TaskPool——适合一次性任务扔进去等结果就行。但当你需要长驻后台线程、维护 WebSocket 连接、做持续的数据处理流水线时TaskPool 的用完即走模式就不够了。Worker 才是这种场景的正解——它创建一个持久线程通过 postMessage/onMessage 双向通信生命周期由你控制。这篇把 Worker 的创建、通信、销毁完整讲清楚。Worker vs TaskPool 怎么选维度TaskPoolWorker生命周期临时任务完成即销毁持久手动创建和销毁通信单次传参返回值持续双向 postMessage适合场景一次性计算、图片处理长连接、数据流、持续监听线程数框架自动管理开发者手动管理数据传递结构化克隆结构化克隆 SharedArrayBuffer决策算一次拿结果→TaskPool。持续跑、持续通信→Worker。Worker 基础Worker 分两端宿主线程主线程和 Worker 线程。两端的代码在不同文件中。主线程创建 Workerimport{worker}fromkit.ArkTSletworkerInstance:worker.ThreadWorkernewworker.ThreadWorker(../workers/MyWorker.ets)注意Worker 脚本路径相对于主线程文件。../workers/MyWorker.ets表示上层 workers 目录。Worker 线程脚本// workers/MyWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPort workerPort.onmessage(e:MessageEvents){letdata:stringe.dataasstring// 处理数据letresult:string处理完成:${data}workerPort.postMessage(result)}双向通信主线程 → WorkerpostMessageWorker → 主线程workerPort.postMessageinterfaceWorkerMessage{type:stringpayload:string}EntryComponentstruct WorkerBasicDemo{StatereceivedMsg:string等待消息...StatesentMsg:stringStateisRunning:booleanfalseprivateworkerInstance:worker.ThreadWorker|nullnullbuild(){Column({space:16}){Text(Worker 双向通信).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Text(收到:${this.receivedMsg}).fontSize(16).padding(12).borderRadius(8).backgroundColor(#E8F5E9).width(100%)Row({space:8}){TextInput({placeholder:发送给Worker}).layoutWeight(1).onChange((value:string){this.sentMsgvalue})Button(发送).onClick((){if(this.workerInstance){this.workerInstance.postMessage(this.sentMsg)}})}Row({space:8}){Button(创建Worker).onClick(()this.createWorker())Button(销毁Worker).onClick(()this.destroyWorker())}Text(this.isRunning?Worker运行中:Worker未创建).fontSize(13).fontColor(this.isRunning?#4CAF50:#999999)}.width(100%).padding(20)}privatecreateWorker():void{this.workerInstancenewworker.ThreadWorker(../workers/MyWorker.ets)this.isRunningtruethis.workerInstance.onmessage(e:MessageEvents){this.receivedMsge.dataasstring}this.workerInstance.onerror(e:ErrorEvent){this.receivedMsgWorker错误:${e.message}}this.workerInstance.onexit(){this.isRunningfalse}}privatedestroyWorker():void{if(this.workerInstance){this.workerInstance.terminate()this.workerInstancenull}}}关键区别主线程用workerInstance.onmessage接收Worker 线程用workerPort.onmessage接收。两个 onmessage 方向相反。Worker 生命周期Worker 的生命周期完全由开发者控制创建 → 通信 → 销毁。事件触发时机主线程Worker线程onmessage收到对方消息workerInstance.onmessageworkerPort.onmessageonerror运行出错workerInstance.onerrorworkerPort.onerroronexitWorker退出workerInstance.onexitworkerPort.onexitonmessageerror无法序列化的消息workerInstance.onmessageerrorworkerPort.onmessageerror销毁方式主线程调用workerInstance.terminate()——主动销毁Worker 线程调用workerPort.close()——自行退出两者效果相同Worker 的 onexit 都会触发重计算实战Fibonacci主线程算 Fibonacci(n40) 会卡死 UIWorker 算则 UI 完全不卡。// workers/FibWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPortfunctionfibonacci(n:number):number{if(n1)returnnreturnfibonacci(n-1)fibonacci(n-2)}workerPort.onmessage(e:MessageEvents){letn:numbere.dataasnumberletstart:numberDate.now()letresult:numberfibonacci(n)letduration:numberDate.now()-start workerPort.postMessage(fib(${n})${result}, 耗时${duration}ms)}EntryComponentstruct FibWorkerDemo{StateresultText:string等待计算...StateinputN:string40StateisComputing:booleanfalseprivatefibWorker:worker.ThreadWorker|nullnullaboutToAppear():void{this.fibWorkernewworker.ThreadWorker(../workers/FibWorker.ets)this.fibWorker.onmessage(e:MessageEvents){this.resultTexte.dataasstringthis.isComputingfalse}}aboutToDisappear():void{if(this.fibWorker){this.fibWorker.terminate()this.fibWorkernull}}build(){Column({space:16}){Text(Fibonacci Worker 计算).