PowerJob版本:4.3.2-main。
PowerJob是全新一代的分布式任务调度与计算框架,官网地址:http://www.powerjob.tech/。其中介绍了PowerJob的功能特点,以及与其他调度框架的对比,这里就不再赘述了。
以上是PowerJob的架构图,取自官网。可以看出,PowerJob是典型的客户端/服务端交互的架构(但是在PowerJob中却没有一般分布式中间件会有的注册中心)。本文就从启动流程出发,来一起探究下PowerJob在启动阶段中都做了些什么动作。
既然要看启动流程源码,那么首先就来看下启动类。PowerJob依赖于Spring Boot,启动类为PowerJobServerApplication:
/**
* powerjob-server entry
*
* @author tjq
* @since 2020/3/29
*/
@Slf4j
@EnableScheduling
@SpringBootApplication
public class PowerJobServerApplication {
private static final String TIPS = "\n\n" +
"******************* PowerJob Tips *******************\n" +
"如果应用无法启动,我们建议您仔细阅读以下文档来解决:\n" +
"if server can't startup, we recommend that you read the documentation to find a solution:\n" +
"https://www.yuque.com/powerjob/guidence/problem\n" +
"******************* PowerJob Tips *******************\n\n";
public static void main(String[] args) {
pre();
// Start SpringBoot application.
try {
SpringApplication.run(PowerJobServerApplication.class, args);
} catch (Throwable t) {
log.error(TIPS);
throw t;
}
}
private static void pre() {
log.info(TIPS);
PropertyUtils.init();
}
}
/**
* 加载配置文件
*
* @author tjq
* @since 2020/5/18
*/
@Slf4j
public class PropertyUtils {
private static final Properties PROPERTIES = new Properties();
public static Properties getProperties() {
return PROPERTIES;
}
public static void init() {
URL propertiesURL = PropertyUtils.class.getClassLoader().getResource("application.properties");
Objects.requireNonNull(propertiesURL);
try (InputStream is = propertiesURL.openStream()) {
PROPERTIES.load(is);
} catch (Exception e) {
ExceptionUtils.rethrow(e);
}
}
}
其中pre方法只是将配置文件内容加载进Properties缓存中。
既然启动类看不出什么逻辑,那么接下来就来看下服务端启动时有没有什么初始化的逻辑:
PowerTransportService类是用作数据传输服务的,也就是客户端和服务端之间的通信。其实现了InitializingBean接口,所以查看下其afterPropertiesSet方法的实现:
/**
* PowerTransportService:
*/
@Override
public void afterPropertiesSet() throws Exception {
log.info("[PowerTransportService] start to initialize whole PowerTransportService!");
log.info("[PowerTransportService] activeProtocols: {}", activeProtocols);
if (StringUtils.isEmpty(activeProtocols)) {
throw new IllegalArgumentException("activeProtocols can't be empty!");
}
for (String protocol : activeProtocols.split(OmsConstant.COMMA)) {
try {
final int port = parseProtocolPort(protocol);
//初始化网络通讯
initRemoteFrameWork(protocol, port);
} catch (Throwable t) {
log.error("[PowerTransportService] initialize protocol[{}] failed. If you don't need to use this protocol, you can turn it off by 'oms.transporter.active.protocols'", protocol);
ExceptionUtils.rethrow(t);
}
}
//选择默认的通信协议,默认为HTTP
choseDefault();
log.info("[PowerTransportService] initialize successfully!");
log.info("[PowerTransportService] ALL_PROTOCOLS: {}", protocolName2Info);
}
/**
* 第18行代码处:
*/
private void initRemoteFrameWork(String protocol, int port) {
// 从构造器注入改为从 applicationContext 获取来避免循环依赖
//获取所有注解了@Actor的bean
final Map beansWithAnnotation = applicationContext.getBeansWithAnnotation(Actor.class);
log.info("[PowerTransportService] find Actor num={},names={}", beansWithAnnotation.size(), beansWithAnnotation.keySet());
Address address = new Address()
.setHost(NetUtils.getLocalHost())
.setPort(port);
EngineConfig engineConfig = new EngineConfig()
.setServerType(ServerType.SERVER)
.setType(protocol.toUpperCase())
.setBindAddress(address)
.setActorList(Lists.newArrayList(beansWithAnnotation.values()));
log.info("[PowerTransportService] start to initialize RemoteEngine[type={},address={}]", protocol, address);
RemoteEngine re = new PowerJobRemoteEngine();
//初始化网络层
final EngineOutput engineOutput = re.start(engineConfig);
log.info("[PowerTransportService] start RemoteEngine[type={},address={}] successfully", protocol, address);
//放入相关缓存中
this.engines.add(re);
this.protocolName2Info.put(protocol, new ProtocolInfo(protocol, address.toFullAddress(), engineOutput.getTransporter()));
}
其中需要说明的一点是:PowerJob网络层使用的协议是Akka或Vert.x:
- Akka是一个在JVM上构建高并发、分布式和弹性消息驱动的应用程序。其是用Scala写的,使用到了Actor模型;
- 而Vert.x是一个在JVM上构建响应式应用程序的工具包,其底层基于Netty(之前我对Netty的源码也进行过分析,感兴趣的话可以查看《较真儿学源码系列-Netty核心流程源码分析》)。
PowerJob屏蔽了底层的实现,用两个自定义的注解@Actor和@Handler进行了统一的封装。客户端传过来的请求会自动跳转到@Actor注解的类、@Handler注解的方法上。拿客户端发送心跳给服务端的逻辑为例,流程如下所示:
下面就来继续看下,在上面的第53行代码处。PowerJob是如何完成这个绑定的:
/**
* PowerJobRemoteEngine:
*/
@Override
public EngineOutput start(EngineConfig engineConfig) {
final String engineType = engineConfig.getType();
EngineOutput engineOutput = new EngineOutput();
log.info("[PowerJobRemoteEngine] [{}] start remote engine with config: {}", engineType, engineConfig);
//获取所有@Actor的类,和其中注解了@Handler的方法
List actorInfos = ActorFactory.load(engineConfig.getActorList());
//遍历获取指定协议的CSInitializer
csInitializer = CSInitializerFactory.build(engineType);
String type = csInitializer.type();
Stopwatch sw = Stopwatch.createStarted();
log.info("[PowerJobRemoteEngine] [{}] try to startup CSInitializer[type={}]", engineType, type);
//CsInitializer初始化,这里以Vert.x为例,查看下其实现
csInitializer.init(new CSInitializerConfig()
.setBindAddress(engineConfig.getBindAddress())
.setServerType(engineConfig.getServerType())
);
// 构建通讯器
Transporter transporter = csInitializer.buildTransporter();
engineOutput.setTransporter(transporter);
log.info("[PowerJobRemoteEngine] [{}] start to bind Handler", engineType);
actorInfos.forEach(actor -> actor.getHandlerInfos().forEach(handlerInfo -> log.info("[PowerJobRemoteEngine] [{}] PATH={}, handler={}", engineType, handlerInfo.getLocation().toPath(), handlerInfo.getMethod())));
// 绑定 handler
csInitializer.bindHandlers(actorInfos);
log.info("[PowerJobRemoteEngine] [{}] startup successfully, cost: {}", engineType, sw);
return engineOutput;
}
/**
* ActorFactory:
* 第11行代码处:
*/
static List load(List
通过上面的buildRequestHandler方法可知,当Vert.x接收到客户端的请求时,会调用到一个回调函数。而这个回调函数会最终通过反射的方式调用到自定义的@Handler注解的方法中来。
上面只是完成了初始化和绑定的操作,还缺少具体调用时候的逻辑(这里属于一个整体流程,所以就一起分析了)。继续拿发送心跳为例,调用的方法是:TransportUtils.reportWorkerHeartbeat:
/**
* TransportUtils:
*/
public static void reportWorkerHeartbeat(WorkerHeartbeat req, String address, Transporter transporter) {
//绑定url调用信息
final URL url = easyBuildUrl(ServerType.SERVER, S4W_PATH, S4W_HANDLER_WORKER_HEARTBEAT, address);
//这里依旧是拿Vert.x的实现来分析
transporter.tell(url, req);
}
/**
* VertxTransporter:
* 第8行代码处:
*/
@Override
public void tell(URL url, PowerSerializable request) {
post(url, request, null);
}
/**
* 这里就涉及到Vert.x的具体调用细节了,不再深入
*/
@SuppressWarnings("unchecked")
private CompletionStage post(URL url, PowerSerializable request, Class clz) {
final String host = url.getAddress().getHost();
final int port = url.getAddress().getPort();
final String path = url.getLocation().toPath();
RequestOptions requestOptions = new RequestOptions()
.setMethod(HttpMethod.POST)
.setHost(host)
.setPort(port)
.setURI(path);
// 获取远程服务器的HTTP连接
Future httpClientRequestFuture = httpClient.request(requestOptions);
// 转换 -> 发送请求获取响应
Future responseFuture = httpClientRequestFuture.compose(httpClientRequest ->
httpClientRequest
.putHeader(HttpHeaderNames.CONTENT_TYPE, HttpHeaderValues.APPLICATION_JSON)
.send(JsonObject.mapFrom(request).toBuffer())
);
return responseFuture.compose(httpClientResponse -> {
// throw exception
final int statusCode = httpClientResponse.statusCode();
if (statusCode != HttpResponseStatus.OK.code()) {
// CompletableFuture.get() 时会传递抛出该异常
throw new RemotingException(String.format("request [host:%s,port:%s,url:%s] failed, status: %d, msg: %s",
host, port, path, statusCode, httpClientResponse.statusMessage()
));
}
return httpClientResponse.body().compose(x -> {
if (clz == null) {
return Future.succeededFuture(null);
}
if (clz.equals(String.class)) {
return Future.succeededFuture((T) x.toString());
}
return Future.succeededFuture(x.toJsonObject().mapTo(clz));
});
}).toCompletionStage();
}
当客户端往服务端发送完请求后,服务端接受到相关的请求,会调用到相应的回调函数,从而最终调用到@Handler注解的方法。整个流程就串起来了。
同PowerTransportService类一样,InstanceMetadataService也实现了InitializingBean接口,所以查看下其afterPropertiesSet方法的实现:
/**
* InstanceMetadataService:
