WorkManager流程分析和源码解析
本文字数：2830字
预计阅读时间：20分钟
01 前言
WorkManager统一了对于Android后台任务的管理。在此之前，从6.0开始Google引入了Doze机制，并且在之后的几个版本对Android的后台行为及广播的限制越来越严格。在Android8.0时Google官方推荐开发者使用JobScheduler来代替Service+Broadcast的后台任务管理方式。为了兼容老版本，android-job, Firebase JobDispatcher和GCMNetworkManager都曾是开发者的选择，而Firebase JobDispatcher和GCMNetworkManager需要支持google play service，并不适用于国内app场景。2018年，google推出的jetpack中包含了WorkManager，之后android-job停止维护，google官方为Firebase JobDispatcher和GCMNetworkManager提出了WorkManager的迁移方案。今年，在Android11的行为变更中提到，如果应用以 API 级别“R”或更高级别为目标平台，则在搭载 Android 6.0（API 级别 23）或更高版本的设备上会停用 Firebase JobDispatcher 和 GcmNetworkManager API 调用。所以在一些场景中，使用WorkManager来维护我们的后台任务可以说是官方推荐的唯一方式。本文将介绍WorkManager的使用方式，并通过剖析WorkManager的内部实现原理，来帮助大家更好的理解WorkManager的实现。
02 WorkManager的特点与适用场景
特点：
保证任务一定会被执行
WorkManager有自己的数据库，每一个任务的信息与任务状态，都会保存在本地数据库中。所以即使程序没有在运行，或者在设备重启等情况下，WorkManager依然可以保证任务的执行，只是不保证任务立即被执行。合理使用设备资源
在执行很多周期性或非立即执行的任务时，WorkManager提供我们API，帮助我们合理利用设备资源，避免不必要的内存，流量，电量等消耗。
适用场景：
可延迟进行的任务
a.满足某些条件才执行的任务，如需要在充电时才执行的任务。
b.用户无感知或可延迟感知的任务，如同步配置信息，同步资源，同步通讯录等。定期重复性任务，但时效性要求不高的，如定期log上传，数据备份等。
退出应用后还应继续执行的未完成任务。
03 WorkManager的使用
WorkManager的使用非常简单，分为如下几个步骤：
创建一个后台任务Worker。
定义WorkRequest，配置运行任务的方式和时间。
将任务提交给系统处理。
观察Worker的进度或状态。
3.1创建后台任务Worker
WorkManager提供了四种Worker的创建模式：
Worker：最简单的实现，WorkManager 会在后台线程上自动运行它。
CoroutineWorker：CoroutineWorker针对后台工作公开挂起函数。默认情况下，它们运行默认的Dispatcher。
RxWorker：如果有很多现有异步代码是用 RxJava 建模的建议使用。与所有 RxJava2 概念一样，可以自由选择所需的线程处理策略。
ListenableWorker：是Worker，CoroutineWorker，RxWorker的基类，为需要与基于回调的异步 API进行交互并且不使用 RxJava2 的 Java 开发者而设计。Worker的创建示例：
class UploadWorker(appContext: Context, workerParams: WorkerParameters)
: Worker(appContext, workerParams) {
override fun doWork(): Result {
// Do the work here--in this case, upload the images.
uploadImages()
// Indicate whether the task finished successfully with the Result
return Result.success()
}
}
3.2配置运行任务方式和时间
3.2.1一次性任务
val uploadWorkRequest = OneTimeWorkRequestBuilder().build()
3.2.2周期性任务
//执行多次任务，每隔12个小时执行一次
val uploadWorkRequest = PeriodicWorkRequestBuilder(12, TimeUnit.HOURS)
.build()
3.2.3带约束条件的任务
// Create a Constraints object that defines when the task should run
val constraints = Constraints.Builder()
.setRequiresDeviceIdle(true)
.setRequiresCharging(true)
.build()
// ...then create a OneTimeWorkRequest that uses those constraints
val compressionWork = OneTimeWorkRequestBuilder()
.setConstraints(constraints)
.build()
3.2.4延迟任务
val uploadWorkRequest = OneTimeWorkRequestBuilder()
.setInitialDelay(10, TimeUnit.SECONDS)//符合触发条件后，延迟10秒执行
.build()
3.3将任务提交给系统处理
WorkManager.getInstance(myContext).enqueue(uploadWorkRequest)
3.3.1多任务调度：
WorkManager.getInstance()
// First, run all the A tasks (in parallel):
.beginWith(workA1, workA2, workA3)
// ...when all A tasks are finished, run the single B task:
.then(workB)
// ...then run the C tasks (in any order):
.then(workC1, workC2)
.enqueue()
3.4观察Worker的进度或状态
WorkManager.getInstance(myContext).getWorkInfoByIdLiveData(uploadWorkRequest.id)
.observe(lifecycleOwner, Observer { workInfo ->
})
04 WorkManager流程分析与源码解析
这个章节将会从以下几个方面梳理WorkManager的流程与源码：
创建
a.WorkManager的初始化
b.WorkRequest的创建非约束条件任务的执行
带约束条件任务的执行
从最基础的流程开始分析：创建一个不带任何约束条件的一次性任务。在doWork()中让线程休息5s。
val work1Request = OneTimeWorkRequestBuilder().build()
WorkManager.getInstance(this).enqueue(work1Request)
class Worker1(appContext: Context, workerParams: WorkerParameters) :
Worker(appContext, workerParams) {
override fun doWork(): Result {
Thread.sleep(5000)
return Result.success()
}
}
4.1创建
首先梳理一下WorkManager的初始化过程。
4.1.1. WorkManager的初始化
在默认的情况下，WorkManager并不是在我们调用WorkManager.getInstance() 时创建的。通过反编译一下apk，会发现在AndroidManifest文件中注册了名为WorkManagerInitializer的ContentProvider。因此WorkManager在app冷启动的时候已经被创建。
//AndroidManifest.xml
WorkManagerInitializer的onCreate()方法：
//WorkManagerInitializer
public boolean onCreate() {
// Initialize WorkManager with the default configuration.
