51

I'm trying to create an app that lets users log routes (locations/GPS). To ensure locations are logged even when the screen is off, I have created a foreground service for the location logging. I store the locations in a Room Database which is injected into my service using Dagger2.

However, this service is killed by Android which is, of course, not good. I could subscribe to low memory warnings but that doesn't solve the underlying problem of my service getting killed after ~30 minutes on a modern high-end phone running Android 8.0

I have created a minimal project with only a "Hello world" activity and the service: https://github.com/RandomStuffAndCode/AndroidForegroundService

The service is started in my Application class, and route logging is started through a Binder:

// Application
@Override
public void onCreate() {
    super.onCreate();
    mComponent = DaggerAppComponent.builder()
            .appModule(new AppModule(this))
            .build();

    Intent startBackgroundIntent = new Intent();
    startBackgroundIntent.setClass(this, LocationService.class);
    startService(startBackgroundIntent);
}

// Binding activity
bindService(new Intent(this, LocationService.class), mConnection, Context.BIND_AUTO_CREATE | Context.BIND_IMPORTANT);
// mConnection starts the route logging through `Binder` once connected. The binder calls startForeground()

I probably don't need the BIND_AUTO_CREATE flag, I've been testing different flags in an attempt to not get my service killed - no luck so far.

Using the profiler it does not seem like I have any memory leaks, memory usage is stable at ~35mb:

profiler

Using adb shell dumpsys activity processes > tmp.txt i can confirm that foregroundServices=true and my service is listed 8th in the LRU list:

Proc # 3: prcp F/S/FGS trm: 0 31592:com.example.foregroundserviceexample/u0a93 (fg-service)

It seems like it is not possible to create a foreground service that you can trust to not get killed. So what can we do? Well...

  1. Put the service in a separate process, in an attempt to let Android kill the UI/Activities while leaving the service alone. Would probably help, but doesn't seem like a guarantee
  2. Persist everything in the service in e.g. a Room database. Every variable, every custom class, every time any of the changes and then start the service with START_STICKY. This seems kind of wasteful and doesn't lead to very beautiful code, but it would probably work... somewhat. Depending on how long it takes for Android to re-create the service after killing it, a large portion of locations may be lost.

Is this really the current state of doing stuff in the background on Android? Isn't there a better way?

EDIT: Whitelisting the app for battery optimization (disabling it) does not stop my service from being killed

EDIT: Using Context.startForegroundService() to start the service does not improve the situation

EDIT: So this indeed only occurs on some devices, but it occurs consistently on them. I guess you have to make a choice of either not supporting a huge number of users or write really ugly code. Awesome.

user1202032
  • 1,430
  • 3
  • 15
  • 36
  • "my service getting killed after ~30 minutes" -- how are you determining this? – CommonsWare Apr 03 '18 at 19:45
  • 2
    The notification for the foreground service is removed after that time and locations stop being inserted to the database. According to @ianhanniballake it seems like you *can* get location data frequently and reliably if you're using a foreground service since its supposed to be unaffected by doze – user1202032 Apr 03 '18 at 19:48
  • 1
    Have you tried starting the service using the new `Context.startForegroundService()`? The service itself still must call `startForeground()`, but maybe the way the service is initially started makes a difference. – StackOverthrow Apr 03 '18 at 19:56
  • Sorry -- I realized that moments after posting the comment, and I wasn't fast enough to delete that portion. Next question: is this device one that shipped with 8.0? Or did it ship with something older and got upgraded by the manufacturer? I ask because it's possible that this is a manufacturer thing more so than an Android thing. Also, does this behavior persist if you (as a user) add your app to the battery optimization whitelist? – CommonsWare Apr 03 '18 at 19:57
  • @TKK I have not tried that, but I will now, thanks for the suggestion. I would prefer to only have the service in the foreground when a route is actively being recorded, not all the time, but its worth a shot! – user1202032 Apr 03 '18 at 19:59
  • @CommonsWare it was upgraded to 8.0. It could very well be a manufacturer thing but ideally i'd like to support all devices running a recent version of android. I will test battery optimization whitelist even though i feel like it should not make a difference for a foreground service. Ill be back with the results – user1202032 Apr 03 '18 at 20:01
  • Related, but it sounds like you're already doing everything in the accepted answer: https://stackoverflow.com/questions/6645193/foreground-service-being-killed-by-android – StackOverthrow Apr 03 '18 at 20:05
  • I have updated the question after testing `Context.startForegroundService()` and battery optimization whitelisting. Spoiler alert: It didnt work :/ – user1202032 Apr 04 '18 at 10:19
  • Try disabling battery save mode. – leonardkraemer Apr 07 '18 at 16:06
  • @leoderprofi Do you mean battery optimization whitelisting? The last comment just before yours i wrote that i already tried that – user1202032 Apr 07 '18 at 16:12
  • I had the same issue on my phone with my running app, and the solution was to disable battery saving mode. Could also be called 'high power mode' or something else. Also try to reproduce the issue on another phone, might be vendor specific. – leonardkraemer Apr 07 '18 at 16:16
  • Have you tried using `JobScheduler` in android which is alternative to service and best for a scenario like yours here is an example https://stackoverflow.com/a/49533086/4762767 – Akshay Katariya Apr 12 '18 at 08:28
  • I think `JobScheduler` is not a good alternative for this. I do not want intelligent scheduling of my job, i was it to be "ON" and running continuously – user1202032 Apr 14 '18 at 16:44
  • On what devices do you have the problem? – leonardkraemer Apr 17 '18 at 10:24
  • @leoderprofi OnePlus but im assuming it happens on other somewhat sketchy OEMs such as Huawei and Xiaomi – user1202032 Apr 17 '18 at 15:52

