This is the last post in this blog series about the Play Services Task API and its use in Firebase. If you missed the first three parts, consider jumping back to those before continuing here. When you're all caught up, let's finish up this series!
Throughout this series, we've only ever talked about units of work that are themselves represented by a Task or a Continuation. In reality, however, there are lots of other ways to get work done. Various utilities and libraries may have their own ways of performing threaded work. You might wonder if you have to switch to the Task API to unify all this if you want to switch to Firebase. But you certainly don't have to. The Task API was designed with the capability of integrating with other ways of doing threaded work.
For example, Java has always had the ability to simply fire up a new thread to process something in parallel with other threads. You can write code like this (though I heartily recommend against it on Android):
new Thread(new Runnable() { @Override public void run() { String result = "the output of some long-running compute"; // now figure out what to do with the result... } }).start();
Here we fire up that new thread from the main thread and do exciting work that ends with a String of interest. All that work that went into creating that string happens in parallel with the main thread, which continued executing after the thread was started. If that threaded work happened to block at any point, the main thread would not be held up by it. However, something must be done to get that String result into the place where it's expected. On Android, if that needs to be back on the main thread, you'll have to write more code to arrange for that to happen. This can get hairy. And we can use Tasks to help.
The Play Services Task API provides a way to make other units of work behave like Tasks, even if they weren't implemented as such. The class of interest here is TaskCompletionSource. This allows you to effectively create a Task "placeholder" that some other bit of code can trigger for success or failure. If you wanted that thread from above to behave like a Task without implementing it as a Task (as we learned last time by passing a Callable to Tasks.call()), you could do this:
final TaskCompletionSource<String> source = new TaskCompletionSource<>(); new Thread(new Runnable() { @Override public void run() { String result = "the output of some long-running compute"; source.setResult(result); } }).start(); Task<String> task = source.getTask(); task.addOnCompleteListener(new OnCompleteListener<String>() { ... });
We now have the thread offering its result String to the TaskCompletionSource using its setResult() method. Then, in the original thread, we simply ask the TaskCompletionSource for its "placeholder" Task, and add a listener to that. The result is now handled inside the listener running on the main thread. You can do the same in the failure case by calling the setException() method on the TaskCompletionSource. That will end up triggering any failure listeners, and they'll get a hold of the exception.
This strategy might seem a little bit silly up front, because there are less verbose ways of putting the result of some work back on the main thread. The value here is in the ability to work with that new placeholder Task along with other Tasks you might be working with in a unified fashion.
Imagine you're writing an app that absolutely depends on some values in Firebase Realtime Database, along with the values in Firebase Remote Config. However, to keep your users entertained while they wait for this data to load, you’d like to create a splash screen that shows some animation until that data is available to work with. Oh, and you don't want that screen to appear and disappear in a jarring way in the event that the data happens to be locally cached, so you want the screen to show for a minimum of 2 seconds. How might you implement this screen?
For starters, you'll need to create a new Activity and design and implement the views for the splash screen. That's straightforward. Then you'll need to coordinate the work between Realtime Database and Remote Config, as well as factor in the two second timer. You'll probably want to kick off all that work during the Activity's onCreate() after you create the splash screen views. You could use a series of Continuations to make sure all these things happen in serial, one after another. But why do that if you could instead start them all at once to run in parallel, and make the user wait only as long as it takes to complete the longest item of work? Let's see how!
The Task API provides a couple methods to help you know when several Tasks are all complete. These static utility methods create a new Task that gets triggered in response to the completion of a collection of Tasks that you provide.
Task<Void> Tasks.whenAll(Collection<? extends Task<?>> tasks) Task<Void> Tasks.whenAll(Task...<?> tasks)
One version of whenAll() accepts a Java Collection (such as a List or Set), and the other uses the varargs style of passing multiple parameters to easily form an array of any length. Either way, the returned Task will now get triggered for success when all the other Tasks succeed, and trigger for failure if any one of them fails. Note that the new Task result is parameterized with Void, meaning it doesn't contain any results directly. If you want the results of each individual Task, you'll have to get the results from each of them directly.
This whenAll() function looks pretty handy for knowing when all our concurrent work is done, so we can move the user past the splash screen. The trick for this case is to somehow get a bunch of Task objects the represent each thing we're waiting on.