WorkManager.initialize(getContext(), new Configuration.Builder().build());
return true;
}
由于WorkManager是个单例，在此时WorkManager就已经被初始化了。在initialize()之前，会创建一个默认的Configuration。Configuration设置了许多属性，用来管理和调度工作的方式。通常我们使用WorkManager默认创建的Configuration即可。如需使用自己的Configuration，可参考官方文档，有明确的使用说明。我们继续看initialize()的实现，由于WorkManager是个抽象类，真正的构造方法是在他的子类WorkManagerImpl实现的：
//WorkManagerImpl
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public static void initialize(@NonNull Context context, @NonNull Configuration configuration) {
synchronized (sLock) {
if (sDelegatedInstance != null && sDefaultInstance != null) {
throw new IllegalStateException("WorkManager is already initialized. Did you "
+ "try to initialize it manually without disabling "
+ "WorkManagerInitializer? See "
+ "WorkManager#initialize(Context, Configuration) or the class level "
+ "Javadoc for more information.");
}
if (sDelegatedInstance == null) {
context = context.getApplicationContext();
if (sDefaultInstance == null) {
sDefaultInstance = new WorkManagerImpl(
context,
configuration,
new WorkManagerTaskExecutor(configuration.getTaskExecutor()));
}
sDelegatedInstance = sDefaultInstance;
}
}
}
此时sDelegatedInstance为null，WorkManager会先创建一个默认的WorkManagerTaskExecutor对象，用来执行WorkManager的任务。之后创建一个WorkManagerImpl对象：
//WorkManagerImpl
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public WorkManagerImpl(
@NonNull Context context,
@NonNull Configuration configuration,
@NonNull TaskExecutor workTaskExecutor) {
this(context,
configuration,
workTaskExecutor,
context.getResources().getBoolean(R.bool.workmanager_test_configuration));
}
//WorkManagerImpl
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public WorkManagerImpl(
@NonNull Context context,
@NonNull Configuration configuration,
@NonNull TaskExecutor workTaskExecutor,
boolean useTestDatabase) {
this(context,
configuration,
workTaskExecutor,
WorkDatabase.create(
context.getApplicationContext(),
workTaskExecutor.getBackgroundExecutor(),
useTestDatabase)
);
}
WorkManager在此时创建了数据库。WorkDatabase.create()将任务列表序列化到本地，记录每一个任务的属性，执行条件，执行顺序及执行状态等。从而保证任务在冷启动或硬件重启后，可以根据条件继续执行。接着看this()的实现：
//WorkManagerImpl
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public WorkManagerImpl(
@NonNull Context context,
@NonNull Configuration configuration,
@NonNull TaskExecutor workTaskExecutor,
@NonNull WorkDatabase database) {
Context applicationContext = context.getApplicationContext();
Logger.setLogger(new Logger.LogcatLogger(configuration.getMinimumLoggingLevel()));
List schedulers = createSchedulers(applicationContext, workTaskExecutor);
Processor processor = new Processor(
context,
configuration,
workTaskExecutor,
database,
schedulers);
internalInit(context, configuration, workTaskExecutor, database, schedulers, processor);
}
到这里有三个重要的初始化步骤。分别是createSchedulers()来根据Build Version创建不同的Schedulers进行任务调度，Processor()用来管理Schedulers的执行，和internalInit()真正的初始化。先看createSchedulers()的实现：
//WorkManagerImpl
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
@NonNull
public List createSchedulers(
@NonNull Context context,
@NonNull TaskExecutor taskExecutor) {
return Arrays.asList(
Schedulers.createBestAvailableBackgroundScheduler(context, this),
// Specify the task executor directly here as this happens before internalInit.
// GreedyScheduler creates ConstraintTrackers and controllers eagerly.
new GreedyScheduler(context, taskExecutor, this));
}
return一个Scheduler数组。其中GreedyScheduler()是常驻的，用来执行没有任何约束的非周期性的任务。接下来看createBestAvailableBackgroundScheduler()的实现。
//Scheduler
@NonNull
static Scheduler createBestAvailableBackgroundScheduler(
@NonNull Context context,
@NonNull WorkManagerImpl workManager) {
Scheduler scheduler;
if (Build.VERSION.SDK_INT >= WorkManagerImpl.MIN_JOB_SCHEDULER_API_LEVEL) {
scheduler = new SystemJobScheduler(context, workManager);
setComponentEnabled(context, SystemJobService.class, true);
Logger.get().debug(TAG, "Created SystemJobScheduler and enabled SystemJobService");
} else {
scheduler = tryCreateGcmBasedScheduler(context);
if (scheduler == null) {
scheduler = new SystemAlarmScheduler(context);
setComponentEnabled(context, SystemAlarmService.class, true);
Logger.get().debug(TAG, "Created SystemAlarmScheduler");
}
}
return scheduler;
}
这段代码对build version进行了判断。若>=23，则返回SystemJobScheduler()，即利用JobScheduler进行任务管理。<23的时候先尝试使用GcmScheduler进行管理。若无法创建GcmScheduler则返回SystemAlarmScheduler()使用AlamManager进行任务管理。返回的这个Scheduler是用来执行周期性，或者有约束性的任务。由此可见，WorkManager创建了两个Scheduler，分别为执行非约束非周期性任务的GreedyScheduler，和执行约束性周期性任务的SystemJobScheduler/GcmBasedScheduler/SystemAlarmScheduler。
这几种Scheduler的构造和执行之后再分析。
之后初始化Processor。Processor存储了Configuration，TaskExecutor，WorkDatabase，schedulers等，用来在适当的时机进行任务调度。再来看internalInit()：
//WorkManagerImpl
private void internalInit(@NonNull Context context,
@NonNull Configuration configuration,
@NonNull TaskExecutor workTaskExecutor,
@NonNull WorkDatabase workDatabase,
@NonNull List schedulers,
@NonNull Processor processor) {
context = context.getApplicationContext();
mContext = context;
mConfiguration = configuration;
mWorkTaskExecutor = workTaskExecutor;
mWorkDatabase = workDatabase;
mSchedulers = schedulers;
mProcessor = processor;
mPreferenceUtils = new PreferenceUtils(workDatabase);
mForceStopRunnableCompleted = false;
// Checks for app force stops.
mWorkTaskExecutor.executeOnBackgroundThread(new ForceStopRunnable(context, this));
}
记录了Configuration，TaskExecutor，WorkDatabase，schedulers，Processor等。然后我们看最后一行执行语句，启动了一个ForceStopRunnable，这个Runnable是干什么用的呢？直接看run()的实现：
//ForceStopRunnable
@Override
public void run() {
// Migrate the database to the no-backup directory if necessary.
