout(() => {
|
||||
Logger.warn('✌️ Worker shutdown timeout reached');
|
||||
Logger.warn('Worker shutdown timeout reached');
|
||||
resolve();
|
||||
}, 10000);
|
||||
}),
|
||||
]);
|
||||
process.exit(0);
|
||||
} catch (error) {
|
||||
Logger.error('✌️ Error during worker shutdown:', error);
|
||||
Logger.error('Error during worker shutdown:', error);
|
||||
process.exit(1);
|
||||
}
|
||||
};
|
||||
@@ -211,11 +211,11 @@ class Application {
|
||||
private async startWorkerProcess() {
|
||||
const serviceType = process.env.SERVICE_TYPE;
|
||||
if (!serviceType || !['http', 'grpc'].includes(serviceType)) {
|
||||
Logger.error('✌️ Invalid SERVICE_TYPE:', serviceType);
|
||||
Logger.error('[boot] Invalid SERVICE_TYPE:', serviceType);
|
||||
process.exit(1);
|
||||
}
|
||||
|
||||
Logger.info(`✌️ ${serviceType} worker started (PID: ${process.pid})`);
|
||||
Logger.info(`[boot] ${serviceType} worker started (PID: ${process.pid})`);
|
||||
|
||||
try {
|
||||
if (serviceType === 'http') {
|
||||
@@ -226,7 +226,7 @@ class Application {
|
||||
|
||||
process.send?.('ready');
|
||||
} catch (error) {
|
||||
Logger.error(`✌️ ${serviceType} worker failed:`, error);
|
||||
Logger.error(`[boot] ${serviceType} worker failed:`, error);
|
||||
process.exit(1);
|
||||
}
|
||||
}
|
||||
@@ -269,13 +269,13 @@ class Application {
|
||||
} else if (msg === 'reregister-crons' && serviceType === 'http') {
|
||||
// Re-register cron jobs when gRPC worker restarts
|
||||
try {
|
||||
Logger.info('✌️ Received reregister-crons message, re-registering cron jobs...');
|
||||
Logger.info('[boot] Received reregister-crons message, re-registering cron jobs...');
|
||||
const CronService = (await import('./services/cron')).default;
|
||||
const cronService = Container.get(CronService);
|
||||
await cronService.autosave_crontab();
|
||||
Logger.info('✌️ Cron jobs re-registered successfully');
|
||||
Logger.info('[boot] Cron jobs re-registered successfully');
|
||||
} catch (error) {
|
||||
Logger.error('✌️ Failed to re-register cron jobs:', error);
|
||||
Logger.error('[boot] Failed to re-register cron jobs:', error);
|
||||
}
|
||||
}
|
||||
});
|
||||
@@ -297,7 +297,7 @@ class Application {
|
||||
}
|
||||
process.exit(0);
|
||||
} catch (error) {
|
||||
Logger.error(`✌️ [${serviceType}] Error during shutdown:`, error);
|
||||
Logger.error(`[${serviceType}] Error during shutdown:`, error);
|
||||
process.exit(1);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -36,7 +36,7 @@ async function tmpFile(prefix: string): Promise<string> {
|
||||
}
|
||||
|
||||
async function generateAllCerts(): Promise<GrpcTlsConfig> {
|
||||
Logger.info('Generating gRPC mTLS certificates...');
|
||||
Logger.info('[boot] Generating gRPC mTLS certificates...');
|
||||
|
||||
const caKeyTmp = await tmpFile('ca_key');
|
||||
const caCertTmp = await tmpFile('ca_cert');
|
||||
@@ -83,7 +83,7 @@ async function generateAllCerts(): Promise<GrpcTlsConfig> {
|
||||
const clientKey = await fs.readFile(clientKeyTmp, 'utf-8');
|
||||