*/
@Override
public void afterPropertiesSet() throws Exception {
instanceId2JobInfoCache = CacheBuilder.newBuilder()
.concurrencyLevel(CACHE_CONCURRENCY_LEVEL)
.maximumSize(instanceMetadataCacheSize)
.softValues()
.build();
}
其中只是初始化了一个本地缓存,没有多余的逻辑。
CoreScheduleTaskManager也实现了InitializingBean接口,查看下其afterPropertiesSet方法的实现:
/**
* CoreScheduleTaskManager:
*/
@SuppressWarnings("AlibabaAvoidManuallyCreateThread")
@Override
public void afterPropertiesSet() {
// 定时调度
coreThreadContainer.add(new Thread(new LoopRunnable("ScheduleCronJob", PowerScheduleService.SCHEDULE_RATE, () -> powerScheduleService.scheduleNormalJob(TimeExpressionType.CRON)), "Thread-ScheduleCronJob"));
coreThreadContainer.add(new Thread(new LoopRunnable("ScheduleDailyTimeIntervalJob", PowerScheduleService.SCHEDULE_RATE, () -> powerScheduleService.scheduleNormalJob(TimeExpressionType.DAILY_TIME_INTERVAL)), "Thread-ScheduleDailyTimeIntervalJob"));
coreThreadContainer.add(new Thread(new LoopRunnable("ScheduleCronWorkflow", PowerScheduleService.SCHEDULE_RATE, powerScheduleService::scheduleCronWorkflow), "Thread-ScheduleCronWorkflow"));
coreThreadContainer.add(new Thread(new LoopRunnable("ScheduleFrequentJob", PowerScheduleService.SCHEDULE_RATE, powerScheduleService::scheduleFrequentJob), "Thread-ScheduleFrequentJob"));
// 数据清理
coreThreadContainer.add(new Thread(new LoopRunnable("CleanWorkerData", PowerScheduleService.SCHEDULE_RATE, powerScheduleService::cleanData), "Thread-CleanWorkerData"));
// 状态检查
coreThreadContainer.add(new Thread(new LoopRunnable("CheckRunningInstance", InstanceStatusCheckService.CHECK_INTERVAL, instanceStatusCheckService::checkRunningInstance), "Thread-CheckRunningInstance"));
coreThreadContainer.add(new Thread(new LoopRunnable("CheckWaitingDispatchInstance", InstanceStatusCheckService.CHECK_INTERVAL, instanceStatusCheckService::checkWaitingDispatchInstance), "Thread-CheckWaitingDispatchInstance"));
coreThreadContainer.add(new Thread(new LoopRunnable("CheckWaitingWorkerReceiveInstance", InstanceStatusCheckService.CHECK_INTERVAL, instanceStatusCheckService::checkWaitingWorkerReceiveInstance), "Thread-CheckWaitingWorkerReceiveInstance"));
coreThreadContainer.add(new Thread(new LoopRunnable("CheckWorkflowInstance", InstanceStatusCheckService.CHECK_INTERVAL, instanceStatusCheckService::checkWorkflowInstance), "Thread-CheckWorkflowInstance"));
coreThreadContainer.forEach(Thread::start);
}
可以看到,其中初始化了一堆的定时任务。从中挑几个定时任务来分析下:
/**
* PowerScheduleService:
*/
public void scheduleNormalJob(TimeExpressionType timeExpressionType) {
long start = System.currentTimeMillis();
// 调度 CRON 表达式 JOB
try {
//获取在PowerJob控制台配置的appId,也就是服务id(PowerJob会使用数据库来存储服务、实例等相关数据。在进行一些操作前,会先落库,然后再执行。以此避免相关操作丢失的情况出现)
final List allAppIds = appInfoRepository.listAppIdByCurrentServer(transportService.defaultProtocol().getAddress());
if (CollectionUtils.isEmpty(allAppIds)) {
log.info("[NormalScheduler] current server has no app's job to schedule.");
return;
}
scheduleNormalJob0(timeExpressionType, allAppIds);
} catch (Exception e) {
log.error("[NormalScheduler] schedule cron job failed.", e);
}
long cost = System.currentTimeMillis() - start;
log.info("[NormalScheduler] {} job schedule use {} ms.", timeExpressionType, cost);
if (cost > SCHEDULE_RATE) {
log.warn("[NormalScheduler] The database query is using too much time({}ms), please check if the database load is too high!", cost);
}
}
/**
* 第14行代码处:
* 调度普通服务端计算表达式类型(CRON、DAILY_TIME_INTERVAL)的任务
*
* @param timeExpressionType 表达式类型
* @param appIds appIds
*/
private void scheduleNormalJob0(TimeExpressionType timeExpressionType, List appIds) {
long nowTime = System.currentTimeMillis();
long timeThreshold = nowTime + 2 * SCHEDULE_RATE;
//分组执行
Lists.partition(appIds, MAX_APP_NUM).forEach(partAppIds -> {
try {
// 查询条件:任务开启 + 使用CRON表达调度时间 + 指定appId + 即将需要调度执行
//获取在PowerJob控制台配置的执行任务
List jobInfos = jobInfoRepository.findByAppIdInAndStatusAndTimeExpressionTypeAndNextTriggerTimeLessThanEqual(partAppIds, SwitchableStatus.ENABLE.getV(), timeExpressionType.getV(), timeThreshold);
if (CollectionUtils.isEmpty(jobInfos)) {
return;
}
// 1. 批量写日志表
Map jobId2InstanceId = Maps.newHashMap();
log.info("[NormalScheduler] These {} jobs will be scheduled: {}.", timeExpressionType.name(), jobInfos);
jobInfos.forEach(jobInfo -> {
//实例表进行落库(任务和实例的关系为:每执行一次任务会生成一个实例)
Long instanceId = instanceService.create(jobInfo.getId(), jobInfo.getAppId(), jobInfo.getJobParams(), null, null, jobInfo.getNextTriggerTime()).getInstanceId();
jobId2InstanceId.put(jobInfo.getId(), instanceId);
});
instanceInfoRepository.flush();
// 2. 推入时间轮中等待调度执行(对时间轮代码的分析请查看https://blog.csdn.net/weixin_30342639/article/details/132732836)
jobInfos.forEach(jobInfoDO -> {
Long instanceId = jobId2InstanceId.get(jobInfoDO.getId());
//获取下次执行时间
long targetTriggerTime = jobInfoDO.getNextTriggerTime();
long delay = 0;
if (targetTriggerTime < nowTime) {
log.warn("[Job-{}] schedule delay, expect: {}, current: {}", jobInfoDO.getId(), targetTriggerTime, System.currentTimeMillis());
} else {
//计算距离下次执行时间所需要延迟的时间
delay = targetTriggerTime - nowTime;
}
//任务实例推入时间轮,等待被执行
InstanceTimeWheelService.schedule(instanceId, delay, () -> dispatchService.dispatch(jobInfoDO, instanceId, Optional.empty(), Optional.empty()));
});
// 3. 计算下一次调度时间(忽略5S内的重复执行,即CRON模式下最小的连续执行间隔为 SCHEDULE_RATE ms)
jobInfos.forEach(jobInfoDO -> {
try {
//重新计算下次执行时间,并落库
//这里会用到策略模式,不同类型的任务(Cron/固定频率/每日固定间隔)会有不同的计算方式,具体不再深入分析,感兴趣可自行查看
refreshJob(timeExpressionType, jobInfoDO);
} catch (Exception e) {
log.error("[Job-{}] refresh job failed.", jobInfoDO.getId(), e);
}
});
jobInfoRepository.flush();
} catch (Exception e) {
log.error("[NormalScheduler] schedule {} job failed.", timeExpressionType.name(), e);
}
});
}
在上面第76行代码处,当任务推入到时间轮之后(对时间轮代码的分析请查看《较真儿学源码系列-PowerJob时间轮源码分析》),等到需要被执行的时候,会调用到DispatchService.dispatch方法来派发任务:
/**
* DispatchService:
* 将任务从Server派发到Worker(TaskTracker)
* 只会派发当前状态为等待派发的任务实例
* **************************************************
* 2021-02-03 modify by Echo009
* 1、移除参数 当前运行次数、工作流实例ID、实例参数
* 更改为从当前任务实例中获取获取以上信息
* 2、移除运行次数相关的(runningTimes)处理逻辑
* 迁移至 {@link InstanceManager#updateStatus} 中处理
* **************************************************
*
* @param jobInfo 任务的元信息
* @param instanceId 任务实例ID
* @param instanceInfoOptional 任务实例信息,可选
* @param overloadOptional 超载信息,可选
*/
@UseCacheLock(type = "processJobInstance", key = "#jobInfo.getMaxInstanceNum() > 0 || T(tech.powerjob.common.enums.TimeExpressionType).FREQUENT_TYPES.contains(#jobInfo.getTimeExpressionType()) ? #jobInfo.getId() : #instanceId", concurrencyLevel = 1024)
public void dispatch(JobInfoDO jobInfo, Long instanceId, Optional instanceInfoOptional, Optional> overloadOptional) {
// 允许从外部传入实例信息,减少 io 次数
// 检查当前任务是否被取消
//获取实例
InstanceInfoDO instanceInfo = instanceInfoOptional.orElseGet(() -> instanceInfoRepository.findByInstanceId(instanceId));
Long jobId = instanceInfo.getJobId();
//...
// 任务信息已经被删除
if (jobInfo.getId() == null) {
log.warn("[Dispatcher-{}|{}] cancel dispatch due to job(id={}) has been deleted!", jobId, instanceId, jobId);
instanceManager.processFinishedInstance(instanceId, instanceInfo.getWfInstanceId(), FAILED, "can't find job by id " + jobId);
return;
}
Date now = new Date();
String dbInstanceParams = instanceInfo.getInstanceParams() == null ? "" : instanceInfo.getInstanceParams();
log.info("[Dispatcher-{}|{}] start to dispatch job: {};instancePrams: {}.", jobId, instanceId, jobInfo, dbInstanceParams);
// 查询当前运行的实例数
long current = System.currentTimeMillis();
Integer maxInstanceNum = jobInfo.getMaxInstanceNum();
// 秒级任务只派发到一台机器,具体的 maxInstanceNum 由 TaskTracker 控制
if (TimeExpressionType.FREQUENT_TYPES.contains(jobInfo.getTimeExpressionType())) {
maxInstanceNum = 1;
}
//...
// 获取当前最合适的 worker 列表
List suitableWorkers = workerClusterQueryService.getSuitableWorkers(jobInfo);
//...
// 判断是否超载,在所有可用 worker 超载的情况下直接跳过当前任务
suitableWorkers = filterOverloadWorker(suitableWorkers);
//...