WorkDatabasePathHelper.migrateDatabase(mContext);
// Clean invalid jobs attributed to WorkManager, and Workers that might have been
// interrupted because the application crashed (RUNNING state).
Logger.get().debug(TAG, "Performing cleanup operations.");
try {
boolean needsScheduling = cleanUp();
if (shouldRescheduleWorkers()) {
Logger.get().debug(TAG, "Rescheduling Workers.");
mWorkManager.rescheduleEligibleWork();
// Mark the jobs as migrated.
mWorkManager.getPreferenceUtils().setNeedsReschedule(false);
} else if (isForceStopped()) {
Logger.get().debug(TAG, "Application was force-stopped, rescheduling.");
mWorkManager.rescheduleEligibleWork();
} else if (needsScheduling) {
Logger.get().debug(TAG, "Found unfinished work, scheduling it.");
Schedulers.schedule(
mWorkManager.getConfiguration(),
mWorkManager.getWorkDatabase(),
mWorkManager.getSchedulers());
}
mWorkManager.onForceStopRunnableCompleted();
} catch (SQLiteCantOpenDatabaseException
| SQLiteDatabaseCorruptException
| SQLiteAccessPermException exception) {
// ForceStopRunnable is usually the first thing that accesses a database (or an app's
// internal data directory). This means that weird PackageManager bugs are attributed
// to ForceStopRunnable, which is unfortunate. This gives the developer a better error
// message.
String message =
"The file system on the device is in a bad state. WorkManager cannot access "
+ "the app's internal data store.";
Logger.get().error(TAG, message, exception);
throw new IllegalStateException(message, exception);
}
}
这段代码的实现细节先不做深究。但是很明显，这个Runnable的作用就是在WorkManager初始化过程中，发现了未完成的，需要重新执行的任务，或者app被强制kill的情况下，直接对Scheduler进行调度。到此，一个WorkManager的初始化流程就完成了。
总结
WorkManager的初始化是在app冷启动后，由WorkManagerInitializer这个ContentProvider执行的。
初始化过程包含了Configuration，WorkManagerTaskExecutor，WorkDatabase，Schedulers，Processor等的初始化过程。
Schedulers有两个。(1) GreedyScheduler：执行没有任何约束的非周期性的任务。(2) SystemJobScheduler/GcmBasedScheduler/SystemAlarmScheduler：执行周期性或者有约束性的任务。优先返回SystemJobScheduler，在build version小于23的情况下先尝试返回GcmBasedScheduler，若返回为空再返回SystemAlarmScheduler。
初始化的最后，会根据情况找到需要被执行的任务进行调度执行。
WorkManager的初始化流程图：
4.1.2.WorkRequest的创建
梳理完WorkManager的初始化过程后，我们回到示例代码，创建一个OneTimeWorkRequest
val work1Request = OneTimeWorkRequestBuilder().build()
//OneTimeWorkRequest.Builder
/**
* Creates a {@link OneTimeWorkRequest}.
*
* @param workerClass The {@link ListenableWorker} class to run for this work
*/
public Builder(@NonNull Class workerClass) {
super(workerClass);
mWorkSpec.inputMergerClassName = OverwritingInputMerger.class.getName();
}
//WorkRequest.Builder
Builder(@NonNull Class workerClass) {
mId = UUID.randomUUID();
mWorkerClass = workerClass;
mWorkSpec = new WorkSpec(mId.toString(), workerClass.getName());
addTag(workerClass.getName());
}
OneTimeWorkRequest为builder对象创建了WorkSpec对象用来保存任务id和类名，其中id是通过UUID自动生成的。request的tag默认是通过类名生成的，外部也可调用addTag()方法设置标签。另外为WorkSpec设置了默认的任务输入流的合并规则：OverwritingInputMerger。接着看build()方法的实现：
//WorkRequest.Builder
public final @NonNull W build() {
W returnValue = buildInternal();
// Create a new id and WorkSpec so this WorkRequest.Builder can be used multiple times.
mId = UUID.randomUUID();
mWorkSpec = new WorkSpec(mWorkSpec);
mWorkSpec.id = mId.toString();
return returnValue;
}
buildInternal()方法返回了一个WorkRequest对象，这是个抽象方法，在子类OneTimeWorkRequest.Builder中的实现如下：
//OneTimeWorkRequest.Builder
@Override
@NonNull OneTimeWorkRequest buildInternal() {
if (mBackoffCriteriaSet
&& Build.VERSION.SDK_INT >= 23
&& mWorkSpec.constraints.requiresDeviceIdle()) {
throw new IllegalArgumentException(
"Cannot set backoff criteria on an idle mode job");
}
if (mWorkSpec.runInForeground
&& Build.VERSION.SDK_INT >= 23
&& mWorkSpec.constraints.requiresDeviceIdle()) {
throw new IllegalArgumentException(
"Cannot run in foreground with an idle mode constraint");
}
return new OneTimeWorkRequest(this);
}
由于我们没有为WorkSpec设置其他属性，目前也没有约束条件，所以直接返回一个OneTimeWorkRequest对象。
//OneTimeWorkRequest
OneTimeWorkRequest(Builder builder) {
super(builder.mId, builder.mWorkSpec, builder.mTags);
}
把Builder的id, WorkSpec对象和tag赋给OneTimeWorkRequest对象。再回到Builder的build()方法：
//OneTimeWorkRequest.Builder
public final @NonNull W build() {
W returnValue = buildInternal();
// Create a new id and WorkSpec so this WorkRequest.Builder can be used multiple times.