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Text(主线程算大数会卡UIWorker算UI完全不卡).fontSize(13).fontColor(#999999).width(100%)Row({space:8}){TextInput({text:this.inputN,placeholder:输入n值}).width(100).type(InputType.Number).onChange((value:string){this.inputNvalue})Button(Worker计算).enabled(!this.isComputing).onClick((){this.isComputingtruethis.resultText计算中...this.fibWorker?.postMessage(parseInt(this.inputN))})}Text(this.resultText).fontSize(16).padding(12).borderRadius(8).backgroundColor(this.isComputing?#FFF3E0:#E8F5E9).width(100%)if(this.isComputing){LoadingProgress().width(40).height(40)Text(UI 仍然可以交互滑动、点击).fontSize(13).fontColor(#4CAF50)}}.width(100%).padding(20)}}要点Worker 计算期间主线程 UI 完全不卡——可以自由滑动、点击、切换页面。计算完成后 onmessage 回调更新结果。数据流实战进度回调Worker 处理多步骤任务时通过多次 postMessage 报告进度。// workers/ProgressWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPort workerPort.onmessage(e:MessageEvents){lettotal:numbere.dataasnumberletsteps:number5letstepSize:numberMath.floor(total/steps)for(leti:number1;isteps;i){// 模拟每个步骤的处理letprocessed:numberi*stepSize// 报告进度workerPort.postMessage(步骤${i}/${steps}完成已处理${processed}条)}workerPort.postMessage(全部完成共处理${total}条数据)}EntryComponentstruct ProgressWorkerDemo{StateprogressList:string[][]StatetotalCount:string1000StateisProcessing:booleanfalseprivateprogressWorker:worker.ThreadWorker|nullnullaboutToAppear():void{this.progressWorkernewworker.ThreadWorker(../workers/ProgressWorker.ets)this.progressWorker.onmessage(e:MessageEvents){letmsg:stringe.dataasstringthis.progressList.push(msg)if(msg.includes(全部完成)){this.isProcessingfalse}}}aboutToDisappear():void{this.progressWorker?.terminate()}build(){Column({space:16}){Text(Worker 进度回调).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Row({space:8}){TextInput({text:this.totalCount,placeholder:数据总量}).width(100).type(InputType.Number).onChange((value:string){this.totalCountvalue})Button(开始处理).enabled(!this.isProcessing).onClick((){this.progressList[]this.isProcessingtruethis.progressWorker?.postMessage(parseInt(this.totalCount))})}ForEach(this.progressList,(msg:string,index:number){Text(msg).fontSize(13).fontColor(msg.includes(全部完成)?#4CAF50:#333333).width(100%).padding(4)},(msg:string,index:number)${index})}.width(100%).padding(20)}}要点Worker 内可以多次调用 workerPort.postMessage主线程每次 onmessage 都能收到。适合多步骤任务的进度汇报。Worker 注意事项Worker 数量有限制——最多创建 8 个 Worker含主线程超过会报错Worker 文件不能引用 ArkUI 装饰器——Component/State 等在 Worker 中不可用数据传递用结构化克隆——函数、DOM 节点、Observed 类实例不能直接传Worker 文件路径——相对于调用者文件建议放在 workers 目录下必须在 aboutToDisappear 中销毁——否则内存泄漏踩坑清单问题原因解决Worker 创建报错路径不对或文件不存在检查相对路径确认文件存在Worker 内引用 State 报错ArkUI 装饰器不能在 Worker 用Worker 只做纯逻辑不碰 UIpostMessage 传 Observed 对象报错对象含不可克隆属性传纯数据对象interface/RecordWorker 超过8个数量限制用 TaskPool 替代或合并 Workeronmessage 收不到通信方向搞反主线程→workerInstanceWorker→workerPortterminate 后 postMessage 崩溃Worker 已销毁terminate 后置 null发送前检查Worker 内 import 报错导入了含装饰器的文件Worker 只 import 纯逻辑模块数据传递慢结构化克隆拷贝大数据用 SharedArrayBuffer 零拷贝Worker onexit 不触发未正确销毁用 terminate() 或 close() 主动退出重复创建 Worker未检查已有实例创建前先销毁旧实例