//获取worker ip
List workerIpList = suitableWorkers.stream().map(WorkerInfo::getAddress).collect(Collectors.toList());
// 构造任务调度请求
ServerScheduleJobReq req = constructServerScheduleJobReq(jobInfo, instanceInfo, workerIpList);
// 发送请求(不可靠,需要一个后台线程定期轮询状态)
//只取第一个worker
WorkerInfo taskTracker = suitableWorkers.get(0);
String taskTrackerAddress = taskTracker.getAddress();
URL workerUrl = ServerURLFactory.dispatchJob2Worker(taskTrackerAddress);
transportService.tell(taskTracker.getProtocol(), workerUrl, req);
log.info("[Dispatcher-{}|{}] send schedule request to TaskTracker[protocol:{},address:{}] successfully: {}.", jobId, instanceId, taskTracker.getProtocol(), taskTrackerAddress, req);
// 修改状态
//修改实例表状态
instanceInfoRepository.update4TriggerSucceed(instanceId, WAITING_WORKER_RECEIVE.getV(), current, taskTrackerAddress, now, instanceInfo.getStatus());
// 装载缓存
instanceMetadataService.loadJobInfo(instanceId, jobInfo);
}
/**
* InstanceManager:
* 第31行代码处:
* 收尾完成的任务实例
*
* @param instanceId 任务实例ID
* @param wfInstanceId 工作流实例ID,非必须
* @param status 任务状态,有 成功/失败/手动停止
* @param result 执行结果
*/
public void processFinishedInstance(Long instanceId, Long wfInstanceId, InstanceStatus status, String result) {
log.info("[Instance-{}] process finished, final status is {}.", instanceId, status.name());
// 上报日志数据
//时间轮延迟执行
HashedWheelTimerHolder.INACCURATE_TIMER.schedule(() -> instanceLogService.sync(instanceId), 60, TimeUnit.SECONDS);
// workflow 特殊处理
if (wfInstanceId != null) {
// 手动停止在工作流中也认为是失败(理论上不应该发生)
workflowInstanceManager.move(wfInstanceId, instanceId, status, result);
}
// 告警
if (status == InstanceStatus.FAILED) {
alert(instanceId, result);
}
// 主动移除缓存,减小内存占用
instanceMetadataService.invalidateJobInfo(instanceId);
}
/**
* InstanceLogService:
* 第96行代码处:
* 将本地的任务实例运行日志同步到 mongoDB 存储,在任务执行结束后异步执行
*
* @param instanceId 任务实例ID
*/
@Async(PJThreadPool.BACKGROUND_POOL)
public void sync(Long instanceId) {
Stopwatch sw = Stopwatch.createStarted();
try {
// 先持久化到本地文件
File stableLogFile = genStableLogFile(instanceId);
// 将文件推送到 MongoDB
if (gridFsManager.available()) {
try {
gridFsManager.store(stableLogFile, GridFsManager.LOG_BUCKET, genMongoFileName(instanceId));
log.info("[InstanceLog-{}] push local instanceLogs to mongoDB succeed, using: {}.", instanceId, sw.stop());
} catch (Exception e) {
log.warn("[InstanceLog-{}] push local instanceLogs to mongoDB failed.", instanceId, e);
}
}
} catch (Exception e) {
log.warn("[InstanceLog-{}] sync local instanceLogs failed.", instanceId, e);
}
// 删除本地数据库数据
try {
instanceId2LastReportTime.remove(instanceId);
CommonUtils.executeWithRetry0(() -> localInstanceLogRepository.deleteByInstanceId(instanceId));
log.info("[InstanceLog-{}] delete local instanceLog successfully.", instanceId);
} catch (Exception e) {
log.warn("[InstanceLog-{}] delete local instanceLog failed.", instanceId, e);
}
}
/**
* WorkerClusterQueryService:
* 第50行代码处:
* get worker for job
*
* @param jobInfo job
* @return worker cluster info, sorted by metrics desc
*/
public List getSuitableWorkers(JobInfoDO jobInfo) {
//获取该集群所有的机器信息
List workers = Lists.newLinkedList(getWorkerInfosByAppId(jobInfo.getAppId()).values());
workers.removeIf(workerInfo -> filterWorker(workerInfo, jobInfo));
DispatchStrategy dispatchStrategy = DispatchStrategy.of(jobInfo.getDispatchStrategy());
switch (dispatchStrategy) {
case RANDOM:
//随机的方式就是打乱顺序
Collections.shuffle(workers);
break;
case HEALTH_FIRST:
//健康优先的方式需要计算下得分,按得分高低排序
workers.sort((o1, o2) -> o2.getSystemMetrics().calculateScore() - o1.getSystemMetrics().calculateScore());
break;
default:
// do nothing
}
// 限定集群大小(0代表不限制)
if (!workers.isEmpty() && jobInfo.getMaxWorkerCount() > 0 && workers.size() > jobInfo.getMaxWorkerCount()) {
workers = workers.subList(0, jobInfo.getMaxWorkerCount());
}
return workers;
}
/**
* SystemMetrics:
* 第171行代码处:
* Calculate score, based on CPU and memory info.
*
* @return score
*/
public int calculateScore() {
if (score > 0) {
return score;
}
// Memory is vital to TaskTracker, so we set the multiplier factor as 2.
//未使用内存指标的权重为2
double memScore = (jvmMaxMemory - jvmUsedMemory) * 2;
// Calculate the remaining load of CPU. Multiplier is set as 1.
//剩余可用cpu数的权重为1
double cpuScore = cpuProcessors - cpuLoad;
// Windows can not fetch CPU load, set cpuScore as 1.
//Windows系统拿不到cpu使用数,所以cpu得分设置为1
if (cpuScore > cpuProcessors) {
cpuScore = 1;
}
//最终得分就是内存得分+cpu得分
score = (int) (memScore + cpuScore);
return score;
}
/**
* ServerURLFactory:
* 第69行代码处:
*/
public static URL dispatchJob2Worker(String address) {
return simileBuild(address, ServerType.WORKER, WORKER_PATH, WTT_HANDLER_RUN_JOB);
}
在上面第216行代码处会将请求跳转到WorkerActor.onReceiveServerScheduleJobReq方法:
/**
* WorkerActor:
*/
@Handler(path = WTT_HANDLER_RUN_JOB)
public void onReceiveServerScheduleJobReq(ServerScheduleJobReq req) {
taskTrackerActor.onReceiveServerScheduleJobReq(req);
}
/**
* TaskTrackerActor:
* 服务器任务调度处理器
*/
@Handler(path = WTT_HANDLER_RUN_JOB)
public void onReceiveServerScheduleJobReq(ServerScheduleJobReq req) {
log.debug("[TaskTrackerActor] server schedule job by request: {}.", req);
Long instanceId = req.getInstanceId();
// 区分轻量级任务模型以及重量级任务模型
//单机执行的OpenApi/Corn/工作流是轻量级任务,其他的是重量级任务
if (isLightweightTask(req)) {
final LightTaskTracker taskTracker = LightTaskTrackerManager.getTaskTracker(instanceId);
if (taskTracker != null) {
log.warn("[TaskTrackerActor] LightTaskTracker({}) for instance(id={}) already exists.", taskTracker, instanceId);
return;
}
// 判断是否已经 overload
if (LightTaskTrackerManager.currentTaskTrackerSize() >= workerRuntime.getWorkerConfig().getMaxLightweightTaskNum() * LightTaskTrackerManager.OVERLOAD_FACTOR) {
// ignore this request
log.warn("[TaskTrackerActor] this worker is overload,ignore this request(instanceId={}),current size = {}!", instanceId, LightTaskTrackerManager.currentTaskTrackerSize());
return;
}
if (LightTaskTrackerManager.currentTaskTrackerSize() >= workerRuntime.getWorkerConfig().getMaxLightweightTaskNum()) {
log.warn("[TaskTrackerActor] this worker will be overload soon,current size = {}!", LightTaskTrackerManager.currentTaskTrackerSize());
}
// 创建轻量级任务
//在轻量级本地缓存中添加TaskTracker
LightTaskTrackerManager.atomicCreateTaskTracker(instanceId, ignore -> LightTaskTracker.create(req, workerRuntime));
} else {
HeavyTaskTracker taskTracker = HeavyTaskTrackerManager.getTaskTracker(instanceId);
if (taskTracker != null) {
log.warn("[TaskTrackerActor] HeavyTaskTracker({}) for instance(id={}) already exists.", taskTracker, instanceId);
return;
}
// 判断是否已经 overload
if (HeavyTaskTrackerManager.currentTaskTrackerSize() >= workerRuntime.getWorkerConfig().getMaxHeavyweightTaskNum()) {
// ignore this request
log.warn("[TaskTrackerActor] this worker is overload,ignore this request(instanceId={})! current size = {},", instanceId, HeavyTaskTrackerManager.currentTaskTrackerSize());
return;
}
// 原子创建,防止多实例的存在
//在重量级本地缓存中添加TaskTracker
HeavyTaskTrackerManager.atomicCreateTaskTracker(instanceId, ignore -> HeavyTaskTracker.create(req, workerRuntime));
}
}
由上可以看到,派发任务就是往TaskTracker里添加一条本地缓存。
这里出现了TaskTracker,TaskTracker、ProcessorTracker和Processor的概念直接参考PowerJob作者写过的文章:
这里我们继续分析下上面第51行代码处的HeavyTaskTracker.create方法:
/**
* HeavyTaskTracker:
* 静态方法创建 TaskTracker
*
* @param req 服务端调度任务请求
* @return API/CRON -> CommonTaskTracker, FIX_RATE/FIX_DELAY -> FrequentTaskTracker
*/
public static HeavyTaskTracker create(ServerScheduleJobReq req, WorkerRuntime workerRuntime) {
try {
TimeExpressionType timeExpressionType = TimeExpressionType.valueOf(req.getTimeExpressionType());
switch (timeExpressionType) {
case FIXED_RATE:
case FIXED_DELAY:
return new FrequentTaskTracker(req, workerRuntime);
default:
//这里我们分析下CommonTaskTracker构造器的实现
return new CommonTaskTracker(req, workerRuntime);
}
} catch (Exception e) {
reportCreateErrorToServer(req, workerRuntime, e);
}
return null;
}
/**
* CommonTaskTracker:
* 第17行代码处:
*/
protected CommonTaskTracker(ServerScheduleJobReq req, WorkerRuntime workerRuntime) {
super(req, workerRuntime);
}
/**
* HeavyTaskTracker:
*/
protected HeavyTaskTracker(ServerScheduleJobReq req, WorkerRuntime workerRuntime) {
// 初始化成员变量
super(req, workerRuntime);
// 赋予时间表达式类型
instanceInfo.setTimeExpressionType(TimeExpressionType.valueOf(req.getTimeExpressionType()).getV());
// 保护性操作
instanceInfo.setThreadConcurrency(Math.max(1, instanceInfo.getThreadConcurrency()));
this.ptStatusHolder = new ProcessorTrackerStatusHolder(instanceId, req.getMaxWorkerCount(), req.getAllWorkerAddress());
this.taskPersistenceService = workerRuntime.getTaskPersistenceService();
// 构建缓存
taskId2BriefInfo = CacheBuilder.newBuilder().maximumSize(1024).build();
// 构建分段锁
