先看play.jobs.JobsPlugin。
public void onApplicationStart() { int core = Integer.parseInt(Play.configuration.getProperty("play.jobs.pool", "10")); executor = new ScheduledThreadPoolExecutor(core, new PThreadFactory("jobs"), new ThreadPoolExecutor.AbortPolicy()); }
在onAppliactionStart()方法中,实例化一个ScheduledThreadPollExecutor做executor。 接受afterApplicationStart事件中,才会处理Job。
public void afterApplicationStart() { List> jobs = new ArrayList>(); for (Class clazz : Play.classloader.getAllClasses()) { if (Job.class.isAssignableFrom(clazz)) { jobs.add(clazz); } } scheduledJobs = new ArrayList(); for (final Class clazz : jobs) { // @OnApplicationStart if (clazz.isAnnotationPresent(OnApplicationStart.class)) { //check if we're going to run the job sync or async OnApplicationStart appStartAnnotation = clazz.getAnnotation(OnApplicationStart.class); if( !appStartAnnotation.async()) { //run job sync try { Job job = ((Job) clazz.newInstance()); scheduledJobs.add(job); job.run(); if(job.wasError) { if(job.lastException != null) { throw job.lastException; } throw new RuntimeException("@OnApplicationStart Job has failed"); } } catch (...) { ... } } else { //run job async try { Job job = ((Job) clazz.newInstance()); scheduledJobs.add(job); //start running job now in the background @SuppressWarnings("unchecked") Callable callable = (Callable)job; executor.submit(callable); } catch (...) { ... } } } // @On if (clazz.isAnnotationPresent(On.class)) { try { Job job = ((Job) clazz.newInstance()); scheduledJobs.add(job); scheduleForCRON(job); } catch (InstantiationException ex) { throw new UnexpectedException("Cannot instanciate Job " + clazz.getName()); } catch (IllegalAccessException ex) { throw new UnexpectedException("Cannot instanciate Job " + clazz.getName()); } } // @Every if (clazz.isAnnotationPresent(Every.class)) { try { Job job = (Job) clazz.newInstance(); scheduledJobs.add(job); String value = job.getClass().getAnnotation(Every.class).value(); if (value.startsWith("cron.")) { value = Play.configuration.getProperty(value); } value = Expression.evaluate(value, value).toString(); if(!"never".equalsIgnoreCase(value)){ executor.scheduleWithFixedDelay(job, Time.parseDuration(value), Time.parseDuration(value), TimeUnit.SECONDS); } } catch (InstantiationException ex) { throw new UnexpectedException("Cannot instanciate Job " + clazz.getName()); } catch (IllegalAccessException ex) { throw new UnexpectedException("Cannot instanciate Job " + clazz.getName()); } } } public static void scheduleForCRON(Job job) { ... try { ... executor.schedule((Callable)job, nextDate.getTime() - now.getTime(), TimeUnit.MILLISECONDS); ... } catch (Exception ex) { throw new UnexpectedException(ex); } }
第一步读取所有类中的Job类,并判断是否是OnApplicationStart标记,如果是sync同步的,就会直接run。如果async异步的,会加入executor中,虽然执行时间间隔为0毫秒,但是实际执行时间由executor决定。 如果被On标记,play会解析On注解中表达式的值。
这里需要注意,如果是以cron开头,就会读取配置文件中的值,这点之前还没发现哈。 然后根据表达式触发时间与目前时间计算延迟时间,并加入executor中。
如果是Every标记,也会像上面的一样处理注解中表达式,不同的是,play会将这个Job和他的执行时间注入到executor,不用再手动规定延迟执行时间,由executor完全接管执行。
那么被On标记的job如何达到周期执行的效果呢?关键在play.jobs.Job类中。
public class Job extends Invoker.Invocation implements Callable { public void doJob() throws Exception { } public V doJobWithResult() throws Exception { doJob(); return null; } private V withinFilter(play.libs.F.Function0 fct) throws Throwable { for (PlayPlugin plugin : Play.pluginCollection.getEnabledPlugins() ){ if (plugin.getFilter() != null) { return (V)plugin.getFilter().withinFilter(fct); } } return null; } public V call() { Monitor monitor = null; try { if (init()) { before(); V result = null; try { ... result = withinFilter(new play.libs.F.Function0() { public V apply() throws Throwable { return doJobWithResult(); } }); ... } catch (...) { ... } after(); return result; } } catch (Throwable e) { onException(e); } finally { ... _finally(); } return null; } @Override public void _finally() { super._finally(); if (executor == JobsPlugin.executor) { JobsPlugin.scheduleForCRON(this); } } }
run()方法中调用call方法。
call方法中的代码也有init/before/after与请求调用过程类似,因为Job也得初始化相应的上下文。也要通过过滤器,这样JPA事务管理也对Job有用,过滤之后会执行doJobWithReslut()方法,这样就出了框架,进入应用了。
在finally模块中进入__finally()方法后,判断job.executor与JobsPlugin.executor的是否一样,因为在上面处理之中,用On标记的赋值了executor,而every标记的没有赋值。这就是区别,相同的话会再次进入JobsPlugin.scheduleForCRON方法,根据表达式触发时间与目前时间计算延迟时间,并加入executor中。
如此一次一次就达到周期执行效果。 Job中其他几种算法:
public void every(String delay) { every(Time.parseDuration(delay)); } public void every(int seconds) { JobsPlugin.executor.scheduleWithFixedDelay(this, seconds, seconds, TimeUnit.SECONDS); }
every方法,可以达到设定every注解一样的效果,我猜测应该是匿名内部类时使用,因为不能直接加注解,可以调用every方法。
public Promise in(String delay) { return in(Time.parseDuration(delay)); } public Promise in(int seconds) { final Promise smartFuture = new Promise(); JobsPlugin.executor.schedule(getJobCallingCallable(smartFuture), seconds, TimeUnit.SECONDS); return smartFuture; }
in方法,设定延迟时间,延迟执行job。
public Promise now() { final Promise smartFuture = new Promise(); JobsPlugin.executor.submit(getJobCallingCallable(smartFuture)); return smartFuture; }
now方法,立刻执行。
public Promise afterRequest() { InvocationContext current = Invoker.InvocationContext.current(); if(current == null || !Http.invocationType.equals(current.getInvocationType())) { return now(); } final Promise smartFuture = new Promise(); Callable callable = getJobCallingCallable(smartFuture); JobsPlugin.addAfterRequestAction(callable); return smartFuture; }
afterRequest方法,请求执行,利用事件机制,在beforeInvocation与afterInvocation事件清除和执行job。
in/now/afterRequest三个方法,返回的是Promise对象,是Play对Promise模式的实现,至于什么是Premise模式,大家可以google一下。