The Remote Config fetch is easy, because it will give you a Task you can use to listen to the availability your values. Let's kick off that task and remember it:
private Task<Void> fetchTask; // during onCreate: fetchTask = FirebaseRemoteConfig.getInstance().fetch();
Realtime Database isn't as easy, because it doesn't provide a Task for triggering on the completion of available data. However, we can use the TaskCompletionSource we just learned about to trigger a placeholder task when the data is available:
private TaskCompletionSource<DataSnapshot> dbSource = new TaskCompletionSource<>(); private Task dbTask = dbSource.getTask(); // during onCreate: DatabaseReference ref = FirebaseDatabase.getInstance().getReference("/data/of/interest"); ref.addListenerForSingleValueEvent(new ValueEventListener() { @Override public void onDataChange(DataSnapshot dataSnapshot) { dbSource.setResult(dataSnapshot); } @Override public void onCancelled(DatabaseError databaseError) { dbSource.setException(databaseError.toException()); } });
Here, we're registering a listener for the data we need to continue launching the app. That listener will then trigger dbTask to success or failure via dbSource depending on the callback it received.
Lastly, there's the minimum two second delay for the splash screen to stay up. We can also represent that delay as a Task using TaskCompletionSource:
private TaskCompletionSource<Void> delaySource = new TaskCompletionSource<>(); private Task<Void> delayTask = delaySource.getTask(); // during onCreate: new Handler().postDelayed(new Runnable() { @Override public void run() { delaySource.setResult(null); } }, 2000);
For the delay, we're just scheduling a Runnable to execute on the main thread after 2000ms, and that Runnable will then trigger delayTask via delaySource.
Now, we have three Tasks, all operating in parallel, and we can use Tasks.whenAll() to create another Task that triggers when they're all successful:
private Task<Void> allTask; // during onCreate(): allTask = Tasks.whenAll(fetchTask, dbTask, delayTask); allTask.addOnSuccessListener(new OnSuccessListener<Void>() { @Override public void onSuccess(Void aVoid) { DataSnapshot data = dbTask.getResult(); // do something with db data? startActivity(new Intent(SplashScreenActivity.this, MainActivity.class)); } }); allTask.addOnFailureListener(new OnFailureListener() { @Override public void onFailure(@NonNull Exception e) { // apologize profusely to the user! } });
And that should do it! When the final allTask succeeds, we can do whatever we need with the data from the database, then we send the user to MainActivity. Without the use of Tasks here, this code becomes more tedious to write because you'd have to check the state of all the other ongoing units of work at the end of each of them, and proceed only when you know they are all done. Here, the Task API handles those details for you. And you can easily add more Tasks as needed without having to change the logic. Just keep adding Tasks to the collection behind allTask.
It's worth noting that there is a way to block the current thread on the result of one or more Tasks. Normally you don't want to block threads at all, if you can help it, but occasionally it's useful when you have to (such as with Loaders). If you do need to wait on the result of a Task, you can use the await() function:
static <TResult> TResult await(Task<TResult> task) static <TResult> TResult await(Task<TResult> task, long timeout, TimeUnit unit)
With await(), the calling thread simply blocks until the task completes, or the given timeout expires. If it was successful, you'll receive the result object, and if it fails, it will throw an ExecutionException which wraps the underlying cause. Please remember that you should never block the main thread! Only use this when you know you're running on some background thread, OK?
Here's what we covered in the four parts of this blog series:
This should be everything you need to know to make effective use of Play Services Task API! I hope you’re able to use Firebase along with the Task API to make efficient and delightful Android apps.
Successful apps turn into successful revenue generating businesses when the right business model is built into the core app development strategy from the very beginning. Since Firebase is designed to help app developers at every part of their lifecycle, from creating high-quality apps to growing and monetizing their app traffic, let’s take a peek at what monetization concepts you could be thinking about now.
Ready to start exploring AdMob?
Sign up for an AdMob account and link it to your Firebase project.
Alrighty! Thanks for joining us for part three of this blog series about the Play services Task API for Android. By now, you've seen the essentials of the API in part one, and how to select the best style of listener in part two. So, at this point, you probably have everything you need to know to make effective use of the Tasks generated by Firebase APIs. But, if you want to press into some advanced usage of Tasks, keep reading!