//SegmentLock是自己实现的ReentrantLock数组(使用数组是为了提高并发度)
segmentLock = new SegmentLock(UPDATE_CONCURRENCY);
// 子类自定义初始化操作
initTaskTracker(req);
log.info("[TaskTracker-{}] create TaskTracker successfully.", instanceId);
}
/**
* CommonTaskTracker:
* 第53行代码处:
*
* @param req 服务器调度任务实例运行请求
*/
@Override
protected void initTaskTracker(ServerScheduleJobReq req) {
// CommonTaskTrackerTimingPool 缩写
String poolName = String.format("ctttp-%d", req.getInstanceId()) + "-%d";
ThreadFactory factory = new ThreadFactoryBuilder().setNameFormat(poolName).build();
this.scheduledPool = Executors.newScheduledThreadPool(2, factory);
// 持久化根任务
persistenceRootTask();
// 开启定时状态检查
int delay = Integer.parseInt(System.getProperty(PowerJobDKey.WORKER_STATUS_CHECK_PERIOD, "13"));
scheduledPool.scheduleWithFixedDelay(new StatusCheckRunnable(), 3, delay, TimeUnit.SECONDS);
// 如果是 MR 任务,则需要启动执行器动态检测装置
ExecuteType executeType = ExecuteType.valueOf(req.getExecuteType());
if (executeType == ExecuteType.MAP || executeType == ExecuteType.MAP_REDUCE) {
scheduledPool.scheduleAtFixedRate(new WorkerDetector(), 1, 1, TimeUnit.MINUTES);
}
// 最后启动任务派发器,否则会出现 TaskTracker 还未创建完毕 ProcessorTracker 已开始汇报状态的情况
scheduledPool.scheduleWithFixedDelay(new Dispatcher(), 10, 5000, TimeUnit.MILLISECONDS);
}
/**
* 第73行代码处:
* 持久化根任务,只有完成持久化才能视为任务开始running(先持久化,再报告server)
*/
private void persistenceRootTask() {
TaskDO rootTask = new TaskDO();
rootTask.setStatus(TaskStatus.WAITING_DISPATCH.getValue());
rootTask.setInstanceId(instanceInfo.getInstanceId());
rootTask.setTaskId(ROOT_TASK_ID);
rootTask.setFailedCnt(0);
rootTask.setAddress(workerRuntime.getWorkerAddress());
//这里需要留意下,根任务的名称为OMS_ROOT_TASK(ROOT_TASK_NAME)
rootTask.setTaskName(TaskConstant.ROOT_TASK_NAME);
rootTask.setCreatedTime(System.currentTimeMillis());
rootTask.setLastModifiedTime(System.currentTimeMillis());
rootTask.setLastReportTime(-1L);
rootTask.setSubInstanceId(instanceId);
if (taskPersistenceService.save(rootTask)) {
log.info("[TaskTracker-{}] create root task successfully.", instanceId);
} else {
log.error("[TaskTracker-{}] create root task failed.", instanceId);
throw new PowerJobException("create root task failed for instance: " + instanceId);
}
}
在上面第77行代码处、第82行代码处和第86行代码处的initTaskTracker方法中,分别开启了三个定时任务,这里我们主要分析下StatusCheckRunnable和Dispatcher(WorkerDetector放在《较真儿学源码系列-PowerJob MapReduce源码分析》中分析):
/**
* StatusCheckRunnable:
*/
@Override
public void run() {
try {
innerRun();
} catch (Exception e) {
log.warn("[TaskTracker-{}] status checker execute failed, please fix the bug (@tjq)!", instanceId, e);
}
}
/**
* 第7行代码处:
*/
@SuppressWarnings("squid:S3776")
private void innerRun() {
//获取任务实例产生的各个Task状态,用于分析任务实例执行情况
InstanceStatisticsHolder holder = getInstanceStatisticsHolder(instanceId);
long finishedNum = holder.succeedNum + holder.failedNum;
long unfinishedNum = holder.waitingDispatchNum + holder.workerUnreceivedNum + holder.receivedNum + holder.runningNum;
log.debug("[TaskTracker-{}] status check result: {}", instanceId, holder);
//组装上报参数
TaskTrackerReportInstanceStatusReq req = new TaskTrackerReportInstanceStatusReq();
req.setAppId(workerRuntime.getAppId());
req.setJobId(instanceInfo.getJobId());
req.setInstanceId(instanceId);
req.setWfInstanceId(instanceInfo.getWfInstanceId());
req.setTotalTaskNum(finishedNum + unfinishedNum);
req.setSucceedTaskNum(holder.succeedNum);
req.setFailedTaskNum(holder.failedNum);
req.setReportTime(System.currentTimeMillis());
req.setStartTime(createTime);
req.setSourceAddress(workerRuntime.getWorkerAddress());
boolean success = false;
String result = null;
// 2. 如果未完成任务数为0,判断是否真正结束,并获取真正结束任务的执行结果
if (unfinishedNum == 0) {
// 数据库中一个任务都没有,说明根任务创建失败,该任务实例失败
if (finishedNum == 0) {
finished.set(true);
result = SystemInstanceResult.TASK_INIT_FAILED;
} else {
ExecuteType executeType = ExecuteType.valueOf(instanceInfo.getExecuteType());
switch (executeType) {
// STANDALONE 只有一个任务,完成即结束
case STANDALONE:
finished.set(true);
List allTask = taskPersistenceService.getAllTask(instanceId, instanceId);
if (CollectionUtils.isEmpty(allTask) || allTask.size() > 1) {
result = SystemInstanceResult.UNKNOWN_BUG;
log.warn("[TaskTracker-{}] there must have some bug in TaskTracker.", instanceId);
} else {
result = allTask.get(0).getResult();
success = allTask.get(0).getStatus() == TaskStatus.WORKER_PROCESS_SUCCESS.getValue();
}
break;
case MAP:
//...
break;
default:
//...
}
}
}
// 3. 检查任务实例整体是否超时
if (isTimeout()) {
finished.set(true);
success = false;
result = SystemInstanceResult.INSTANCE_EXECUTE_TIMEOUT;
}
// 4. 执行完毕,报告服务器
if (finished.get()) {
req.setResult(result);
// 上报追加的工作流上下文信息
req.setAppendedWfContext(appendedWfContext);
req.setInstanceStatus(success ? InstanceStatus.SUCCEED.getV() : InstanceStatus.FAILED.getV());
reportFinalStatusThenDestroy(workerRuntime, req);
return;
}
// 5. 未完成,上报状态
req.setInstanceStatus(InstanceStatus.RUNNING.getV());
TransportUtils.ttReportInstanceStatus(req, workerRuntime.getServerDiscoveryService().getCurrentServerAddress(), workerRuntime.getTransporter());
// 6.1 定期检查 -> 重试派发后未确认的任务
long currentMS = System.currentTimeMillis();
if (holder.workerUnreceivedNum != 0) {
taskPersistenceService.getTaskByStatus(instanceId, TaskStatus.DISPATCH_SUCCESS_WORKER_UNCHECK, 100).forEach(uncheckTask -> {
long elapsedTime = currentMS - uncheckTask.getLastModifiedTime();
if (elapsedTime > DISPATCH_TIME_OUT_MS) {
TaskDO updateEntity = new TaskDO();
updateEntity.setStatus(TaskStatus.WAITING_DISPATCH.getValue());
// 特殊任务只能本机执行
if (!TaskConstant.LAST_TASK_NAME.equals(uncheckTask.getTaskName())) {
updateEntity.setAddress(RemoteConstant.EMPTY_ADDRESS);
}
// 失败次数 + 1
updateEntity.setFailedCnt(uncheckTask.getFailedCnt() + 1);
taskPersistenceService.updateTask(instanceId, uncheckTask.getTaskId(), updateEntity);
log.warn("[TaskTracker-{}] task(id={},name={}) try to dispatch again due to unreceived the response from ProcessorTracker.",
instanceId, uncheckTask.getTaskId(), uncheckTask.getTaskName());
}
});
}
// 6.2 定期检查 -> 重新执行被派发到宕机ProcessorTracker上的任务
List disconnectedPTs = ptStatusHolder.getAllDisconnectedProcessorTrackers();
if (!disconnectedPTs.isEmpty()) {
log.warn("[TaskTracker-{}] some ProcessorTracker disconnected from TaskTracker,their address is {}.", instanceId, disconnectedPTs);
if (taskPersistenceService.updateLostTasks(instanceId, disconnectedPTs, true)) {
ptStatusHolder.remove(disconnectedPTs);
log.warn("[TaskTracker-{}] removed these ProcessorTracker from StatusHolder: {}", instanceId, disconnectedPTs);
}
}
}
上报到服务端的逻辑这里就不再看了(更新instance_info实例表数据)。整体上来说,StatusCheckRunnable的作用就是检查任务的执行情况,并上报到服务端。
/**
* Dispatcher:
*/
@Override
public void run() {
if (finished.get()) {
return;
}
Stopwatch stopwatch = Stopwatch.createStarted();
// 1. 获取可以派发任务的 ProcessorTracker
List availablePtIps = ptStatusHolder.getAvailableProcessorTrackers();
// 2. 没有可用 ProcessorTracker,本次不派发
if (availablePtIps.isEmpty()) {
log.debug("[TaskTracker-{}] no available ProcessorTracker now.", instanceId);
return;
}
// 3. 避免大查询,分批派发任务
long currentDispatchNum = 0;
//这里需要留意下,最大分发次数=可用ProcessorTracker数量*实例并发度*2
long maxDispatchNum = availablePtIps.size() * instanceInfo.getThreadConcurrency() * 2L;
AtomicInteger index = new AtomicInteger(0);
// 4. 循环查询数据库,获取需要派发的任务
while (maxDispatchNum > currentDispatchNum) {
//每次最多查询100个任务
int dbQueryLimit = Math.min(DB_QUERY_LIMIT, (int) maxDispatchNum);
//获取等待调度器调度的任务
List needDispatchTasks = taskPersistenceService.getTaskByStatus(instanceId, TaskStatus.WAITING_DISPATCH, dbQueryLimit);
currentDispatchNum += needDispatchTasks.size();
needDispatchTasks.forEach(task -> {
// 获取 ProcessorTracker 地址,如果 Task 中自带了 Address,则使用该 Address
String ptAddress = task.getAddress();
if (StringUtils.isEmpty(ptAddress) || RemoteConstant.EMPTY_ADDRESS.equals(ptAddress)) {
//否则,从可用ProcessorTracker中取余获取一个
ptAddress = availablePtIps.get(index.getAndIncrement() % availablePtIps.size());
}
//分发任务
dispatchTask(task, ptAddress);
});
// 数量不足 或 查询失败,则终止循环
if (needDispatchTasks.size() < dbQueryLimit) {
break;
}
}
log.debug("[TaskTracker-{}] dispatched {} tasks,using time {}.", instanceId, currentDispatchNum, stopwatch.stop());
}
/**
* ProcessorTrackerStatusHolder:
* 第14行代码处:
* 获取可用 ProcessorTracker 的IP地址
*/
public List getAvailableProcessorTrackers() {
List result = Lists.newLinkedList();
address2Status.forEach((address, ptStatus) -> {
if (ptStatus.available()) {
result.add(address);
}
});
return result;
}
/**
* ProcessorTrackerStatus:
* 第66行代码处:
* 是否可用
*/
public boolean available() {
// 未曾派发过,默认可用
if (!dispatched) {
return true;
}
// 已派发但未收到响应,则不可用
if (!connected) {
return false;
}
// 长时间未收到心跳消息,则不可用
if (isTimeout()) {
return false;
}
// 留有过多待处理任务,则不可用