mId = UUID.randomUUID();
mWorkSpec = new WorkSpec(mWorkSpec);
mWorkSpec.id = mId.toString();
return returnValue;
}
在buildInternal()拿到OneTimeWorkRequest对象之后，为Builder创建了一个新的WorkSpec对象，并赋予了新的UUID。虽然与原先的WorkSpec对象中每个属性的值是一致的，但指向了不同的内存地址。这么做的目的是为了这个Builder对象可被重复利用。好了，现在我们一个任务的WorkRequest创建就完成了。
总结
WorkRequest的创建是为了持有三个重要的成员变量。分别是：
mId：由UUID生成的任务id。
mWorkSpec：每个任务的属性。
mTags：每个任务的标签。
WorkRequest创建的流程图
4.2非约束条件任务的执行过程
执行OneTimeWorkRequest
WorkManager.getInstance(this).enqueue(work1Request)
根据第一节的分析，WorkManager是个单例，在app启动的时候就已经被初始化了。所以直接看enqueue()的实现：
//WorkManager
@NonNull
public final Operation enqueue(@NonNull WorkRequest workRequest) {
return enqueue(Collections.singletonList(workRequest));
}
//WorkManager
@NonNull
public abstract Operation enqueue(@NonNull List requests);
//WorkManagerImpl
@NonNull
public Operation enqueue(
@NonNull List workRequests) {
// This error is not being propagated as part of the Operation, as we want the
// app to crash during development. Having no workRequests is always a developer error.
if (workRequests.isEmpty()) {
throw new IllegalArgumentException(
"enqueue needs at least one WorkRequest.");
}
return new WorkContinuationImpl(this, workRequests).enqueue();
}
创建一个WorkContinuationImpl()对象，再执行enqueue()方法。WorkContinuationImpl是WorkContinuation的子类。用来把多个OneTimeWorkRequest根据需求串行，并行或合并处理。我们熟悉的then()，combine()，enqueue()等都是这个类的方法。
//WorkContinuationImpl
WorkContinuationImpl(@NonNull WorkManagerImpl workManagerImpl,
String name,
ExistingWorkPolicy existingWorkPolicy,
@NonNull List work,
@Nullable List parents) {
mWorkManagerImpl = workManagerImpl;
mName = name;
mExistingWorkPolicy = existingWorkPolicy;
mWork = work;
mParents = parents;
mIds = new ArrayList<>(mWork.size());
mAllIds = new ArrayList<>();
if (parents != null) {
for (WorkContinuationImpl parent : parents) {
mAllIds.addAll(parent.mAllIds);
}
}
for (int i = 0; i < work.size(); i++) {
String id = work.get(i).getStringId();
mIds.add(id);
mAllIds.add(id);
}
}
WorkContinuation保存了任务相关的所有信息，如WorkManager，WorkRequest，父WorkContinuation等。继续看WorkContinuationImpl的enqueue()方法的实现：
//WorkContinuationImpl
@Override
public @NonNull Operation enqueue() {
// Only enqueue if not already enqueued.
if (!mEnqueued) {
// The runnable walks the hierarchy of the continuations
// and marks them enqueued using the markEnqueued() method, parent first.
EnqueueRunnable runnable = new EnqueueRunnable(this);
mWorkManagerImpl.getWorkTaskExecutor().executeOnBackgroundThread(runnable);
mOperation = runnable.getOperation();
} else {
Logger.get().warning(TAG,
String.format("Already enqueued work ids (%s)", TextUtils.join(", ", mIds)));
}
return mOperation;
}
WorkManager的TaskExecutor执行了EnqueueRunnable。EnqueueRunnable中run()的实现：
//EnqueueRunnable
@Override
public void run() {
try {
if (mWorkContinuation.hasCycles()) {
throw new IllegalStateException(
String.format("WorkContinuation has cycles (%s)", mWorkContinuation));
}
boolean needsScheduling = addToDatabase();
if (needsScheduling) {
// Enable RescheduleReceiver, only when there are Worker's that need scheduling.
final Context context =
mWorkContinuation.getWorkManagerImpl().getApplicationContext();
PackageManagerHelper.setComponentEnabled(context, RescheduleReceiver.class, true);
scheduleWorkInBackground();
}
mOperation.setState(Operation.SUCCESS);
} catch (Throwable exception) {
mOperation.setState(new Operation.State.FAILURE(exception));
}
}
addToDatabase()的作用是把WorkSpec存入到数据库,并对任务的状态进行校验。当前的case会返回true。PackageManagerHelper.setComponentEnabled()开启了RescheduleReceiver。通过反编译我们得知这个Receiver是在AndroidManifest中注册的，默认是disable的。监听了开机，时间变化，时区变化这三个广播。
//AndroidManifest
scheduleWorkInBackground()的实现：
//EnqueueRunnable
/**
* Schedules work on the background scheduler.
*/
@VisibleForTesting
public void scheduleWorkInBackground() {
WorkManagerImpl workManager = mWorkContinuation.getWorkManagerImpl();
Schedulers.schedule(
workManager.getConfiguration(),
workManager.getWorkDatabase(),
workManager.getSchedulers());
}
这部分就是任务调度的实现。拿到WorkManager对象后调用了Schedulers.schedule()方法，传入了Configuration, WorkDatabase, Scheduler这三个对象。执行schedule()方法：
//Schedulers
public static void schedule(
@NonNull Configuration configuration,
@NonNull WorkDatabase workDatabase,
List schedulers) {
if (schedulers == null || schedulers.size() == 0) {
return;
}
WorkSpecDao workSpecDao = workDatabase.workSpecDao();
List eligibleWorkSpecs;
workDatabase.beginTransaction();
try {
eligibleWorkSpecs = workSpecDao.getEligibleWorkForScheduling(
configuration.getMaxSchedulerLimit());
if (eligibleWorkSpecs != null && eligibleWorkSpecs.size() > 0) {
long now = System.currentTimeMillis();
// Mark all the WorkSpecs as scheduled.