We know that some of the Firebase features for Android will do work for you and notify a Task upon completion. But, what if you want to create your own Tasks to perform threaded work? The Task API gives you the tools for this. If you want to work with the Task API without having to integrate Firebase into your app, you can get the library with a dependency in your build.gradle:
compile 'com.google.android.gms:play-services-tasks:9.6.1'
But, if you are integrating Firebase, you'll get this library included for free, so no need to call it out specifically in that case.
There is just one method (with two variants) you can use to kick off a new Task. You can use the static method named "call" on the Tasks utility class for this. The variants are as follows:
Task<TResult> call(Callable<TResult> callable) Task<TResult> call(Executor executor, Callable<TResult> callable)
Just like addOnSuccessListener(), you have a version of call() that executes the work on the main thread and another that submits the work to an Executor. You specify the work to perform inside the passed Callable. A Java Callable is similar to a Runnable, except it's parameterized by some result type, and that type becomes the returned object type of its call() method. This result type then becomes the type of the Task returned by call(). Here's a really simple Callable that just returns a String:
public class CarlyCallable implements Callable<String> { @Override public String call() throws Exception { return "Call me maybe"; } }
Notice that CarlyCallable is parameterized by String, which means its call() method must return a String. Now, you can create a Task out of it with a single line:
Task<String> task = Tasks.call(new CarlyCallable());
After this line executes, you can be certain that the call() method on the CarlyCallable will be invoked on the main thread, and you can add a listener to the Task to find the result (even though that result is totally predictable). More interesting Callables might actually load some data from a database or a network endpoint, and you'd want to have those blocking Callables run on an Executor using the second form of call() that accepts the Executor as the first argument.
Let's say, for the sake of example, you want to process the String result of the CarlyCallable Task after it's been generated. Imagine that we're not so much interested in the text of the resulting String itself, and more interested in a List of individual words in the String. But, we don't necessarily want to modify CarlyCallable because it's doing exactly what it's supposed to, and it could be used in other places as it’s written now. Instead, we'd rather encapsulate the logic that splits words into its own class, and use that after the CarlyCallable returns its String. We can do this with a Continuation. An implementation of the Continuation interface takes the output of one Task, does some processing on it, and returns a result object, not necessarily of the same type. Here's a Continuation that splits a string of words into an List of Strings with each word:
public class SeparateWays implements Continuation<String, List<String>> { @Override public List<String> then(Task<String> task) throws Exception { return Arrays.asList(task.getResult().split(" +")); } }
Notice that the Continuation interface being implemented here is parameterized by two types, an input type (String) and an output type (List). The input and output types are used in the signature of the lone method then() to define what it's supposed to do. Of particular note is the parameter passed to then(). It's a Task, and the String there must match the input type of the Continuation interface. This is how the Continuation gets its input - it pulls the finished result out of the completed Task.
Here's another Continuation that randomizes a List of Strings:
public class AllShookUp implements Continuation<List<String>, List<String>> { @Override public List<String> then(@NonNull Task<List<String>> task) throws Exception { // Randomize a copy of the List, not the input List itself, since it could be immutable final ArrayList<String> shookUp = new ArrayList<>(task.getResult()); Collections.shuffle(shookUp); return shookUp; } }
And another one that joins a List of Strings into a single space-separated String:
private static class ComeTogether implements Continuation<List<String>, String> { @Override public String then(@NonNull Task<List<String>> task) throws Exception { StringBuilder sb = new StringBuilder(); for (String word : task.getResult()) { if (sb.length() > 0) { sb.append(' '); } sb.append(word); } return sb.toString(); } }
Maybe you can see where I'm going with this! Let's tie them all together into a chain of operations that randomizes the word order of a String from a starting Task, and generates a new String with that result:
Task<String> playlist = Tasks.call(new CarlyCallable()) .continueWith(new SeparateWays()) .continueWith(new AllShookUp()) .continueWith(new ComeTogether()); playlist.addOnSuccessListener(new OnSuccessListener<String>() { @Override public void onSuccess(String message) { // The final String with all the words randomized is here } });
The continueWith() method on Task returns a new Task that represents the computation of the prior Task after it’s been processed by the given Continuation. So, what we’re doing here is chaining calls to continueWith() to form a pipeline of operations that culminates in a final Task that waits for each stage to complete before completing.