|
||||
await cleanup();
|
||||
Logger.info('gRPC mTLS certificates generated successfully');
|
||||
Logger.info('[boot] gRPC mTLS certificates generated successfully');
|
||||
|
||||
return { caCert, serverCert, serverKey, clientCert, clientKey };
|
||||
} catch (e) {
|
||||
@@ -101,7 +101,7 @@ async function saveCerts(tlsConfig: GrpcTlsConfig): Promise<void> {
|
||||
await fs.writeFile(clientCertPath, tlsConfig.clientCert, { mode: 0o644 });
|
||||
await fs.writeFile(clientKeyPath, tlsConfig.clientKey, { mode: 0o600 });
|
||||
|
||||
Logger.info(`gRPC mTLS certificates saved to ${certDir}`);
|
||||
Logger.info(`[boot] gRPC mTLS certificates saved to ${certDir}`);
|
||||
}
|
||||
|
||||
async function loadExistingCerts(): Promise<GrpcTlsConfig | null> {
|
||||
@@ -125,7 +125,7 @@ async function loadExistingCerts(): Promise<GrpcTlsConfig | null> {
|
||||
fs.readFile(clientKeyPath, 'utf-8'),
|
||||
]);
|
||||
|
||||
Logger.info('Loaded existing gRPC mTLS certificates from disk');
|
||||
Logger.info('[boot] Loaded existing gRPC mTLS certificates from disk');
|
||||
return { caCert, serverCert, serverKey, clientCert, clientKey };
|
||||
}
|
||||
|
||||
|
||||
+6
-6
@@ -9,20 +9,20 @@ import initFile from './initFile';
|
||||
|
||||
export default async ({ app }: { app: Application }) => {
|
||||
depInjectorLoader();
|
||||
Logger.info('✌️ Dependency loaded');
|
||||
Logger.info('[boot] Dependency loaded');
|
||||
|
||||
await linkDeps();
|
||||
Logger.info('✌️ Link deps loaded');
|
||||
Logger.info('[boot] Link deps loaded');
|
||||
|
||||
await initFile();
|
||||
Logger.info('✌️ Init file loaded');
|
||||
Logger.info('[boot] Init file loaded');
|
||||
|
||||
await initData();
|
||||
Logger.info('✌️ Init data loaded');
|
||||
Logger.info('[boot] Init data loaded');
|
||||
|
||||
initTask();
|
||||
Logger.info('✌️ Init task loaded');
|
||||
Logger.info('[boot] Init task loaded');
|
||||
|
||||
expressLoader({ app });
|
||||
Logger.info('✌️ Express loaded');
|
||||
Logger.info('[boot] Express loaded');
|
||||
};
|
||||
|
||||
+2
-2
@@ -58,8 +58,8 @@ export default async () => {
|
||||
}
|
||||
}
|
||||
|
||||
Logger.info('✌️ DB loaded');
|
||||
Logger.info('[boot] DB loaded');
|
||||
} catch (error) {
|
||||
Logger.error('✌️ DB load failed', error);
|
||||
Logger.error('[boot] DB load failed', error);
|
||||
}
|
||||
};
|
||||
|
||||
@@ -125,5 +125,5 @@ export default async () => {
|
||||
}
|
||||
}
|
||||
|
||||
Logger.info('✌️ Init file down');
|
||||
Logger.info('[boot] Init file down');
|
||||
};
|
||||
|
||||
@@ -4,7 +4,7 @@ import Sock from './sock';
|
||||
|
||||
export default async ({ server }: { server: Server }) => {
|
||||
await Sock({ server });
|
||||
Logger.info('✌️ Sock loaded');
|
||||
Logger.info('[boot] Sock loaded');
|
||||
|
||||
process.on('uncaughtException', (error) => {
|
||||
Logger.error('Uncaught exception:', error);
|
||||
|
||||
+99
-15
@@ -1,16 +1,79 @@
|
||||
import { ServerUnaryCall, sendUnaryData } from '@grpc/grpc-js';
|
||||