if (remainTaskNum >= DISPATCH_THRESHOLD) {
return false;
}
// TODO:后续考虑加上机器健康度等信息
return true;
}
/**
* 第91行代码处:
* 是否超时(超过一定时间没有收到心跳)
*/
public boolean isTimeout() {
if (dispatched) {
//系统当前时间-上次活跃时间>心跳超时时间
return System.currentTimeMillis() - lastActiveTime > HEARTBEAT_TIMEOUT_MS;
}
// 未曾派发过任务的机器,不用处理
return false;
}
/**
* HeavyTaskTracker:
* 第45行代码处:
* 派发任务到 ProcessorTracker
*
* @param task 需要被执行的任务
* @param processorTrackerAddress ProcessorTracker的地址(IP:Port)
*/
protected void dispatchTask(TaskDO task, String processorTrackerAddress) {
// 1. 持久化,更新数据库(如果更新数据库失败,可能导致重复执行,先不处理)
TaskDO updateEntity = new TaskDO();
updateEntity.setStatus(TaskStatus.DISPATCH_SUCCESS_WORKER_UNCHECK.getValue());
// 写入处理该任务的 ProcessorTracker
updateEntity.setAddress(processorTrackerAddress);
boolean success = taskPersistenceService.updateTask(instanceId, task.getTaskId(), updateEntity);
if (!success) {
log.warn("[TaskTracker-{}] dispatch task(taskId={},taskName={}) failed due to update task status failed.", instanceId, task.getTaskId(), task.getTaskName());
return;
}
// 2. 更新 ProcessorTrackerStatus 状态
ptStatusHolder.getProcessorTrackerStatus(processorTrackerAddress).setDispatched(true);
// 3. 初始化缓存
taskId2BriefInfo.put(task.getTaskId(), new TaskBriefInfo(task.getTaskId(), TaskStatus.DISPATCH_SUCCESS_WORKER_UNCHECK, -1L));
// 4. 任务派发
TaskTrackerStartTaskReq startTaskReq = new TaskTrackerStartTaskReq(instanceInfo, task, workerRuntime.getWorkerAddress());
TransportUtils.ttStartPtTask(startTaskReq, processorTrackerAddress, workerRuntime.getTransporter());
log.debug("[TaskTracker-{}] dispatch task(taskId={},taskName={}) successfully.", instanceId, task.getTaskId(), task.getTaskName());
}
/**
* TransportUtils:
* 第146行代码处:
*/
public static void ttStartPtTask(TaskTrackerStartTaskReq req, String address, Transporter transporter) {
final URL url = easyBuildUrl(ServerType.WORKER, WPT_PATH, WPT_HANDLER_START_TASK, address);
transporter.tell(url, req);
}
/**
* ProcessorTrackerActor:
* 第157行代码处:
* 处理来自TaskTracker的task执行请求
*
* @param req 请求
*/
@Handler(path = RemoteConstant.WPT_HANDLER_START_TASK, processType = ProcessType.NO_BLOCKING)
public void onReceiveTaskTrackerStartTaskReq(TaskTrackerStartTaskReq req) {
Long instanceId = req.getInstanceInfo().getInstanceId();
// 创建 ProcessorTracker 一定能成功
ProcessorTracker processorTracker = ProcessorTrackerManager.getProcessorTracker(
instanceId,
req.getTaskTrackerAddress(),
() -> new ProcessorTracker(req, workerRuntime));
TaskDO task = new TaskDO();
task.setTaskId(req.getTaskId());
task.setTaskName(req.getTaskName());
task.setTaskContent(req.getTaskContent());
task.setFailedCnt(req.getTaskCurrentRetryNums());
task.setSubInstanceId(req.getSubInstanceId());
processorTracker.submitTask(task);
}
/**
* ProcessorTracker:
* 提交任务到线程池执行
* 1.0版本:TaskTracker有任务就dispatch,导致 ProcessorTracker 本地可能堆积过多的任务,造成内存压力。为此 ProcessorTracker 在线程
* 池队列堆积到一定程度时,会将数据持久化到DB,然后通过异步线程定时从数据库中取回任务,重新提交执行。
* 联动:数据库的SPID设计、TaskStatus段落设计等,全部取消...
* last commitId: 341953aceceafec0fbe7c3d9a3e26451656b945e
* 2.0版本:ProcessorTracker定时向TaskTracker发送心跳消息,心跳消息中包含了当前线程池队列任务个数,TaskTracker根据ProcessorTracker
* 的状态判断能否继续派发任务。因此,ProcessorTracker本地不会堆积过多任务,故删除 持久化机制 ╥﹏╥...!
*
* @param newTask 需要提交到线程池执行的任务
*/
public void submitTask(TaskDO newTask) {
// 一旦 ProcessorTracker 出现异常,所有提交到此处的任务直接返回失败,防止形成死锁
// 死锁分析:TT创建PT,PT创建失败,无法定期汇报心跳,TT长时间未收到PT心跳,认为PT宕机(确实宕机了),无法选择可用的PT再次派发任务,死锁形成,GG斯密达 T_T
if (lethal) {
ProcessorReportTaskStatusReq report = new ProcessorReportTaskStatusReq()
.setInstanceId(instanceId)
.setSubInstanceId(newTask.getSubInstanceId())
.setTaskId(newTask.getTaskId())
.setStatus(TaskStatus.WORKER_PROCESS_FAILED.getValue())
.setResult(lethalReason)
.setReportTime(System.currentTimeMillis());
TransportUtils.ptReportTask(report, taskTrackerAddress, workerRuntime);
return;
}
boolean success = false;
// 1. 设置值并提交执行
newTask.setInstanceId(instanceInfo.getInstanceId());
newTask.setAddress(taskTrackerAddress);
HeavyProcessorRunnable heavyProcessorRunnable = new HeavyProcessorRunnable(instanceInfo, taskTrackerAddress, newTask, processorBean, omsLogger, statusReportRetryQueue, workerRuntime);
try {
threadPool.submit(heavyProcessorRunnable);
success = true;
} catch (RejectedExecutionException ignore) {
log.warn("[ProcessorTracker-{}] submit task(taskId={},taskName={}) to ThreadPool failed due to ThreadPool has too much task waiting to process, this task will dispatch to other ProcessorTracker.",
instanceId, newTask.getTaskId(), newTask.getTaskName());
} catch (Exception e) {
log.error("[ProcessorTracker-{}] submit task(taskId={},taskName={}) to ThreadPool failed.", instanceId, newTask.getTaskId(), newTask.getTaskName(), e);
}
// 2. 回复接收成功
if (success) {
ProcessorReportTaskStatusReq reportReq = new ProcessorReportTaskStatusReq();
reportReq.setInstanceId(instanceId);
reportReq.setSubInstanceId(newTask.getSubInstanceId());
reportReq.setTaskId(newTask.getTaskId());
reportReq.setStatus(TaskStatus.WORKER_RECEIVED.getValue());
reportReq.setReportTime(System.currentTimeMillis());
TransportUtils.ptReportTask(reportReq, taskTrackerAddress, workerRuntime);
log.debug("[ProcessorTracker-{}] submit task(taskId={}, taskName={}) success, current queue size: {}.",
instanceId, newTask.getTaskId(), newTask.getTaskName(), threadPool.getQueue().size());
}
}
可以看到,在最后,Dispatcher会创建出一个HeavyProcessorRunnable的线程来执行,里面存放着需要执行的任务实例、执行地址、执行处理器等信息。接下来看下其实现:
/**
* HeavyProcessorRunnable:
*/
@Override
@SuppressWarnings("squid:S2142")
public void run() {
// 切换线程上下文类加载器(否则用的是 Worker 类加载器,不存在容器类,在序列化/反序列化时会报 ClassNotFoundException)
Thread.currentThread().setContextClassLoader(processorBean.getClassLoader());
try {
innerRun();
} catch (InterruptedException ignore) {
// ignore
} catch (Throwable e) {
reportStatus(TaskStatus.WORKER_PROCESS_FAILED, e.toString(), null, null);
log.error("[ProcessorRunnable-{}] execute failed, please contact the author(@KFCFans) to fix the bug!", task.getInstanceId(), e);
} finally {
ThreadLocalStore.clear();
}
}
/**
* 第10行代码处:
*/
public void innerRun() throws InterruptedException {
//获取执行处理器
final BasicProcessor processor = processorBean.getProcessor();
String taskId = task.getTaskId();
Long instanceId = task.getInstanceId();
log.debug("[ProcessorRunnable-{}] start to run task(taskId={}&taskName={})", instanceId, taskId, task.getTaskName());
//缓存
ThreadLocalStore.setTask(task);
ThreadLocalStore.setRuntimeMeta(workerRuntime);
// 0. 构造任务上下文
WorkflowContext workflowContext = constructWorkflowContext();
TaskContext taskContext = constructTaskContext();
taskContext.setWorkflowContext(workflowContext);
// 1. 上报执行信息
reportStatus(TaskStatus.WORKER_PROCESSING, null, null, null);
ProcessResult processResult;
ExecuteType executeType = ExecuteType.valueOf(instanceInfo.getExecuteType());
// 2. 根任务 & 广播执行 特殊处理
if (TaskConstant.ROOT_TASK_NAME.equals(task.getTaskName()) && executeType == ExecuteType.BROADCAST) {
// 广播执行:先选本机执行 preProcess,完成后 TaskTracker 再为所有 Worker 生成子 Task
handleBroadcastRootTask(instanceId, taskContext);
return;
}
// 3. 最终任务特殊处理(一定和 TaskTracker 处于相同的机器)
if (TaskConstant.LAST_TASK_NAME.equals(task.getTaskName())) {
handleLastTask(taskId, instanceId, taskContext, executeType);
return;
}
// 4. 正式提交运行
try {
//这里的process方法也就是我们自己写的业务方法
processResult = processor.process(taskContext);
if (processResult == null) {
processResult = new ProcessResult(false, "ProcessResult can't be null");
}
} catch (Throwable e) {
log.warn("[ProcessorRunnable-{}] task(id={},name={}) process failed.", instanceId, taskContext.getTaskId(), taskContext.getTaskName(), e);
processResult = new ProcessResult(false, e.toString());
}
//上报执行结果
reportStatus(processResult.isSuccess() ? TaskStatus.WORKER_PROCESS_SUCCESS : TaskStatus.WORKER_PROCESS_FAILED, suit(processResult.getMsg()), null, workflowContext.getAppendedContextData());
}
Dispatcher的作用就是分发任务,并执行任务。也就是会有一个单独的线程(HeavyProcessorRunnable),会定时从任务表中拿取任务,然后执行我们自己实现的process方法,并会在执行前和执行后上报执行信息。
/**
* InstanceStatusCheckService:
* 检查运行中的实例
* RUNNING 超时:TaskTracker down,断开与 server 的心跳连接
*/
public void checkRunningInstance() {
Stopwatch stopwatch = Stopwatch.createStarted();
// 查询 DB 获取该 Server 需要负责的 AppGroup
//获取appId
List allAppIds = appInfoRepository.listAppIdByCurrentServer(transportService.defaultProtocol().getAddress());
if (CollectionUtils.isEmpty(allAppIds)) {
log.info("[InstanceStatusChecker] current server has no app's job to check");
return;
}
try {
// 检查 RUNNING 状态的任务(一定时间没收到 TaskTracker 的状态报告,视为失败)
Lists.partition(allAppIds, MAX_BATCH_NUM_APP).forEach(this::handleRunningInstance);
} catch (Exception e) {
log.error("[InstanceStatusChecker] RunningInstance status check failed.", e);