3 Answers3

40

A service started by startForeground belongs to the second most important group visible process:

  1. A visible process is doing work that the user is currently aware of, so killing it would have a noticeable negative impact on the user experience. A process is considered visible in the following conditions:

  2. It is running an Activity that is visible to the user on-screen but not in the foreground (its onPause() method has been called). This may occur, for example, if the foreground Activity is displayed as a dialog that allows the previous Activity to be seen behind it.

  3. It has a Service that is running as a foreground service, through Service.startForeground() (which is asking the system to treat the service as something the user is aware of, or essentially visible to them).

  4. It is hosting a service that the system is using for a particular feature that the user is aware, such as a live wallpaper, input method service, etc.

The number of these processes running in the system is less bounded than foreground processes, but still relatively controlled. These processes are considered extremely important and will not be killed unless doing so is required to keep all foreground processes running.

That being said, you can never be sure that your service is not killed at any time. E.g. memory pressure, low battery etc. See who-lives-and-who-dies.


For how to handle it, basically you answered the question yourself. The way to go is START_STICKY:

For started services, there are two additional major modes of operation they can decide to run in, depending on the value they return from onStartCommand(): START_STICKY is used for services that are explicitly started and stopped as needed, while START_NOT_STICKY or START_REDELIVER_INTENT are used for services that should only remain running while processing any commands sent to them. See the linked documentation for more detail on the semantics.

As a general guideline you should do as little as possible in the background (ore foreground) service, i.e. only do the location tracking and keep everything else in your foreground activity. Only the tracking should require very little configuration an can be loaded quickly. Also the smaller your service is the less likely it is to be killed. Your activity will be restored by the system in the state that is was before it went into background, as long as it is not killed as well. A "cold-start" of the foreground activity on the other hand should not be a problem.
I don't consider that as ugly, because this guarantees that the phone always provides the best experience to the user. This is the most important thing it has to do. That some devices close services after 30 minutes (possibly without user interaction) is unfortunate.

So, as you stated, you have to

Persist everything in the service in e.g. a Room database. Every variable, every custom class, every time any of them changes and then start the service with START_STICKY.

See creating a never ending service

Implicit question:

Depending on how long it takes for Android to re-create the service after killing it, a large portion of locations may be lost.

This usually takes only a really short time. Especially because you can use the Fused Location Provider Api for the location updates, which is an independent system service and very unlikely to be killed. So it mainly depends on the time you need to recreate the service in onStartCommand.

Also take note that from Android 8.0 onwards you need to use a forground service because of the background location limits.


Edit: As recently covered in the news: Some manufacturers may give you a hard time to keep your service running. The site https://dontkillmyapp.com/ keeps track of the manufacturers and possible mitigations for your device. Oneplus is currently (29.01.19) one of the worst offenders.

When releasing their 1+5 and 1+6 phones, OnePlus introduced one of the most severe background limits on the market to date, dwarfing even those performed by Xiaomi or Huawei. Not only did users need to enable extra settings to make their apps work properly, but those settings even get reset with firmware update so that apps break again and users are required to re-enable those settings on a regular basis.