// Calls to Scheduler#schedule() could potentially result in more schedules
// on a separate thread. Therefore, this needs to be done first.
for (WorkSpec workSpec : eligibleWorkSpecs) {
workSpecDao.markWorkSpecScheduled(workSpec.id, now);
}
}
workDatabase.setTransactionSuccessful();
} finally {
workDatabase.endTransaction();
}
if (eligibleWorkSpecs != null && eligibleWorkSpecs.size() > 0) {
WorkSpec[] eligibleWorkSpecsArray = eligibleWorkSpecs.toArray(new WorkSpec[0]);
// Delegate to the underlying scheduler.
for (Scheduler scheduler : schedulers) {
scheduler.schedule(eligibleWorkSpecsArray);
}
}
}
先进行了一系列的数据库操作，然后开始根据条件每个任务进行调度。其中eligibleWorkSpecs返回的是在ENQUEUED状态下，未被执行且未被取消的WorkSpec列表，然后更新这些任务的request状态到数据库。最后遍历schedulers调用scheduler.schedule()对每个任务进行调度处理。由于示例代码创建的是没有约束的一次性任务，所以看一下GreedyScheduler对于schedule()方法的实现：
//GreedyScheduler
@Override
public void schedule(@NonNull WorkSpec... workSpecs) {
if (mIsMainProcess == null) {
// The default process name is the package name.
mIsMainProcess = TextUtils.equals(mContext.getPackageName(), getProcessName());
}
if (!mIsMainProcess) {
Logger.get().info(TAG, "Ignoring schedule request in non-main process");
return;
}
registerExecutionListenerIfNeeded();
// Keep track of the list of new WorkSpecs whose constraints need to be tracked.
// Add them to the known list of constrained WorkSpecs and call replace() on
// WorkConstraintsTracker. That way we only need to synchronize on the part where we
// are updating mConstrainedWorkSpecs.
List constrainedWorkSpecs = new ArrayList<>();
List constrainedWorkSpecIds = new ArrayList<>();
for (WorkSpec workSpec : workSpecs) {
if (workSpec.state == WorkInfo.State.ENQUEUED
&& !workSpec.isPeriodic()
&& workSpec.initialDelay == 0L
&& !workSpec.isBackedOff()) {
if (workSpec.hasConstraints()) {
if (SDK_INT >= 23 && workSpec.constraints.requiresDeviceIdle()) {
// Ignore requests that have an idle mode constraint.
Logger.get().debug(TAG,
String.format("Ignoring WorkSpec %s, Requires device idle.",
workSpec));
} else if (SDK_INT >= 24 && workSpec.constraints.hasContentUriTriggers()) {
// Ignore requests that have content uri triggers.
Logger.get().debug(TAG,
String.format("Ignoring WorkSpec %s, Requires ContentUri triggers.",
workSpec));
} else {
constrainedWorkSpecs.add(workSpec);
constrainedWorkSpecIds.add(workSpec.id);
}
} else {
Logger.get().debug(TAG, String.format("Starting work for %s", workSpec.id));
mWorkManagerImpl.startWork(workSpec.id);
}
}
}
// onExecuted() which is called on the main thread also modifies the list of mConstrained
// WorkSpecs. Therefore we need to lock here.
synchronized (mLock) {
if (!constrainedWorkSpecs.isEmpty()) {
Logger.get().debug(TAG, String.format("Starting tracking for [%s]",
TextUtils.join(",", constrainedWorkSpecIds)));
mConstrainedWorkSpecs.addAll(constrainedWorkSpecs);
mWorkConstraintsTracker.replace(mConstrainedWorkSpecs);
}
}
}
在: (1) WorkSpec是ENQUEUED的状态 (2) 非周期性任务 (3) 非延迟任务 (4) 非撤销的任务 (5) 没有其它约束的任务 满足这五个条件后，直接调用：
//GreedyScheduler
mWorkManagerImpl.startWork(workSpec.id);
让WorkManager直接去执行任务。继续看startWork()的实现：
//WorkManagerImpl
/**
* @param workSpecId The {@link WorkSpec} id to start
* @hide
*/
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public void startWork(@NonNull String workSpecId) {
startWork(workSpecId, null);
}
//WorkManagerImpl
/**
* @param workSpecId The {@link WorkSpec} id to start
* @param runtimeExtras The {@link WorkerParameters.RuntimeExtras} associated with this work
* @hide
*/
@RestrictTo(RestrictTo.Scope.LIBRARY_GROUP)
public void startWork(
@NonNull String workSpecId,
@Nullable WorkerParameters.RuntimeExtras runtimeExtras) {
mWorkTaskExecutor
.executeOnBackgroundThread(
new StartWorkRunnable(this, workSpecId, runtimeExtras));
}
WorkTaskExecutor对任务进行了调度。StartWorkRunnable的run()的实现：
//StartWorkRunnable
@Override
public void run() {
mWorkManagerImpl.getProcessor().startWork(mWorkSpecId, mRuntimeExtras);
}
StartWorkRunnable会将任务的信息交给Processor，由Processor调用startWork()去执行任务:
//Processor
/**
* Starts a given unit of work in the background.
*
* @param id The work id to execute.
* @param runtimeExtras The {@link WorkerParameters.RuntimeExtras} for this work, if any.
* @return {@code true} if the work was successfully enqueued for processing
*/
public boolean startWork(
@NonNull String id,
@Nullable WorkerParameters.RuntimeExtras runtimeExtras) {
WorkerWrapper workWrapper;
synchronized (mLock) {
// Work may get triggered multiple times if they have passing constraints
// and new work with those constraints are added.
if (mEnqueuedWorkMap.containsKey(id)) {
Logger.get().debug(
TAG,
String.format("Work %s is already enqueued for processing", id));
return false;
}
workWrapper =
new WorkerWrapper.Builder(
mAppContext,
mConfiguration,
mWorkTaskExecutor,
this,
mWorkDatabase,
id)
.withSchedulers(mSchedulers)
.withRuntimeExtras(runtimeExtras)
.build();
ListenableFuture future = workWrapper.getFuture();
future.addListener(
new FutureListener(this, id, future),
mWorkTaskExecutor.getMainThreadExecutor());
mEnqueuedWorkMap.put(id, workWrapper);
}
mWorkTaskExecutor.getBackgroundExecutor().execute(workWrapper);
Logger.get().debug(TAG, String.format("%s: processing %s", getClass().getSimpleName(), id));
return true;
}
startWork()方法中创建了一个WorkerWrapper的Runnable对象，交由WorkTaskExecutor调度处理。WorkerWrapper的run()方法的实现:
//WorkerWrapper
@WorkerThread
@Override
public void run() {
mTags = mWorkTagDao.getTagsForWorkSpecId(mWorkSpecId);
mWorkDescription = createWorkDescription(mTags);
runWorker();
}
//WorkerWrapper
private void runWorker() {
if (tryCheckForInterruptionAndResolve()) {
return;
}
mWorkDatabase.beginTransaction();
try {
mWorkSpec = mWorkSpecDao.getWorkSpec(mWorkSpecId);
...
mWorkDatabase.setTransactionSuccessful();
} finally {
mWorkDatabase.endTransaction();
}
// Merge inputs. This can be potentially expensive code, so this should not be done inside
// a database transaction.