This chain of operations could be problematic if these they have to deal with large Strings, so let's modify it to do all the processing on other threads so we don't block up the main thread:
Executor executor = ... // you decide! Task<String> playlist = Tasks.call(executor, new CarlyCallable()) .continueWith(executor, new SeparateWays()) .continueWith(executor, new AllShookUp()) .continueWith(executor, new ComeTogether()); playlist.addOnSuccessListener(executor, new OnSuccessListener() { @Override public void onSuccess(String message) { // Do something with the output of this playlist! } });
Now, the Callable, all of the Continuations, and the final Task listener will each run on some thread determined by the Executor, freeing up the main thread to deal with UI stuff while this happens. It should be totally jank-free.
At first blush, it could seem a bit foolish to separate all these operations into all the different classes. You could just as easily write this as a few lines in a single method that do only what's required. So, keep in mind that this is a simplified example intended to highlight how Tasks can work for you. The benefit of chaining of Tasks and Continuations (even for relatively simple functions) becomes more evident when you consider the following:
Practically speaking, you're more likely to use Task continuations to perform a series of modular chain of filter, map, and reduce functions on a set of data, and keep those units of work off the main thread, if the collections can be large. But, I had fun with music theme here!
One last thing to know about Continuations. If a runtime exception is thrown during processing at any stage along the way, that exception will normally propagate all the way down to the failure listeners on the final Task in the chain. You can check for this yourself in any Continuation by asking the input Task if it completed successfully with the isSuccessful() method. Or, you can just blindly call getResult() (as is the case in the above samples), and if there was previously a failure, it will get re-thrown and automatically end up in the next Continuation. The listeners on the final Task in the chain should always check for failure, though, if failure is an option.
So, for example, if the CarlyCallable in the above chain returned null, that would cause the SeparateWays continuation to throw a NullPointerException, which would propagate to the end of the Task. And if we had an OnFailureListener registered, that would get invoked with the same exception instance.
What's the most efficient way, with the above chain, of finding out the number of words in the original string, without modifying any of the processing components? Take a moment to think about it before reading on!
The answer is probably more simple than you'd imagine. The most obvious solution is to count the number of words in the final output string, since their order only got randomized. But there is one more trick. Each call to continueWith() returns a new Task instance, but those are all invisible here because we used a chaining syntax to assemble them into the final Task. So you can intercept any of those those tasks and add another listener to it, in addition to the next Continuation:
Task<List<String>> split_task = Tasks.call(new CarlyCallable()) .continueWith(executor, new SeparateWays()); split_task = .continueWith(executor, new AllShookUp()) .continueWith(executor, new ComeTogether()); split_task.addOnCompleteListener(executor, new OnCompleteListener<List<String>>() { @Override public void onComplete(@NonNull Task<List<String>> task) { // Find the number of words just by checking the size of the List int size = task.getResult().size(); } }); playlist.addOnCompleteListener( /* as before... */ );
When a Task finishes, it will trigger both of the Continuations on it, as well as all of the added listeners. All we've done here is intercept the Task that captures the output of the SeparateWays continuation, and listen to the output of that directly, without affecting the chain of continuations. With this intercepted task, we only have to call size() on the List to get the count.
All joking aside, the Task API makes it relatively easy for you to express and execute a sequential pipeline of processing in a modular fashion, while giving you the ability to specify which Executor is used at each stage in the process. You can do this /with or without/ Firebase integrated into your app, using your own Tasks or those that come from Firebase APIs. For the next and final part to this series, we'll look at how Tasks can be used in parallel to kick off multiple units of work simultaneously.
As usual, if you have any questions, consider using Twitter with the #AskFirebase hashtag or the firebase-talk Google Group. We also have a dedicated Firebase Slack channel. And you can follow me @CodingDoug on Twitter to get notified of the next post in this series.
Lastly, if you're wondering about all the songs I referenced in this post, you can find them here:
This is our third post in the Pirate Metrics with Firebase series. In the first post, we gave an overview of what Pirate Metrics are and why they’re important. In the second, we showed how you can use Firebase to improve your acquisition strategy.
Once you acquire a user, your main goal is to make them use your product. Users often install an app but never get hooked. They would have the app around for a day or two, if you’re lucky, before either forgetting about it or, worse, uninstalling it. All that effort you put into your acquisition goes down the drain.
The first few days are, hence, crucial. Through your data, you want to find a pattern to determine at what point is a user activated, and look at ways to get more users past that point. Examples can be the number of friends on a social networking application or the number of levels crossed in a video game. Devising the right “activation strategy” always involves a lot of experimentation.