import { ServerUnaryCall, sendUnaryData, status } from '@grpc/grpc-js';
|
||||
import { AddCronRequest, AddCronResponse } from '../protos/cron';
|
||||
import nodeSchedule from 'node-schedule';
|
||||
import { scheduleStacks } from './data';
|
||||
import { runCron } from '../shared/runCron';
|
||||
import Logger from '../loaders/logger';
|
||||
import { tf } from '../shared/i18n';
|
||||
|
||||
/**
|
||||
* 预校验 cron 表达式,检测 node-schedule 会拒绝但 cron-parser 会接受的 pattern。
|
||||
* node-schedule 对 bare /N(字段以 / 开头,如前无星号/数字前缀的 /6)返回 null,
|
||||
* 提前拦截避免走 scheduleJob 后才发现无效。
|
||||
*/
|
||||
const isValidCronField = (cron: string): boolean => {
|
||||
// 检测 bare /N 模式:字段以 / 开头如 "/6",或空格后紧跟 "/6"
|
||||
// node-schedule 会对这种字段返回 null
|
||||
if (/\s\/\d/.test(cron) || /^\/\d/.test(cron)) {
|
||||
return false;
|
||||
}
|
||||
// 检测 ? 字符:Quartz cron 语法,node-schedule 在大多数位置返回 null
|
||||
// cron-parser 接受但 node-schedule 拒绝,提前拦截
|
||||
if (/\?/.test(cron)) {
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
};
|
||||
|
||||
const addCron = (
|
||||
call: ServerUnaryCall<AddCronRequest, AddCronResponse>,
|
||||
callback: sendUnaryData<AddCronResponse>,
|
||||
) => {
|
||||
// ===== 第一遍:预校验所有 cron 表达式 =====
|
||||
const validationErrors: string[] = [];
|
||||
|
||||
for (const item of call.request.crons) {
|
||||
const { id, schedule, extra_schedules } = item;
|
||||
|
||||
if (!isValidCronField(schedule)) {
|
||||
validationErrors.push(
|
||||
tf(
|
||||
'任务ID %s: 无效的 cron 表达式 "%s"(不支持裸 /N 步长和 ? 字符)',
|
||||
String(id),
|
||||
schedule,
|
||||
),
|
||||
);
|
||||
}
|
||||
|
||||
if (extra_schedules?.length) {
|
||||
extra_schedules.forEach((x) => {
|
||||
if (!isValidCronField(x.schedule)) {
|
||||
validationErrors.push(
|
||||
tf(
|
||||
'任务ID %s (extra_schedule): 无效的 cron 表达式 "%s"(不支持裸 /N 步长和 ? 字符)',
|
||||
String(id),
|
||||
x.schedule,
|
||||
),
|
||||
);
|
||||
}
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
if (validationErrors.length > 0) {
|
||||
const details = validationErrors.join('\n');
|
||||
const err: any = new Error(details);
|
||||
err.code = status.INVALID_ARGUMENT;
|
||||
err.details = details;
|
||||
callback(err, null);
|
||||
return;
|
||||
}
|
||||
|
||||
// ===== 第二遍:注册所有任务 =====
|
||||
for (const item of call.request.crons) {
|
||||
const { id, schedule, command, extra_schedules, name } = item;
|
||||
|
||||
// 取消该 id 已有的旧任务
|
||||
if (scheduleStacks.has(id)) {
|
||||
scheduleStacks.get(id)?.forEach((x) => x.cancel());
|
||||
}
|
||||
@@ -35,20 +98,41 @@ const addCron = (
|
||||
});
|
||||
}
|
||||
|
||||
scheduleStacks.set(id, [
|
||||
nodeSchedule.scheduleJob(id, schedule, async () => {
|
||||
Logger.info(`[schedule][准备运行任务] 命令: ${command}`);
|
||||
runCron(command, item);
|
||||
}),
|
||||
...(extra_schedules?.length
|
||||
? extra_schedules.map((x) =>
|
||||
nodeSchedule.scheduleJob(id, x.schedule, async () => {
|
||||
Logger.info(`[schedule][准备运行任务] 命令: ${command}`);
|
||||
runCron(command, item);
|
||||
}),
|
||||
)
|
||||
: []),
|
||||
]);
|
||||