...
WorkerParameters params = new WorkerParameters(
UUID.fromString(mWorkSpecId),
input,
mTags,
mRuntimeExtras,
mWorkSpec.runAttemptCount,
mConfiguration.getExecutor(),
mWorkTaskExecutor,
mConfiguration.getWorkerFactory(),
new WorkProgressUpdater(mWorkDatabase, mWorkTaskExecutor),
new WorkForegroundUpdater(mForegroundProcessor, mWorkTaskExecutor));
// Not always creating a worker here, as the WorkerWrapper.Builder can set a worker override
// in test mode.
if (mWorker == null) {
mWorker = mConfiguration.getWorkerFactory().createWorkerWithDefaultFallback(
mAppContext,
mWorkSpec.workerClassName,
params);
}
...
// Try to set the work to the running state. Note that this may fail because another thread
// may have modified the DB since we checked last at the top of this function.
if (trySetRunning()) {
if (tryCheckForInterruptionAndResolve()) {
return;
}
final SettableFuture future = SettableFuture.create();
// Call mWorker.startWork() on the main thread.
mWorkTaskExecutor.getMainThreadExecutor()
.execute(new Runnable() {
@Override
public void run() {
try {
Logger.get().debug(TAG, String.format("Starting work for %s",
mWorkSpec.workerClassName));
mInnerFuture = mWorker.startWork();
future.setFuture(mInnerFuture);
} catch (Throwable e) {
future.setException(e);
}
}
});
// Avoid synthetic accessors.
...
}
这段代码很长，我们省略了一些判断步骤和与示例无关的参数设置。先创建一个WorkerParameters对象。然后调用mConfiguration.getWorkerFactory().createWorkerWithDefaultFallback()方法创建Worker对象。这个方法我们不展开了，返回的就是我们自己的Woker对象，即Worker1的实例。之后交由WorkTaskExecutor调度处理。在run()方法的实现，我们看到调用了mWorker.startWork()方法：
//ListenableWorker
@MainThread
public abstract @NonNull ListenableFuture startWork();
ListenableWorker是个抽象类，是所有Worker的父类。Worker1也继承Worker类，startWork()在Worker中的实现：
//Worker
@Override
public final @NonNull ListenableFuture startWork() {
mFuture = SettableFuture.create();
getBackgroundExecutor().execute(new Runnable() {
@Override
public void run() {
try {
Result result = doWork();
mFuture.set(result);
} catch (Throwable throwable) {
mFuture.setException(throwable);
}
}
});
return mFuture;
}
在run()的实现执行了doWork()方法，即执行了我们Worker1的doWork()方法。
//Worker1
class Worker1(appContext: Context, workerParams: WorkerParameters) :
Worker(appContext, workerParams) {
override fun doWork(): Result {
Thread.sleep(5000)
return Result.success()
}
}
在执行完这个任务后，返回了success。Worker也就执行完成了。回到WorkerWrapper的runWorker()方法看接下来的处理：
//WorkerWrapper
private void runWorker() {
...
final SettableFuture future = SettableFuture.create();
// Call mWorker.startWork() on the main thread.
mWorkTaskExecutor.getMainThreadExecutor()
.execute(new Runnable() {
@Override
public void run() {
try {
Logger.get().debug(TAG, String.format("Starting work for %s",
mWorkSpec.workerClassName));
mInnerFuture = mWorker.startWork();
future.setFuture(mInnerFuture);
} catch (Throwable e) {
future.setException(e);
}
}
});
// Avoid synthetic accessors.
final String workDescription = mWorkDescription;
future.addListener(new Runnable() {
@Override
@SuppressLint("SyntheticAccessor")
public void run() {
try {
// If the ListenableWorker returns a null result treat it as a failure.
ListenableWorker.Result result = future.get();
if (result == null) {
Logger.get().error(TAG, String.format(
"%s returned a null result. Treating it as a failure.",
mWorkSpec.workerClassName));
} else {
Logger.get().debug(TAG, String.format("%s returned a %s result.",
mWorkSpec.workerClassName, result));
mResult = result;
}
} catch (CancellationException exception) {
// Cancellations need to be treated with care here because innerFuture
// cancellations will bubble up, and we need to gracefully handle that.
Logger.get().info(TAG, String.format("%s was cancelled", workDescription),
exception);
} catch (InterruptedException | ExecutionException exception) {
Logger.get().error(TAG,
String.format("%s failed because it threw an exception/error",
workDescription), exception);
} finally {
onWorkFinished();
}
}
}, mWorkTaskExecutor.getBackgroundExecutor());
} else {
resolveIncorrectStatus();
}
}
startWork()返回了一个Future对象mInnerFuture，调用future.setFuture(mInnerFuture)去处理doWork()返回的result。再经过一系列判断后，最终执行了onWorkFinished()方法：
//WorkerWrapper
void onWorkFinished() {
boolean isWorkFinished = false;
if (!tryCheckForInterruptionAndResolve()) {
mWorkDatabase.beginTransaction();
try {
WorkInfo.State state = mWorkSpecDao.getState(mWorkSpecId);
mWorkDatabase.workProgressDao().delete(mWorkSpecId);
if (state == null) {
// state can be null here with a REPLACE on beginUniqueWork().
// Treat it as a failure, and rescheduleAndResolve() will
// turn into a no-op. We still need to notify potential observers
// holding on to wake locks on our behalf.
resolve(false);
isWorkFinished = true;
} else if (state == RUNNING) {
handleResult(mResult);
// Update state after a call to handleResult()
state = mWorkSpecDao.getState(mWorkSpecId);
isWorkFinished = state.isFinished();
} else if (!state.isFinished()) {
rescheduleAndResolve();
}
mWorkDatabase.setTransactionSuccessful();
} finally {
mWorkDatabase.endTransaction();
}
}
// Try to schedule any newly-unblocked workers, and workers requiring rescheduling (such as
// periodic work using AlarmManager). This code runs after runWorker() because it should
// happen in its own transaction.