TaskPool——适合一次性任务扔进去等结果就行。但当你需要长驻后台线程、维护 WebSocket 连接、做持续的数据处理流水线时TaskPool 的用完即走模式就不够了。Worker 才是这种场景的正解——它创建一个持久线程通过 postMessage/onMessage 双向通信生命周期由你控制。这篇把 Worker 的创建、通信、销毁完整讲清楚。Worker vs TaskPool 怎么选维度TaskPoolWorker生命周期临时任务完成即销毁持久手动创建和销毁通信单次传参返回值持续双向 postMessage适合场景一次性计算、图片处理长连接、数据流、持续监听线程数框架自动管理开发者手动管理数据传递结构化克隆结构化克隆 SharedArrayBuffer决策算一次拿结果→TaskPool。持续跑、持续通信→Worker。Worker 基础Worker 分两端宿主线程主线程和 Worker 线程。两端的代码在不同文件中。主线程创建 Workerimport{worker}fromkit.ArkTSletworkerInstance:worker.ThreadWorkernewworker.ThreadWorker(../workers/MyWorker.ets)注意Worker 脚本路径相对于主线程文件。../workers/MyWorker.ets表示上层 workers 目录。Worker 线程脚本// workers/MyWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPort workerPort.onmessage(e:MessageEvents){letdata:stringe.dataasstring// 处理数据letresult:string处理完成:${data}workerPort.postMessage(result)}双向通信主线程 → WorkerpostMessageWorker → 主线程workerPort.postMessageinterfaceWorkerMessage{type:stringpayload:string}EntryComponentstruct WorkerBasicDemo{StatereceivedMsg:string等待消息...StatesentMsg:stringStateisRunning:booleanfalseprivateworkerInstance:worker.ThreadWorker|nullnullbuild(){Column({space:16}){Text(Worker 双向通信).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Text(收到:${this.receivedMsg}).fontSize(16).padding(12).borderRadius(8).backgroundColor(#E8F5E9).width(100%)Row({space:8}){TextInput({placeholder:发送给Worker}).layoutWeight(1).onChange((value:string){this.sentMsgvalue})Button(发送).onClick((){if(this.workerInstance){this.workerInstance.postMessage(this.sentMsg)}})}Row({space:8}){Button(创建Worker).onClick(()this.createWorker())Button(销毁Worker).onClick(()this.destroyWorker())}Text(this.isRunning?Worker运行中:Worker未创建).fontSize(13).fontColor(this.isRunning?#4CAF50:#999999)}.width(100%).padding(20)}privatecreateWorker():void{this.workerInstancenewworker.ThreadWorker(../workers/MyWorker.ets)this.isRunningtruethis.workerInstance.onmessage(e:MessageEvents){this.receivedMsge.dataasstring}this.workerInstance.onerror(e:ErrorEvent){this.receivedMsgWorker错误:${e.message}}this.workerInstance.onexit(){this.isRunningfalse}}privatedestroyWorker():void{if(this.workerInstance){this.workerInstance.terminate()this.workerInstancenull}}}关键区别主线程用workerInstance.onmessage接收Worker 线程用workerPort.onmessage接收。两个 onmessage 方向相反。Worker 生命周期Worker 的生命周期完全由开发者控制创建 → 通信 → 销毁。事件触发时机主线程Worker线程onmessage收到对方消息workerInstance.onmessageworkerPort.onmessageonerror运行出错workerInstance.onerrorworkerPort.onerroronexitWorker退出workerInstance.onexitworkerPort.onexitonmessageerror无法序列化的消息workerInstance.onmessageerrorworkerPort.onmessageerror销毁方式主线程调用workerInstance.terminate()——主动销毁Worker 线程调用workerPort.close()——自行退出两者效果相同Worker 的 onexit 都会触发重计算实战Fibonacci主线程算 Fibonacci(n40) 会卡死 UIWorker 算则 UI 完全不卡。// workers/FibWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPortfunctionfibonacci(n:number):number{if(n1)returnnreturnfibonacci(n-1)fibonacci(n-2)}workerPort.onmessage(e:MessageEvents){letn:numbere.dataasnumberletstart:numberDate.now()letresult:numberfibonacci(n)letduration:numberDate.now()-start workerPort.postMessage(fib(${n})${result}, 耗时${duration}ms)}EntryComponentstruct FibWorkerDemo{StateresultText:string等待计算...StateinputN:string40StateisComputing:booleanfalseprivatefibWorker:worker.ThreadWorker|nullnullaboutToAppear():void{this.fibWorkernewworker.ThreadWorker(../workers/FibWorker.ets)this.fibWorker.onmessage(e:MessageEvents){this.resultTexte.dataasstringthis.isComputingfalse}}aboutToDisappear():void{if(this.fibWorker){this.fibWorker.terminate()this.fibWorkernull}}build(){Column({space:16}){Text(Fibonacci Worker 计算).