To carry out these experiments, we have just the right tool for you - Firebase’s Remote Configuration. Remote Config allows you to set certain key/value pairs on the server, and use them to vary the experience inside of your application. These values when updated on the Firebase console reflect inside of your own application, allowing you to change the experience for users without releasing an update.
If you use this capability of Remote Config, and set values using the “random percentile” targeting, you essentially have an A/B test setup. You can then see the impact on your analytics, and change these values dynamically on the server itself, increasing the rollout for experiments that have proven to work.. It makes for a great solution for A/B testing.
To optimize your testing, we recommend first defining the data points you want to improve (such as an increase in users signing up on the first app open). Then, ideate on the experiments you want to run to improve these data points. These might be experiments that track the impact of different tutorials or signup methods for apps or difficulty settings for initial levels in a game that can ultimately improve your activation percentage.
Today we're excited to announce the availability of HTTP/2 on Firebase Hosting. HTTP/2 is a new version of the HTTP protocol that is already supported by 77% of browsers worldwide. It offers some exciting advantages for web developers:
Taken together, these add up to significant performance advantages and lots of opportunity to make your web applications load faster on mobile devices with slow connections.
HTTP/2 is currently enabled for all *.firebaseapp.com traffic as well as newly-provisioned custom domains. If you already have a custom domain on Firebase, see Custom Domain Migration below.
*.firebaseapp.com
To utilize HTTP/2 on Firebase Hosting, you don't have to do anything! It will automatically be served if the user's browser supports it. However, there are some best practices you should keep in mind if you want to optimize for HTTP/2:
The above guidelines aren't hard and fast rules -- as with any performance optimization, you should experiment with different combinations of optimizations to see which ones deliver the best result for your app's specific needs.
Firebase Hosting has experimental support for HTTP/2 server push using Link headers. Server push allows a server to automatically send the contents for additional resources when an initial request is made. The most common use for server push is to send down associated assets (like JavaScript or CSS files) when a page is loaded.
To configure server push on Firebase Hosting, you need to add the Link header to your firebase.json configuration like so:
firebase.json
{ "hosting": { "headers": [ { "source": "/", "headers": [{"key": "Link", "value": "</js/app.js>;rel=preload;as=script,</css/app.css>;rel=preload;as=style"}] }, { "source": "/users/*", "headers": [{"key": "Link", "value": "</js/app.js>;rel=preload;as=script,</css/app.css>;rel=preload;as=style;nopush,</css/users.css>;rel=preload;as=style"}] } ] } }
Here we are using server push to preload /js/app.js and /css/app.css on the root path, and additionally /css/users.css on any path matching /users/*. You can use the nopush directive (like on app.css in the second example) to preload the asset without HTTP/2 push for files that are likely to already be in the browser cache.
/js/app.js
/css/app.css
/css/users.css
/users/*
app.css
It's still early days for server push, and there are a few things to keep in mind:
;nopush
We're excited about HTTP/2's potential to improve that first-load experience, and we're still exploring additional ways to make server push simple, reliable, and effective for your site.
With our migration to HTTP/2 we're also moving to Server Name Indication (SNI) for our SSL certificates. SNI enables us to continue to scale our infrastructure more reliably and is supported by nearly 98% of browsers worldwide. Because this change has the possibility of affecting user traffic, we are not automatically switching over existing custom domains until December 2016.
If you have a custom domain on Firebase Hosting from before August 11, 2016, you will need to update your DNS records to take advantage of HTTP/2. You can check if you're already on SNI by running dig <your-site>.firebaseapp.com. If you see s-sni.firebaseapp.com in the result, your site is already migrated.
dig <your-site>.firebaseapp.com
s-sni.firebaseapp.com
To migrate if you're using a CNAME, update your DNS to point to s-sni.firebaseapp.com. If you're using A records, update them to:
151.101.1.195 151.101.65.195
Once you've changed over your DNS and it's had the chance to propagate, your site will be live with HTTP/2! We will be transitioning all Firebase Hosting traffic to HTTP/2 and SNI by the end of the year, so please reach out to support if you're worried about how SNI might affect your users.
Our goal with Firebase Hosting is to bring the best practices of Progressive Web App development within reach of everyone. HTTP/2 is another step along that path, and we're excited to see what you build with it!