const mainJob = nodeSchedule.scheduleJob(id, schedule, async () => {
|
||||
Logger.info(`[schedule][准备运行任务] 命令: ${command}`);
|
||||
runCron(command, item);
|
||||
});
|
||||
|
||||
if (!mainJob) {
|
||||
Logger.warn(
|
||||
'[schedule][创建定时任务] scheduleJob 返回 null(不符合预期,已通过预校验): 任务ID: %s, cron: %s',
|
||||
id,
|
||||
schedule,
|
||||
);
|
||||
}
|
||||
|
||||
const extraJobs = extra_schedules?.length
|
||||
? extra_schedules.map((x) => {
|
||||
const job = nodeSchedule.scheduleJob(id, x.schedule, async () => {
|
||||
Logger.info(`[schedule][准备运行任务] 命令: ${command}`);
|
||||
runCron(command, item);
|
||||
});
|
||||
if (!job) {
|
||||
Logger.warn(
|
||||
'[schedule][创建定时任务] scheduleJob 返回 null(不符合预期,已通过预校验): 任务ID: %s, cron: %s',
|
||||
id,
|
||||
x.schedule,
|
||||
);
|
||||
}
|
||||
return job;
|
||||
})
|
||||
: [];
|
||||
|
||||
// 过滤 null(兜底保护,正常情况下预校验已拦截)
|
||||
const jobs = [mainJob, ...extraJobs].filter((x) => x != null);
|
||||
if (jobs.length > 0) {
|
||||
scheduleStacks.set(id, jobs);
|
||||
}
|
||||
}
|
||||
|
||||
callback(null, null);
|
||||
|
||||
+40
-8
@@ -117,12 +117,16 @@ export default class CronService {
|
||||
},
|
||||
]);
|
||||
} catch (error: any) {
|
||||
// 调度器注册为尽力而为,失败时不阻断 crontab.list 与系统 crontab 的同步,
|
||||
// 调度器重启后会重新注册(见 autosave_crontab)
|
||||
this.logger.warn(
|
||||
'[crontab] Failed to register cron job in scheduler:',
|
||||
// gRPC 注册失败时回滚 DB 记录,避免产生"僵尸任务"
|
||||
// (DB 和 crontab.list 有记录但调度器永远不会执行)
|
||||
await CrontabModel.destroy({ where: { id: doc.id } });
|
||||
this.logger.error(
|
||||
'[crontab] Failed to register cron job in scheduler, task creation rolled back:',
|
||||
error?.message || error,
|
||||
);
|
||||
throw new Error(
|
||||
`${t('调度器注册失败,任务创建已回滚')}: ${(error as any)?.details || error?.message}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -166,10 +170,33 @@ export default class CronService {
|
||||
},
|
||||
]);
|
||||
} catch (error: any) {
|
||||
this.logger.warn(
|
||||
'[crontab] Failed to register cron job in scheduler:',
|
||||
// gRPC 注册新任务失败 → 回滚 DB 到旧数据,并尝试恢复旧调度注册
|
||||
await CrontabModel.update(doc, { where: { id: doc.id } });
|
||||
if (this.shouldUseCronClient(doc)) {
|
||||
try {
|
||||
await cronClient.addCron([
|
||||
{
|
||||
name: doc.name || '',
|
||||
id: String(doc.id),
|
||||
schedule: doc.schedule!,
|
||||
command: this.makeCommand(doc),
|
||||
extra_schedules: doc.extra_schedules || [],
|
||||
},
|
||||
]);
|
||||
} catch (_recoveryError: any) {
|
||||
this.logger.warn(
|
||||
'[crontab] Failed to restore old cron job in scheduler after rollback:',
|
||||
_recoveryError?.message || _recoveryError,
|
||||
);
|
||||
}
|
||||
}
|
||||
this.logger.error(
|
||||
'[crontab] Failed to register updated cron job in scheduler, update rolled back:',
|
||||
error?.message || error,
|
||||
);
|
||||
throw new Error(
|
||||
`${t('调度器注册失败,任务更新已回滚')}: ${(error as any)?.details || error?.message}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -726,10 +753,15 @@ export default class CronService {
|
||||
try {
|
||||