// Cancel this work in other schedulers. For example, if this work was
// completed by GreedyScheduler, we should make sure JobScheduler is informed
// that it should remove this job and AlarmManager should remove all related alarms.
if (mSchedulers != null) {
if (isWorkFinished) {
for (Scheduler scheduler : mSchedulers) {
scheduler.cancel(mWorkSpecId);
}
}
Schedulers.schedule(mConfiguration, mWorkDatabase, mSchedulers);
}
}
在onWorkFinished()会对刚刚执行完毕的任务作进一步处理。首先获取任务的当前状态state，然后从db中删除这个任务，再根据state作进一步处理。在我们的示例中，这时候state应该是RUNNING，我们看一下handleResult(mResult)的实现：
//WorkerWrapper
private void handleResult(ListenableWorker.Result result) {
if (result instanceof ListenableWorker.Result.Success) {
Logger.get().info(
TAG,
String.format("Worker result SUCCESS for %s", mWorkDescription));
if (mWorkSpec.isPeriodic()) {
resetPeriodicAndResolve();
} else {
setSucceededAndResolve();
}
} else if (result instanceof ListenableWorker.Result.Retry) {
Logger.get().info(
TAG,
String.format("Worker result RETRY for %s", mWorkDescription));
rescheduleAndResolve();
} else {
Logger.get().info(
TAG,
String.format("Worker result FAILURE for %s", mWorkDescription));
if (mWorkSpec.isPeriodic()) {
resetPeriodicAndResolve();
} else {
setFailedAndResolve();
}
}
}
在handleResult()方法中会根据任务类型和result结果进行不同的处理。例如周期性的任务会重新将这个任务的状态设置为ENQUEUED，更新其他相关参数，并更新数据库。我们示例中已经完成的一次性任务将会被更新成SUCCEEDED的状态，具体的处理的过程就不展开了。handleResult()执行完毕后更新isWorkFinished。如果isWorkFinished为true，由于我们在GreedyScheduler已经处理了这个任务，为了避免这个任务被其他schedulers处理，WorkManager遍历了mSchedulers列表，并将这个任务从其他schedulers中移除。最后再次执行Schedulers.schedule()方法，schedule下一个任务。
总结
在WorkManager执行了enqueue()后，创建WorkContinuationImpl对象执行enqueue()方法。
WorkContinuationImpl持有的EnqueueRunnable对象将任务添加到db，并交给Schedulers去调度。
Schedulers将任务交给每一个Scheduler去处理。在我们的示例中，GreedyScheduler会先处理这个任务。
GreedyScheduler经过一系列判断后，调用WorkManager的startWork()方法执行这种一次性，非延迟，无约束的任务。
WorkManager持有的StartWorkRunnable对象会将任务交给Processor去处理，执行startWork()方法。
Processor创建一个WorkerWrapper对象，由它去调用Worker的startWork()方法，执行我们自定义worker的任务，并返回相应的result。
任务完成后，WorkerWrapper会根据result对任务状态，db等进行更新，然后schedule下一个任务。
WorkManager任务执行流程图：
4.3带约束的任务的执行过程
上面的章节我们分析了一个没有约束条件的一次性任务的执行过程。接下来我们来分析一个带约束条件的任务的执行过程。
创建一个非低电量才能执行的任务：
val constraints = Constraints.Builder()
.setRequiresBatteryNotLow(true)
.build()
val work2Request = OneTimeWorkRequestBuilder()
.setConstraints(constraints)
.build()
WorkManager.getInstance(this).enqueue(work2Request)
任务的创建过程中，会为WorkSpec添加Constraints属性。
public final @NonNull B setConstraints(@NonNull Constraints constraints) {
mWorkSpec.constraints = constraints;
return getThis();
}
在任务执行的过程中，由于增加了约束条件，根据之前章节的分析，常驻的GreedyScheduler的schedule()方法将不会startWork()，而是根据build version交由SystemJobScheduler或SystemAlarmScheduler进行处理。先来看使用SystemJobScheduler的情况：
4.3.1.SystemJobScheduler(Build Version >=23)
SystemJobScheduler使用的是JobScheduler来调度执行任务。由于JobScheduler的实现过程分析不在本文的讨论范围，所以只看WorkManager是如何使用JobScheduler进行任务调度的。通常JobScheduler的使用步骤如下：
创建JobService。
配置JobInfo。
执行。
SystemJobService： SystemJobService是执行任务的服务类，在onStartJob()中，会调用WorkManagerImpl的startWork()执行任务。
//SystemJobService
@Override
public boolean onStartJob(@NonNull JobParameters params) {
... ...
String workSpecId = getWorkSpecIdFromJobParameters(params);
... ...
synchronized (mJobParameters) {
... ...
mJobParameters.put(workSpecId, params);
}
... ...
mWorkManagerImpl.startWork(workSpecId, runtimeExtras);
return true;
}
SystemJobScheduler： 在初始化SystemJobScheduler的时候会获取JobScheduler对象：
//SystemJobScheduler
public SystemJobScheduler(@NonNull Context context, @NonNull WorkManagerImpl workManager) {
this(context,
workManager,
(JobScheduler) context.getSystemService(JOB_SCHEDULER_SERVICE),
new SystemJobInfoConverter(context));
}
SystemJobScheduler的schedule()方法执行了scheduleInternal()：
//SystemJobScheduler
public void scheduleInternal(WorkSpec workSpec, int jobId) {
JobInfo jobInfo = mSystemJobInfoConverter.convert(workSpec, jobId);
Logger.get().debug(
TAG,
String.format("Scheduling work ID %s Job ID %s", workSpec.id, jobId));
try {
mJobScheduler.schedule(jobInfo);
} catch (IllegalStateException e) {
... ...
throw new IllegalStateException(message, e);
} catch (Throwable throwable) {
// OEM implementation bugs in JobScheduler cause the app to crash. Avoid crashing.