}
log.info("[InstanceStatusChecker] RunningInstance status check using {}.", stopwatch.stop());
}
/**
* 第17行代码处:
*/
private void handleRunningInstance(List partAppIds) {
// 3. 检查 RUNNING 状态的任务(一定时间没收到 TaskTracker 的状态报告,视为失败)
long threshold = System.currentTimeMillis() - RUNNING_TIMEOUT_MS;
//查找修改时间距离现在超过1分钟的实例
List failedInstances = instanceInfoRepository.selectBriefInfoByAppIdInAndStatusAndGmtModifiedBefore(partAppIds, InstanceStatus.RUNNING.getV(), new Date(threshold), PageRequest.of(0, MAX_BATCH_NUM_INSTANCE));
while (!failedInstances.isEmpty()) {
// collect job id
Set jobIds = failedInstances.stream().map(BriefInstanceInfo::getJobId).collect(Collectors.toSet());
// query job info and map
Map jobInfoMap = jobInfoRepository.findByIdIn(jobIds).stream().collect(Collectors.toMap(JobInfoDO::getId, e -> e));
log.warn("[InstanceStatusCheckService] find some instances have not received status report for a long time : {}", failedInstances.stream().map(BriefInstanceInfo::getInstanceId).collect(Collectors.toList()));
failedInstances.forEach(instance -> {
Optional jobInfoOpt = Optional.ofNullable(jobInfoMap.get(instance.getJobId()));
if (!jobInfoOpt.isPresent()) {
final Optional opt = instanceInfoRepository.findById(instance.getId());
opt.ifPresent(e -> updateFailedInstance(e, SystemInstanceResult.REPORT_TIMEOUT));
return;
}
TimeExpressionType timeExpressionType = TimeExpressionType.of(jobInfoOpt.get().getTimeExpressionType());
SwitchableStatus switchableStatus = SwitchableStatus.of(jobInfoOpt.get().getStatus());
// 如果任务已关闭,则不进行重试,将任务置为失败即可;秒级任务也直接置为失败,由派发器重新调度
if (switchableStatus != SwitchableStatus.ENABLE || TimeExpressionType.FREQUENT_TYPES.contains(timeExpressionType.getV())) {
final Optional opt = instanceInfoRepository.findById(instance.getId());
opt.ifPresent(e -> updateFailedInstance(e, SystemInstanceResult.REPORT_TIMEOUT));
return;
}
// CRON 和 API一样,失败次数 + 1,根据重试配置进行重试
if (instance.getRunningTimes() < jobInfoOpt.get().getInstanceRetryNum()) {
//更新实例表状态为等待派发
dispatchService.redispatchAsync(instance.getInstanceId(), InstanceStatus.RUNNING.getV());
} else {
final Optional opt = instanceInfoRepository.findById(instance.getId());
opt.ifPresent(e -> updateFailedInstance(e, SystemInstanceResult.REPORT_TIMEOUT));
}
});
threshold = System.currentTimeMillis() - RUNNING_TIMEOUT_MS;
//重新查找修改时间距离现在超过1分钟的实例,并循环
failedInstances = instanceInfoRepository.selectBriefInfoByAppIdInAndStatusAndGmtModifiedBefore(partAppIds, InstanceStatus.RUNNING.getV(), new Date(threshold), PageRequest.of(0, MAX_BATCH_NUM_INSTANCE));
}
}
客户端的启动类同样没有什么逻辑,查看下初始化类PowerJobSpringWorker:
/**
* PowerJobSpringWorker:
*/
@Override
public void afterPropertiesSet() throws Exception {
powerJobWorker = new PowerJobWorker(config);
powerJobWorker.init();
}
/**
* PowerJobWorker:
*/
public void init() throws Exception {
if (!initialized.compareAndSet(false, true)) {
log.warn("[PowerJobWorker] please do not repeat the initialization");
return;
}
Stopwatch stopwatch = Stopwatch.createStarted();
log.info("[PowerJobWorker] start to initialize PowerJobWorker...");
PowerJobWorkerConfig config = workerRuntime.getWorkerConfig();
CommonUtils.requireNonNull(config, "can't find PowerJobWorkerConfig, please set PowerJobWorkerConfig first");
try {
//打印banner
PowerBannerPrinter.print();
// 校验 appName
if (!config.isEnableTestMode()) {
assertAppName();
} else {
log.warn("[PowerJobWorker] using TestMode now, it's dangerous if this is production env.");
}
// 初始化元数据
String workerAddress = NetUtils.getLocalHost() + ":" + config.getPort();
workerRuntime.setWorkerAddress(workerAddress);
// 初始化 线程池
final ExecutorManager executorManager = new ExecutorManager(workerRuntime.getWorkerConfig());
workerRuntime.setExecutorManager(executorManager);
// 初始化 ProcessorLoader
//动态加载类的工厂,可以不依赖于Spring的实现
ProcessorLoader processorLoader = buildProcessorLoader(workerRuntime);
workerRuntime.setProcessorLoader(processorLoader);
// 初始化 actor
TaskTrackerActor taskTrackerActor = new TaskTrackerActor(workerRuntime);
ProcessorTrackerActor processorTrackerActor = new ProcessorTrackerActor(workerRuntime);
WorkerActor workerActor = new WorkerActor(workerRuntime, taskTrackerActor);
// 初始化通讯引擎
EngineConfig engineConfig = new EngineConfig()
.setType(config.getProtocol().name())
.setServerType(ServerType.WORKER)
.setBindAddress(new Address().setHost(NetUtils.getLocalHost()).setPort(config.getPort()))
.setActorList(Lists.newArrayList(taskTrackerActor, processorTrackerActor, workerActor));
//之前服务端启动的时候也会执行PowerJobRemoteEngine.start方法
EngineOutput engineOutput = remoteEngine.start(engineConfig);
workerRuntime.setTransporter(engineOutput.getTransporter());
// 连接 server
ServerDiscoveryService serverDiscoveryService = new ServerDiscoveryService(workerRuntime.getAppId(), workerRuntime.getWorkerConfig());
serverDiscoveryService.start(workerRuntime.getExecutorManager().getCoreExecutor());
workerRuntime.setServerDiscoveryService(serverDiscoveryService);
log.info("[PowerJobWorker] PowerJobRemoteEngine initialized successfully.");
// 初始化日志系统
OmsLogHandler omsLogHandler = new OmsLogHandler(workerAddress, workerRuntime.getTransporter(), serverDiscoveryService);
workerRuntime.setOmsLogHandler(omsLogHandler);
// 初始化存储
TaskPersistenceService taskPersistenceService = new TaskPersistenceService(workerRuntime.getWorkerConfig().getStoreStrategy());
taskPersistenceService.init();
workerRuntime.setTaskPersistenceService(taskPersistenceService);
log.info("[PowerJobWorker] local storage initialized successfully.");
// 初始化定时任务
workerRuntime.getExecutorManager().getCoreExecutor().scheduleAtFixedRate(new WorkerHealthReporter(workerRuntime), 0, config.getHealthReportInterval(), TimeUnit.SECONDS);
workerRuntime.getExecutorManager().getCoreExecutor().scheduleWithFixedDelay(omsLogHandler.logSubmitter, 0, 5, TimeUnit.SECONDS);
log.info("[PowerJobWorker] PowerJobWorker initialized successfully, using time: {}, congratulations!", stopwatch);
} catch (Exception e) {
log.error("[PowerJobWorker] initialize PowerJobWorker failed, using {}.", stopwatch, e);
throw e;
}
}
/**
* 第31行代码处:
*/
@SuppressWarnings("rawtypes")
private void assertAppName() {
PowerJobWorkerConfig config = workerRuntime.getWorkerConfig();
String appName = config.getAppName();
Objects.requireNonNull(appName, "appName can't be empty!");
String url = "http://%s/server/assert?appName=%s";
for (String server : config.getServerAddress()) {
String realUrl = String.format(url, server, appName);
try {
//客户端启动的时候连接一下服务端,看是否能连通
String resultDTOStr = CommonUtils.executeWithRetry0(() -> HttpUtils.get(realUrl));
ResultDTO resultDTO = JsonUtils.parseObject(resultDTOStr, ResultDTO.class);
if (resultDTO.isSuccess()) {
Long appId = Long.valueOf(resultDTO.getData().toString());
log.info("[PowerJobWorker] assert appName({}) succeed, the appId for this application is {}.", appName, appId);
workerRuntime.setAppId(appId);
return;
} else {
log.error("[PowerJobWorker] assert appName failed, this appName is invalid, please register the appName {} first.", appName);
throw new PowerJobException(resultDTO.getMessage());
}
} catch (PowerJobException oe) {
throw oe;
} catch (Exception ignore) {
log.warn("[PowerJobWorker] assert appName by url({}) failed, please check the server address.", realUrl);
}
}
log.error("[PowerJobWorker] no available server in {}.", config.getServerAddress());
throw new PowerJobException("no server available!");
}
/**
* ExecutorManager:
* 第41行代码处:
*/
public ExecutorManager(PowerJobWorkerConfig workerConfig) {
//可用cpu数
final int availableProcessors = Runtime.getRuntime().availableProcessors();
// 初始化定时线程池
ThreadFactory coreThreadFactory = new ThreadFactoryBuilder().setNameFormat("powerjob-worker-core-%d").build();
coreExecutor = new ScheduledThreadPoolExecutor(3, coreThreadFactory);
ThreadFactory lightTaskReportFactory = new ThreadFactoryBuilder().setNameFormat("powerjob-worker-light-task-status-check-%d").build();
// 都是 io 密集型任务
lightweightTaskStatusCheckExecutor = new ScheduledThreadPoolExecutor(availableProcessors * 10, lightTaskReportFactory);
ThreadFactory lightTaskExecuteFactory = new ThreadFactoryBuilder().setNameFormat("powerjob-worker-light-task-execute-%d").build();
// 大部分任务都是 io 密集型
lightweightTaskExecutorService = new ThreadPoolExecutor(availableProcessors * 10, availableProcessors * 10, 120L, TimeUnit.SECONDS,
new ArrayBlockingQueue<>((workerConfig.getMaxLightweightTaskNum() * 2), true), lightTaskExecuteFactory, new ThreadPoolExecutor.AbortPolicy());
}
/**
* ServerDiscoveryService:
* 第68行代码处:
*/
public void start(ScheduledExecutorService timingPool) {
this.currentServerAddress = discovery();