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Text(主线程算大数会卡UIWorker算UI完全不卡).fontSize(13).fontColor(#999999).width(100%)Row({space:8}){TextInput({text:this.inputN,placeholder:输入n值}).width(100).type(InputType.Number).onChange((value:string){this.inputNvalue})Button(Worker计算).enabled(!this.isComputing).onClick((){this.isComputingtruethis.resultText计算中...this.fibWorker?.postMessage(parseInt(this.inputN))})}Text(this.resultText).fontSize(16).padding(12).borderRadius(8).backgroundColor(this.isComputing?#FFF3E0:#E8F5E9).width(100%)if(this.isComputing){LoadingProgress().width(40).height(40)Text(UI 仍然可以交互滑动、点击).fontSize(13).fontColor(#4CAF50)}}.width(100%).padding(20)}}要点Worker 计算期间主线程 UI 完全不卡——可以自由滑动、点击、切换页面。计算完成后 onmessage 回调更新结果。数据流实战进度回调Worker 处理多步骤任务时通过多次 postMessage 报告进度。// workers/ProgressWorker.etsimport{worker,MessageEvents,ThreadWorkerGlobalScope}fromkit.ArkTSconstworkerPort:ThreadWorkerGlobalScopeworker.workerPort workerPort.onmessage(e:MessageEvents){lettotal:numbere.dataasnumberletsteps:number5letstepSize:numberMath.floor(total/steps)for(leti:number1;isteps;i){// 模拟每个步骤的处理letprocessed:numberi*stepSize// 报告进度workerPort.postMessage(步骤${i}/${steps}完成已处理${processed}条)}workerPort.postMessage(全部完成共处理${total}条数据)}EntryComponentstruct ProgressWorkerDemo{StateprogressList:string[][]StatetotalCount:string1000StateisProcessing:booleanfalseprivateprogressWorker:worker.ThreadWorker|nullnullaboutToAppear():void{this.progressWorkernewworker.ThreadWorker(../workers/ProgressWorker.ets)this.progressWorker.onmessage(e:MessageEvents){letmsg:stringe.dataasstringthis.progressList.push(msg)if(msg.includes(全部完成)){this.isProcessingfalse}}}aboutToDisappear():void{this.progressWorker?.terminate()}build(){Column({space:16}){Text(Worker 进度回调).fontSize(22).fontWeight(FontWeight.Bold).width(100%)Row({space:8}){TextInput({text:this.totalCount,placeholder:数据总量}).width(100).type(InputType.Number).onChange((value:string){this.totalCountvalue})Button(开始处理).enabled(!this.isProcessing).onClick((){this.progressList[]this.isProcessingtruethis.progressWorker?.postMessage(parseInt(this.totalCount))})}ForEach(this.progressList,(msg:string,index:number){Text(msg).fontSize(13).fontColor(msg.includes(全部完成)?#4CAF50:#333333).width(100%).padding(4)},(msg:string,index:number)${index})}.width(100%).padding(20)}}要点Worker 内可以多次调用 workerPort.postMessage主线程每次 onmessage 都能收到。适合多步骤任务的进度汇报。Worker 注意事项Worker 数量有限制——最多创建 8 个 Worker含主线程超过会报错Worker 文件不能引用 ArkUI 装饰器——Component/State 等在 Worker 中不可用数据传递用结构化克隆——函数、DOM 节点、Observed 类实例不能直接传Worker 文件路径——相对于调用者文件建议放在 workers 目录下必须在 aboutToDisappear 中销毁——否则内存泄漏踩坑清单问题原因解决Worker 创建报错路径不对或文件不存在检查相对路径确认文件存在Worker 内引用 State 报错ArkUI 装饰器不能在 Worker 用Worker 只做纯逻辑不碰 UIpostMessage 传 Observed 对象报错对象含不可克隆属性传纯数据对象interface/RecordWorker 超过8个数量限制用 TaskPool 替代或合并 Workeronmessage 收不到通信方向搞反主线程→workerInstanceWorker→workerPortterminate 后 postMessage 崩溃Worker 已销毁terminate 后置 null发送前检查Worker 内 import 报错导入了含装饰器的文件Worker 只 import 纯逻辑模块数据传递慢结构化克隆拷贝大数据用 SharedArrayBuffer 零拷贝Worker onexit 不触发未正确销毁用 terminate() 或 close() 主动退出重复创建 Worker未检查已有实例创建前先销毁旧实例