Logger.get().error(TAG, String.format("Unable to schedule %s", workSpec), throwable);
}
}
SystemJobInfoConverter.convert()方法就是创建了一个JobInfo，并将Constraints里的约束条件赋予JobInfo对象，之后便执行了JobScheduler.schedule()，根据约束条件对任务进行调度。
4.3.2. SystemAlarmScheduler(Build Version <23)
SystemAlarmScheduler使用的是AlarmManager来调度执行任务。由于AlarmManager的实现过程分析不在本文的讨论范围，所以只看WorkManager是如何使用AlarmManager进行任务调度的。反编译apk后，在AndroidManifest里有如下receiver注册：
在电量变化时，收到BATTERY_LOW的广播。在BatteryNotLowProxy的onReceive()进行处理：
//ConstraintProxy
public static class BatteryNotLowProxy extends ConstraintProxy {
}
@Override
public void onReceive(Context context, Intent intent) {
Logger.get().debug(TAG, String.format("onReceive : %s", intent));
Intent constraintChangedIntent = CommandHandler.createConstraintsChangedIntent(context);
context.startService(constraintChangedIntent);
}
createConstraintsChangedIntent()的执行如下：
//ConstraintProxy
static Intent createConstraintsChangedIntent(@NonNull Context context) {
Intent intent = new Intent(context, SystemAlarmService.class);
intent.setAction(ACTION_CONSTRAINTS_CHANGED);
return intent;
}
SystemAlarmService的onStartCommand()处理如下：
@Override
public int onStartCommand(Intent intent, int flags, int startId) {
super.onStartCommand(intent, flags, startId);
... ...
if (intent != null) {
mDispatcher.add(intent, startId);
}
// If the service were to crash, we want all unacknowledged Intents to get redelivered.
return Service.START_REDELIVER_INTENT;
}
调用了SystemAlarmDispatcher.add()方法。
//SystemAlarmDispatcher
@MainThread
public boolean add(@NonNull final Intent intent, final int startId) {
... ...
if (CommandHandler.ACTION_CONSTRAINTS_CHANGED.equals(action)
&& hasIntentWithAction(CommandHandler.ACTION_CONSTRAINTS_CHANGED)) {
return false;
}
intent.putExtra(KEY_START_ID, startId);
synchronized (mIntents) {
boolean hasCommands = !mIntents.isEmpty();
mIntents.add(intent);
if (!hasCommands) {
// Only call processCommand if this is the first command.
// The call to dequeueAndCheckForCompletion will process the remaining commands
// in the order that they were added.
processCommand();
}
}
return true;
}
add()方法中执行了processCommand()，这段代码的核心执行语句是：
//SystemAlarmDispatcher
mCommandHandler.onHandleIntent(mCurrentIntent, startId,
SystemAlarmDispatcher.this);
在CommandHandler的onHandleIntent()方法中，action为ACTION_CONSTRAINTS_CHANGED的执行是：
//CommandHandler
if (ACTION_CONSTRAINTS_CHANGED.equals(action)) {
handleConstraintsChanged(intent, startId, dispatcher);
}
//CommandHandler
private void handleConstraintsChanged(
@NonNull Intent intent, int startId,
@NonNull SystemAlarmDispatcher dispatcher) {
Logger.get().debug(TAG, String.format("Handling constraints changed %s", intent));
// Constraints changed command handler is synchronous. No cleanup
// is necessary.
ConstraintsCommandHandler changedCommandHandler =
new ConstraintsCommandHandler(mContext, startId, dispatcher);
changedCommandHandler.handleConstraintsChanged();
}
在handleConstraintsChanged()方法的执行中，会创建一个action为ACTION_DELAY_MET的Intent然后由SystemAlarmDispatcher发送出去，实际上也是调用了SystemAlarmDispatcher.add()方法。回到SystemAlarmDispatcher的add()流程。
//ConstraintsCommandHandler
Intent intent = CommandHandler.createDelayMetIntent(mContext, workSpecId);
Logger.get().debug(TAG, String.format(
"Creating a delay_met command for workSpec with id (%s)", workSpecId));
mDispatcher.postOnMainThread(
new SystemAlarmDispatcher.AddRunnable(mDispatcher, intent, mStartId));
回到onHandleIntent()方法，在CommandHandler的onHandleIntent()方法中，action为ACTION_DELAY_MET的执行是：
//CommandHandler
else if (ACTION_DELAY_MET.equals(action)) {
handleDelayMet(intent, startId, dispatcher);
}
handleDelayMet()的执行过程，会调用DelayMetCommandHandler的handleProcessWork()方法，接着执行onAllConstraintsMet()：
@Override
public void onAllConstraintsMet(@NonNull List workSpecIds) {
... ...
synchronized (mLock) {
if (mCurrentState == STATE_INITIAL) {
... ...
boolean isEnqueued = mDispatcher.getProcessor().startWork(mWorkSpecId);
... ...
} else {
Logger.get().debug(TAG, String.format("Already started work for %s", mWorkSpecId));
}
}
}
到这里终于看到由SystemAlarmDispatcher调用了Processor的startWork()方法，回到了之前章节分析的任务执行流程。到此为止，一个任务在不同条件下的创建，执行流程就分析完毕。
05 结语
WorkManager的使用方法简单，但是在使用时还是要分清场景，适用于可延迟，周期性，必须执行完成的任务。通过对源码的分析，WorkManager会针对不同Android版本的选择适当的策略。细致阅读代码，会发现针对指定的系统版本还有一些小的优化点。WorkManager目前已经比较稳定，所以如果在场景适合的情况下，推荐使用WorkManager来代替原有的任务管理方案。
06 参考文献
[1]https://developer.android.google.cn/topic/libraries/architecture/workmanager?
[2]https://developer.android.google.cn/preview/behavior-changes-11
◆ ◆ ◆
上期获奖名单公布
恭喜“阿策~”、“beatyou1”、“天天向上”、“Howie”、“哈哈”！以上读者请添加小编微信：sohu-tech20兑奖~
加入搜狐技术作者天团
千元稿费等你来！
???? 戳这里！
也许你还想看
（▼点击文章标题或封面查看）
Cocos Engine-节点与组件 2020-10-15
全面详细的java线程池解密，看我就够了！ 2020-09-03
【周年福利Round3】Coroutines（协程）我是这样理解的！ 2020-08-20
RecyclerView的曝光统计 2020-07-16
安卓自定义view中绘画几何图形和文字及圆角ImageView图片等api使用及举例 2020-07-02
Android死锁初探 2020-05-21