await cronClient.addCron(crons);
|
||||
} catch (error: any) {
|
||||
this.logger.warn(
|
||||
'[crontab] Failed to register cron job in scheduler:',
|
||||
// gRPC 注册失败 → 回滚启用状态,避免 DB 显示已启用但调度器未注册
|
||||
await CrontabModel.update({ isDisabled: 1 }, { where: { id: ids } });
|
||||
this.logger.error(
|
||||
'[crontab] Failed to register cron job in scheduler, enable rolled back:',
|
||||
error?.message || error,
|
||||
);
|
||||
throw new Error(
|
||||
`${t('调度器注册失败,任务启用已回滚')}: ${(error as any)?.details || error?.message}`,
|
||||
);
|
||||
}
|
||||
await this.setCrontab();
|
||||
}
|
||||
|
||||
@@ -50,7 +50,7 @@ export class GrpcServerService {
|
||||
try {
|
||||
const address = this.formatGrpcAddress(host, grpcPort);
|
||||
await bindAsync(address, credentials);
|
||||
Logger.debug(`✌️ gRPC service started successfully on ${address}`);
|
||||
Logger.debug(`[boot] gRPC service started successfully on ${address}`);
|
||||
metricsService.record('grpc_service_start', 1, {
|
||||
port: grpcPort.toString(),
|
||||
host
|
||||
|
||||
@@ -20,7 +20,7 @@ export class HttpServerService {
|
||||
for (const host of hostsToTry) {
|
||||
try {
|
||||
const server = await this.tryListen(expressApp, port, host);
|
||||
Logger.debug(`✌️ HTTP service started successfully on ${host}:${port}`);
|
||||
Logger.debug(`[boot] HTTP service started successfully on ${host}:${port}`);
|
||||
metricsService.record('http_service_start', 1, {
|
||||
port: port.toString(),
|
||||
host
|
||||
|
||||
@@ -131,6 +131,16 @@ const messages: Record<string, Record<string, string>> = {
|
||||
'执行before命令...': 'Execute before command...',
|
||||
'执行after命令...': 'Execute after command...',
|
||||
'## 执行结束... %s 耗时 %s 秒': '## Execution finished... %s elapsed %s seconds',
|
||||
'调度器注册失败,任务创建已回滚':
|
||||
'Scheduler registration failed, task creation rolled back',
|
||||
'调度器注册失败,任务更新已回滚':
|
||||
'Scheduler registration failed, task update rolled back',
|
||||
'调度器注册失败,任务启用已回滚':
|
||||
'Scheduler registration failed, task enable rolled back',
|
||||
'任务ID %s: 无效的 cron 表达式 "%s"(不支持裸 /N 步长和 ? 字符)':
|
||||
'Task ID %s: invalid cron expression "%s" (bare /N steps and ? character not supported)',
|
||||
'任务ID %s (extra_schedule): 无效的 cron 表达式 "%s"(不支持裸 /N 步长和 ? 字符)':
|
||||
'Task ID %s (extra_schedule): invalid cron expression "%s" (bare /N steps and ? character not supported)',
|
||||
},
|
||||
};
|
||||
|
||||
|
||||
@@ -12,6 +12,16 @@ const validateSchedule = (value: string, helpers: any) => {
|
||||
return value;
|
||||
}
|
||||
|
||||
// 检测裸 /N 模式:cron-parser 会接受,但 node-schedule 会返回 null
|
||||
// 提前拦截,避免任务入库后调度器注册失败
|
||||
if (/\s\/\d/.test(value) || /^\/\d/.test(value)) {
|
||||
return helpers.error('any.invalid');
|
||||
}
|
||||
// 检测 ? 字符:Quartz cron 语法,node-schedule 在大多数字段上返回 null
|
||||
if (/\?/.test(value)) {
|
||||
return helpers.error('any.invalid');
|
||||
}
|
||||
|
||||
try {
|
||||
if (CronExpressionParser.parse(value).hasNext()) {
|
||||
return value;
|
||||
|
||||
Reference in New Issue
Block a user