if (StringUtils.isEmpty(this.currentServerAddress) && !config.isEnableTestMode()) {
throw new PowerJobException("can't find any available server, this worker has been quarantined.");
}
// 这里必须保证成功
timingPool.scheduleAtFixedRate(() -> {
try {
this.currentServerAddress = discovery();
} catch (Exception e) {
log.error("[PowerDiscovery] fail to discovery server!", e);
}
}
, 10, 10, TimeUnit.SECONDS);
}
/**
* 第159行代码处和第166行代码处:
*/
private String discovery() {
if (ip2Address.isEmpty()) {
config.getServerAddress().forEach(x -> ip2Address.put(x.split(":")[0], x));
}
String result = null;
// 先对当前机器发起请求
String currentServer = currentServerAddress;
if (!StringUtils.isEmpty(currentServer)) {
String ip = currentServer.split(":")[0];
// 直接请求当前Server的HTTP服务,可以少一次网络开销,减轻Server负担
String firstServerAddress = ip2Address.get(ip);
if (firstServerAddress != null) {
result = acquire(firstServerAddress);
}
}
for (String httpServerAddress : config.getServerAddress()) {
if (StringUtils.isEmpty(result)) {
result = acquire(httpServerAddress);
} else {
break;
}
}
if (StringUtils.isEmpty(result)) {
log.warn("[PowerDiscovery] can't find any available server, this worker has been quarantined.");
// 在 Server 高可用的前提下,连续失败多次,说明该节点与外界失联,Server已经将秒级任务转移到其他Worker,需要杀死本地的任务
if (FAILED_COUNT++ > MAX_FAILED_COUNT) {
log.warn("[PowerDiscovery] can't find any available server for 3 consecutive times, It's time to kill all frequent job in this worker.");
List frequentInstanceIds = HeavyTaskTrackerManager.getAllFrequentTaskTrackerKeys();
if (!CollectionUtils.isEmpty(frequentInstanceIds)) {
frequentInstanceIds.forEach(instanceId -> {
HeavyTaskTracker taskTracker = HeavyTaskTrackerManager.removeTaskTracker(instanceId);
taskTracker.destroy();
log.warn("[PowerDiscovery] kill frequent instance(instanceId={}) due to can't find any available server.", instanceId);
});
}
FAILED_COUNT = 0;
}
return null;
} else {
// 重置失败次数
FAILED_COUNT = 0;
log.debug("[PowerDiscovery] current server is {}.", result);
return result;
}
}
/**
* 第192行代码处和第198行代码处
*/
@SuppressWarnings("rawtypes")
private String acquire(String httpServerAddress) {
String result = null;
//构建url参数
String url = buildServerDiscoveryUrl(httpServerAddress);
try {
//请求服务端ServerController.acquireServer方法
result = CommonUtils.executeWithRetry0(() -> HttpUtils.get(url));
} catch (Exception ignore) {
}
if (!StringUtils.isEmpty(result)) {
try {
ResultDTO resultDTO = JsonUtils.parseObject(result, ResultDTO.class);
if (resultDTO.isSuccess()) {
return resultDTO.getData().toString();
}
} catch (Exception ignore) {
}
}
return null;
}
/**
* ServerController:
* 第241行代码处:
*/
@GetMapping("/acquire")
public ResultDTO acquireServer(ServerDiscoveryRequest request) {
return ResultDTO.success(serverElectionService.elect(request));
}
/**
* ServerElectionService:
*/
public String elect(ServerDiscoveryRequest request) {
if (!accurate()) {
final String currentServer = request.getCurrentServer();
// 如果是本机,就不需要查数据库那么复杂的操作了,直接返回成功
Optional localProtocolInfoOpt = Optional.ofNullable(transportService.allProtocols().get(request.getProtocol()));
//如果不是精确地请求,并且请求的参数中直接带有当前服务端的地址,则直接返回其地址即可
if (localProtocolInfoOpt.isPresent() && localProtocolInfoOpt.get().getAddress().equals(currentServer)) {
log.debug("[ServerElectionService] this server[{}] is worker's current server, skip check", currentServer);
return currentServer;
}
}
//上面的条件不满足,则走常规的选举流程
return getServer0(request);
}
/**
* 第269行代码处:
*/
private boolean accurate() {
//accurateSelectServerPercentage默认为50,这里是在随机判断是否是精确的请求
return ThreadLocalRandom.current().nextInt(100) < accurateSelectServerPercentage;
}
/**
* 第280行代码处:
*/
private String getServer0(ServerDiscoveryRequest discoveryRequest) {
final Long appId = discoveryRequest.getAppId();
final String protocol = discoveryRequest.getProtocol();
Set downServerCache = Sets.newHashSet();
for (int i = 0; i < RETRY_TIMES; i++) {
// 无锁获取当前数据库中的Server
Optional appInfoOpt = appInfoRepository.findById(appId);
if (!appInfoOpt.isPresent()) {
throw new PowerJobException(appId + " is not registered!");
}
String appName = appInfoOpt.get().getAppName();
String originServer = appInfoOpt.get().getCurrentServer();
String activeAddress = activeAddress(originServer, downServerCache, protocol);
if (StringUtils.isNotEmpty(activeAddress)) {
return activeAddress;
}
// 无可用Server,重新进行Server选举,需要加锁
String lockName = String.format(SERVER_ELECT_LOCK, appId);
//文件锁
boolean lockStatus = lockService.tryLock(lockName, 30000);
if (!lockStatus) {
try {
Thread.sleep(500);
} catch (Exception ignore) {
}
continue;
}
try {
// 可能上一台机器已经完成了Server选举,需要再次判断
//双重检查加锁
AppInfoDO appInfo = appInfoRepository.findById(appId).orElseThrow(() -> new RuntimeException("impossible, unless we just lost our database."));
String address = activeAddress(appInfo.getCurrentServer(), downServerCache, protocol);
if (StringUtils.isNotEmpty(address)) {
return address;
}
// 篡位,如果本机存在协议,则作为Server调度该 worker
//优先本机调度
final ProtocolInfo targetProtocolInfo = transportService.allProtocols().get(protocol);
if (targetProtocolInfo != null) {
// 注意,写入 AppInfoDO#currentServer 的永远是 default 的地址,仅在返回的时候特殊处理为协议地址
appInfo.setCurrentServer(transportService.defaultProtocol().getAddress());
appInfo.setGmtModified(new Date());
appInfoRepository.saveAndFlush(appInfo);
log.info("[ServerElection] this server({}) become the new server for app(appId={}).", appInfo.getCurrentServer(), appId);
return targetProtocolInfo.getAddress();
}
} catch (Exception e) {
log.error("[ServerElection] write new server to db failed for app {}.", appName, e);
} finally {
lockService.unlock(lockName);
}
}
throw new PowerJobException("server elect failed for app " + appId);
}
/**
* 第309行代码处:
* 判断指定server是否存活
*
* @param serverAddress 需要检测的server地址
* @param downServerCache 缓存,防止多次发送PING(这个QPS其实还蛮爆表的...)
* @param protocol 协议,用于返回指定的地址
* @return null or address
*/
private String activeAddress(String serverAddress, Set downServerCache, String protocol) {
if (downServerCache.contains(serverAddress)) {
return null;
}
if (StringUtils.isEmpty(serverAddress)) {
return null;
}
Ping ping = new Ping();
ping.setCurrentTime(System.currentTimeMillis());
//构建服务端url参数
URL targetUrl = ServerURLFactory.ping2Friend(serverAddress);
try {
//这里会跳转到FriendActor.onReceivePing方法
AskResponse response = transportService.ask(Protocol.HTTP.name(), targetUrl, ping, AskResponse.class)
.toCompletableFuture()
.get(PING_TIMEOUT_MS, TimeUnit.MILLISECONDS);
if (response.isSuccess()) {
// 检测通过的是远程 server 的暴露地址,需要返回 worker 需要的协议地址
final JSONObject protocolInfo = JsonUtils.parseObject(response.getData(), JSONObject.class).getJSONObject(protocol);
if (protocolInfo != null) {
downServerCache.remove(serverAddress);
final String protocolAddress = protocolInfo.toJavaObject(ProtocolInfo.class).getAddress();
log.info("[ServerElection] server[{}] is active, it will be the master, final protocol address={}", serverAddress, protocolAddress);
return protocolAddress;
} else {
log.warn("[ServerElection] server[{}] is active but don't have target protocol", serverAddress);
}
}
} catch (TimeoutException te) {
log.warn("[ServerElection] server[{}] was down due to ping timeout!", serverAddress);
} catch (Exception e) {
log.warn("[ServerElection] server[{}] was down with unknown case!", serverAddress, e);
}
downServerCache.add(serverAddress);
return null;
}
/**
* FriendActor:
* 第381行代码处:
* 处理存活检测的请求
*/
@Handler(path = S4S_HANDLER_PING, processType = ProcessType.NO_BLOCKING)
public AskResponse onReceivePing(Ping ping) {
return AskResponse.succeed(transportService.allProtocols());
}
/**
* PowerTransportService:
*/
@Override
public Map allProtocols() {
//这里就是从protocolName2Info缓存中取数据
return protocolName2Info;
}
由上可知,在客户端启动的时候,会通过选举的方式来选出一台服务器作为自己的server(也可能不选举,直接选择当前的服务端),并且会定时发送心跳数据来进行保活。
在上面第85行代码处和第86行代码处,客户端启动的时候也会有两个定时任务,查看其实现:
WorkerHealthReporter是用来给客户端健康度定时上报用的:
/**
* WorkerHealthReporter:
*/
@Override
public void run() {
// 没有可用Server,无法上报
String currentServer = workerRuntime.getServerDiscoveryService().getCurrentServerAddress();
if (StringUtils.isEmpty(currentServer)) {
log.warn("[WorkerHealthReporter] no available server,fail to report health info!");
return;
}
SystemMetrics systemMetrics;
if (workerRuntime.getWorkerConfig().getSystemMetricsCollector() == null) {
systemMetrics = SystemInfoUtils.getSystemMetrics();
} else {
systemMetrics = workerRuntime.getWorkerConfig().getSystemMetricsCollector().collect();
}
WorkerHeartbeat heartbeat = new WorkerHeartbeat();
heartbeat.setSystemMetrics(systemMetrics);
heartbeat.setWorkerAddress(workerRuntime.getWorkerAddress());
heartbeat.setAppName(workerRuntime.getWorkerConfig().getAppName());
