refactor: migrate cron scheduling to node-cron (#3079)

* refactor: migrate cron scheduling to node-cron

* feat: support annually midnight and minutely cron macros

* fix: harden cron recovery and system scheduler compatibility
This commit is contained in:
whyour
2026-09-27 22:18:39 +08:00
committed by GitHub
parent bf63425805
commit f168efcdae
20 changed files with 1955 additions and 201 deletions
+48 -73
View File
@@ -1,6 +1,6 @@
import { ServerUnaryCall, sendUnaryData, status } from '@grpc/grpc-js';
import { AddCronRequest, AddCronResponse } from '../protos/cron';
import nodeSchedule from 'node-schedule';
import { createCronJob, CronJob } from '../shared/cronScheduler';
import { isValidCronSchedule } from '../shared/cronSchedule';
import { scheduleStacks } from './data';
import { runCron } from '../shared/runCron';
@@ -24,11 +24,7 @@ const addCron = (
if (!isValidCronField(schedule)) {
validationErrors.push(
tf(
'任务ID %s: 无效的 cron 表达式 "%s"',
String(id),
schedule,
),
tf('任务ID %s: 无效的 cron 表达式 "%s"', String(id), schedule),
);
}
@@ -56,79 +52,58 @@ const addCron = (
return;
}
// Recovery replaces the whole snapshot, including deletions and disabled jobs.
// Validation above must finish before touching the previous schedule.
if (call.request.replace) {
for (const jobs of scheduleStacks.values()) {
for (const job of jobs) job?.cancel();
// Prepare stopped jobs first: construction errors must preserve the old snapshot.
const prepared = new Map<string, CronJob[]>();
try {
for (const item of call.request.crons) {
const jobs: CronJob[] = [];
prepared.get(item.id)?.forEach((job) => job.cancel());
prepared.set(item.id, jobs);
for (const schedule of [
item.schedule,
...(item.extra_schedules || []).map((x) => x.schedule),
]) {
jobs.push(
createCronJob(
schedule,
async () => {
Logger.info('[schedule][准备运行任务] 命令: %s', item.command);
await runCron(item.command, item);
},
{
name: `${item.id}: ${item.name || ''}`,
logger: Logger,
start: false,
},
),
);
}
}
scheduleStacks.clear();
} catch (error) {
for (const jobs of prepared.values()) jobs.forEach((job) => job.cancel());
const err: any = new Error(
error instanceof Error ? error.message : String(error),
);
err.code = status.INVALID_ARGUMENT;
err.details = err.message;
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());
}
if (call.request.replace) {
for (const jobs of scheduleStacks.values())
jobs.forEach((job) => job?.cancel());
scheduleStacks.clear();
}
for (const [id, jobs] of prepared) {
scheduleStacks.get(id)?.forEach((job) => job.cancel());
scheduleStacks.set(id, jobs);
jobs.forEach((job) => job.start());
Logger.info(
'[schedule][创建定时任务] 任务ID: %s, 名称: %s, cron: %s, 执行命令: %s',
'[schedule][创建定时任务] 任务ID: %s, 规则数: %s',
id,
name,
schedule,
command,
jobs.length,
);
if (extra_schedules?.length) {
extra_schedules.forEach((x) => {
Logger.info(
'[schedule][创建定时任务] 任务ID: %s, 名称: %s, cron: %s, 执行命令: %s',
id,
name,
x.schedule,
command,
);
});
}
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);
+7 -1
View File
@@ -67,11 +67,17 @@ class Client {
replace = false
): Promise<AddCronResponse> {
await this.waitForReady(2000);
// Include every rule in a recovery snapshot, but bound a stuck worker.
const ruleCount = request.reduce(
(count, item) => count + 1 + (item.extra_schedules?.length || 0),
0,
);
const registrationTimeoutMs = Math.min(120000, 5000 + ruleCount * 5);
return new Promise((resolve, reject) => {
this.client.addCron(
{ crons: request, replace },
new Metadata(),
{ deadline: Date.now() + 5000 },
{ deadline: Date.now() + registrationTimeoutMs },
(err, res) => {
if (err) {
if (err.code === status.UNAVAILABLE || err.code === status.DEADLINE_EXCEEDED) {
+2 -2
View File
@@ -1,6 +1,6 @@
import nodeSchedule from 'node-schedule';
import { CronJob } from '../shared/cronScheduler';
import { ToadScheduler } from 'toad-scheduler';
export const scheduleStacks = new Map<string, nodeSchedule.Job[]>();
export const scheduleStacks = new Map<string, CronJob[]>();
export const intervalSchedule = new ToadScheduler();
+1 -1
View File
@@ -13,7 +13,7 @@ const delCron = (
'[schedule][取消定时任务] 任务ID: %s',
id,
);
// 过滤掉 nodeSchedule.scheduleJob() 对无效表达式返回的 null,
// 防御性过滤历史调度栈中的空任务,
// 否则对 null 调 cancel() 会让整个取消流程抛出 UNKNOWN 错误,
// 进而导致 HTTP 端的 remove() 跳过 setCrontab(),造成 crontab.list 残留。
scheduleStacks.get(id)?.filter((x) => x != null).forEach((x) => {