heartbeat.setAppId(workerRuntime.getAppId());
heartbeat.setHeartbeatTime(System.currentTimeMillis());
heartbeat.setVersion(PowerJobWorkerVersion.getVersion());
heartbeat.setProtocol(workerRuntime.getWorkerConfig().getProtocol().name());
heartbeat.setClient("KingPenguin");
heartbeat.setTag(workerRuntime.getWorkerConfig().getTag());
// 上报 Tracker 数量
heartbeat.setLightTaskTrackerNum(LightTaskTrackerManager.currentTaskTrackerSize());
heartbeat.setHeavyTaskTrackerNum(HeavyTaskTrackerManager.currentTaskTrackerSize());
// 是否超载
if (workerRuntime.getWorkerConfig().getMaxLightweightTaskNum() <= LightTaskTrackerManager.currentTaskTrackerSize() || workerRuntime.getWorkerConfig().getMaxHeavyweightTaskNum() <= HeavyTaskTrackerManager.currentTaskTrackerSize()) {
heartbeat.setOverload(true);
}
// 获取当前加载的容器列表
heartbeat.setContainerInfos(OmsContainerFactory.getDeployedContainerInfos());
// 发送请求
if (StringUtils.isEmpty(currentServer)) {
return;
}
// log
log.info("[WorkerHealthReporter] report health status,appId:{},appName:{},isOverload:{},maxLightweightTaskNum:{},currentLightweightTaskNum:{},maxHeavyweightTaskNum:{},currentHeavyweightTaskNum:{}",
heartbeat.getAppId(),
heartbeat.getAppName(),
heartbeat.isOverload(),
workerRuntime.getWorkerConfig().getMaxLightweightTaskNum(),
heartbeat.getLightTaskTrackerNum(),
workerRuntime.getWorkerConfig().getMaxHeavyweightTaskNum(),
heartbeat.getHeavyTaskTrackerNum()
);
TransportUtils.reportWorkerHeartbeat(heartbeat, currentServer, workerRuntime.getTransporter());
}
/**
* SystemInfoUtils:
* 第17行代码处:
*/
public static SystemMetrics getSystemMetrics() {
SystemMetrics metrics = new SystemMetrics();
//赋值cpu指标
fillCPUInfo(metrics);
//赋值内存指标
fillMemoryInfo(metrics);
//赋值磁盘指标
fillDiskInfo(metrics);
// 在Worker完成分数计算,减小Server压力
metrics.calculateScore();
return metrics;
}
/**
* TransportUtils:
* 第58行代码处:
*/
public static void reportWorkerHeartbeat(WorkerHeartbeat req, String address, Transporter transporter) {
//绑定url调用信息
final URL url = easyBuildUrl(ServerType.SERVER, S4W_PATH, S4W_HANDLER_WORKER_HEARTBEAT, address);
transporter.tell(url, req);
}
其中SystemInfoUtils.getSystemMetrics方法是在获取系统的一些指标数据(调用的都是Java的底层Api,这里就不再详细查看了),并且计算健康度的得分(之前在第2.3.1小节中已经查看过该方法的实现了)。基于此,我们就知道了控制台首页的worker指标数据是怎么来的了:
在上面第88行代码处向服务端发送了心跳数据,接下来就来看下服务端是如何处理的:
/**
* AbWorkerRequestHandler:
*/
@Override
@Handler(path = S4W_HANDLER_WORKER_HEARTBEAT, processType = ProcessType.NO_BLOCKING)
public void processWorkerHeartbeat(WorkerHeartbeat heartbeat) {
long startMs = System.currentTimeMillis();
WorkerHeartbeatEvent event = new WorkerHeartbeatEvent()
.setAppName(heartbeat.getAppName())
.setAppId(heartbeat.getAppId())
.setVersion(heartbeat.getVersion())
.setProtocol(heartbeat.getProtocol())
.setTag(heartbeat.getTag())
.setWorkerAddress(heartbeat.getWorkerAddress())
.setDelayMs(startMs - heartbeat.getHeartbeatTime())
.setScore(heartbeat.getSystemMetrics().getScore());
processWorkerHeartbeat0(heartbeat, event);
//默认实现是写入日志进行监控
monitorService.monitor(event);
}
/**
* WorkerRequestHandlerImpl:
* 第17行代码处:
*/
@Override
protected void processWorkerHeartbeat0(WorkerHeartbeat heartbeat, WorkerHeartbeatEvent event) {
WorkerClusterManagerService.updateStatus(heartbeat);
}
/**
* WorkerClusterManagerService:
* 更新状态
*
* @param heartbeat Worker的心跳包
*/
public static void updateStatus(WorkerHeartbeat heartbeat) {
Long appId = heartbeat.getAppId();
String appName = heartbeat.getAppName();
ClusterStatusHolder clusterStatusHolder = APP_ID_2_CLUSTER_STATUS.computeIfAbsent(appId, ignore -> new ClusterStatusHolder(appName));
clusterStatusHolder.updateStatus(heartbeat);
}
/**
* ClusterStatusHolder:
* 更新 worker 机器的状态
*
* @param heartbeat 心跳请求
*/
public void updateStatus(WorkerHeartbeat heartbeat) {
String workerAddress = heartbeat.getWorkerAddress();
long heartbeatTime = heartbeat.getHeartbeatTime();
WorkerInfo workerInfo = address2WorkerInfo.computeIfAbsent(workerAddress, ignore -> {
WorkerInfo wf = new WorkerInfo();
wf.refresh(heartbeat);
return wf;
});
long oldTime = workerInfo.getLastActiveTime();
//过期心跳数据,不处理
if (heartbeatTime < oldTime) {
log.warn("[ClusterStatusHolder-{}] receive the expired heartbeat from {}, serverTime: {}, heartTime: {}", appName, heartbeat.getWorkerAddress(), System.currentTimeMillis(), heartbeat.getHeartbeatTime());
return;
}
workerInfo.refresh(heartbeat);
List containerInfos = heartbeat.getContainerInfos();
if (!CollectionUtils.isEmpty(containerInfos)) {
containerInfos.forEach(containerInfo -> {
Map infos = containerId2Infos.computeIfAbsent(containerInfo.getContainerId(), ignore -> Maps.newConcurrentMap());
infos.put(workerAddress, containerInfo);
});
}
}
/**
* WorkerInfo:
* 第57行代码处和第67行代码处:
* 刷新服务端记录的客户端的数据
*/
public void refresh(WorkerHeartbeat workerHeartbeat) {
address = workerHeartbeat.getWorkerAddress();
lastActiveTime = workerHeartbeat.getHeartbeatTime();
protocol = workerHeartbeat.getProtocol();
client = workerHeartbeat.getClient();
tag = workerHeartbeat.getTag();
systemMetrics = workerHeartbeat.getSystemMetrics();
containerInfos = workerHeartbeat.getContainerInfos();
lightTaskTrackerNum = workerHeartbeat.getLightTaskTrackerNum();
heavyTaskTrackerNum = workerHeartbeat.getHeavyTaskTrackerNum();
if (workerHeartbeat.isOverload()) {
overloading = true;
lastOverloadTime = workerHeartbeat.getHeartbeatTime();
log.warn("[WorkerInfo] worker {} is overload!", getAddress());
} else {
overloading = false;
}
}
/**
* LogSubmitter:
*/
@Override
public void run() {
boolean lockResult = reportLock.tryLock();
if (!lockResult) {
return;
}
try {
final String currentServerAddress = serverDiscoveryService.getCurrentServerAddress();
// 当前无可用 Server
if (StringUtils.isEmpty(currentServerAddress)) {
if (!logQueue.isEmpty()) {
logQueue.clear();
log.warn("[OmsLogHandler] because there is no available server to report logs which leads to queue accumulation, oms discarded all logs.");
}
return;
}
List logs = Lists.newLinkedList();
while (!logQueue.isEmpty()) {
try {
//从日志队列头部取数据
InstanceLogContent logContent = logQueue.poll(100, TimeUnit.MILLISECONDS);
logs.add(logContent);
//批处理
if (logs.size() >= BATCH_SIZE) {
WorkerLogReportReq req = new WorkerLogReportReq(workerAddress, Lists.newLinkedList(logs));
// 不可靠请求,WEB日志不追求极致
TransportUtils.reportLogs(req, currentServerAddress, transporter);
logs.clear();
}
} catch (Exception ignore) {
break;
}
}
if (!logs.isEmpty()) {
WorkerLogReportReq req = new WorkerLogReportReq(workerAddress, logs);
TransportUtils.reportLogs(req, currentServerAddress, transporter);
}
} finally {
reportLock.unlock();
}
}
/**
* TransportUtils:
* 第36行代码处和第47行代码处:
*/
public static void reportLogs(WorkerLogReportReq req, String address, Transporter transporter) {
final URL url = easyBuildUrl(ServerType.SERVER, S4W_PATH, S4W_HANDLER_REPORT_LOG, address);
transporter.tell(url, req);
}
上面第61行代码处会将请求发送到服务端的AbWorkerRequestHandler.processWorkerLogReport方法中:
/**
* AbWorkerRequestHandler:
*/
@Override
@Handler(path = S4W_HANDLER_REPORT_LOG, processType = ProcessType.NO_BLOCKING)
public void processWorkerLogReport(WorkerLogReportReq req) {
WorkerLogReportEvent event = new WorkerLogReportEvent()
.setWorkerAddress(req.getWorkerAddress())
.setLogNum(req.getInstanceLogContents().size());
try {
processWorkerLogReport0(req, event);
event.setStatus(WorkerLogReportEvent.Status.SUCCESS);
} catch (RejectedExecutionException re) {
event.setStatus(WorkerLogReportEvent.Status.REJECTED);
} catch (Throwable t) {
event.setStatus(WorkerLogReportEvent.Status.EXCEPTION);
log.warn("[WorkerRequestHandler] process worker report failed!", t);
} finally {
//日志监控
monitorService.monitor(event);
}
}
/**
* WorkerRequestHandlerImpl:
* 第12行代码处:
*/
@Override
protected void processWorkerLogReport0(WorkerLogReportReq req, WorkerLogReportEvent event) {
// 这个效率应该不会拉垮吧...也就是一些判断 + Map#get 吧...
instanceLogService.submitLogs(req.getWorkerAddress(), req.getInstanceLogContents());
}
/**
* InstanceLogService:
* 提交日志记录,持久化到本地数据库中
*
* @param workerAddress 上报机器地址
* @param logs 任务实例运行时日志
*/
@Async(value = PJThreadPool.LOCAL_DB_POOL)
public void submitLogs(String workerAddress, List logs) {
List logList = logs.stream().map(x -> {
instanceId2LastReportTime.put(x.getInstanceId(), System.currentTimeMillis());
LocalInstanceLogDO y = new LocalInstanceLogDO();
BeanUtils.copyProperties(x, y);
y.setWorkerAddress(workerAddress);
return y;
}).collect(Collectors.toList());
try {
//插入到local_instance_log表中
CommonUtils.executeWithRetry0(() -> localInstanceLogRepository.saveAll(logList));
} catch (Exception e) {
log.warn("[InstanceLogService] persistent instance logs failed, these logs will be dropped: {}.", logs, e);
}
}
原创不易,未得准许,请勿转载,翻版必究