Solution for users

Turn off System Settings > Apps > Gear Icon > Special Access > Battery Optimization.

sadly there is

No known solution on the developer end

Pavel Chuchuva
  • 22,633
  • 10
  • 99
  • 115
leonardkraemer
  • 6,573
  • 1
  • 31
  • 54
  • Good post thank you. Regarding Fused Provider API being killed.... I dont expect that system to be killed, but my reference to the Fused Provider Listener is in my service so when my service is killed i cannot receive updates until the service is restarted, and my FusedProviderListener is re-created and re-subscribed – user1202032 Apr 12 '18 at 12:22
  • That's true. You can use the [last known location](https://developer.android.com/training/location/retrieve-current.html) to try to fill in the blank (a bit). Unfortunately google does not allow programmatic access to the users [location history](https://www.google.de/maps/timeline?authuser=0). – leonardkraemer Apr 12 '18 at 12:39
  • 6
    Update: `START_STICKY` is completely unreliable. Sometimes the service is restarted but often its not... its killed and never ever restarted. The "Neverending service" link you posed is not going to work consistently as `onDestroy()` is not called consistently (by design) – user1202032 Apr 16 '18 at 18:04
  • Good to know, do you know the exact circumstance when it is not restarted? Probably it's vendor specific though. As M D P suggested you can use `AlarmManager` to make sure the service is triggered/restarted at a constant interval. It seems to be the critical part, not sure about the other stuff. – leonardkraemer Apr 16 '18 at 20:35
  • @user1202032 I remember from the 4.x days that on some devices, a sticky service would restart within 3-5 seconds, while on others it would take 20-30 seconds. How long did you wait to decide it's "never" restarted? – StackOverthrow Apr 16 '18 at 23:18
  • @leoderprofi I've replied to him, his approach will not work. – user1202032 Apr 17 '18 at 06:52
  • 1
    @TKK I waited more than 1 hour – user1202032 Apr 17 '18 at 06:52
  • 1
    I guess, but considering no solution that works exists, this came close. Lots of good info – user1202032 May 01 '18 at 21:41
  • @Steve M Do you mean in the last bit? Yes it is, some manufacturers even stop foreground services, regardless of any battery exemption lists or settings. – leonardkraemer Feb 15 '19 at 15:28
  • @leonardkraemer no I mean the first quoted block is not relevant since it's not about foreground services and not a good answer to the OP's question – Steve M Feb 15 '19 at 15:33
  • @SteveM you were right, that part came from trying to provide a quick answer. I assumed `startForeground` had not been called. I updated the answer in a way, that fits the question better. – leonardkraemer Feb 15 '19 at 15:57
13

I know it's late but this may help some one. I too faced the same problem of keeping the foreground service alive without being killed by OS from different manufacturers. Most of the Chinese manufacturer's OS kills the foreground service even if it is added to the exception lists (Battery, Cleaner, etc) and allowed to auto start.

I found this link which solved me this long time problem of keeping the service alive.

All you have to do is run your foreground service in a seperate process. That's it.

You can do that by adding android:process to your service in your AndroidManifest.xml.

For Example:

<service android:name=".YourService"
        android:process=":yourProcessName" />

You can refer the docs to know more about android:process

EDIT: SharedPreferences won't work across multiple processes. In this case, you must go for IPC (Inter Process Communication) methods or you can use ContentProviders for storing and accessing your data to be used across processes. Referred from docs.

Joshua
  • 1,167
  • 1
  • 14
  • 32
  • 1
    Thank you, my foreground service is no longer terminated on my Redmi Note 6 Pro – k8C Mar 24 '19 at 04:48
  • @KhoaChuAnh It will work like a charm, until you don't use the same `SharedPreferences` across two processes. – Joshua Mar 26 '19 at 09:00
  • I use SharedPreferences in my service with the android:process attribute without problems. As I have read somewhere on the Internet, you can still use BroadcastReceiver, ResultReceiver or Messenger to communicate between activity and service running in another process – k8C Mar 27 '19 at 10:02
  • 1
    Yes, `Messenger` and `ResultReceiver` are some of the IPC methods. But those doesn't seem to work in the background. `BroadcastReceiver` works in background but from oreo and up, it has some limitations like only few events per hour will be received in the background (when it's in the background for a long time) even if it's an explicit broadcast (That's what I read somewhere). – Joshua Mar 27 '19 at 11:37
  • 1
    What would be the ramifications of running in a separate process? – IgorGanapolsky Apr 13 '20 at 11:42
  • Among all the threads this was gold. I'd like to add that this allow the service not to be cleared by the "sweep away background apps" action. – Sandor Mezil Dec 19 '20 at 09:36
-1

I suggest you use these: AlarmManager, PowerManager, WakeLock, Thread, WakefulBroadcastReceiver, Handler, Looper

I assume you already are using those "separate process" and other tweaks too.

So in your Application class:

MyApp.java:

import android.app.AlarmManager;
import android.app.Application;
import android.app.PendingIntent;
import android.content.Context;
import android.content.Intent;
import android.os.PowerManager;
import android.util.Log; 

public final class MyApp extends Application{

public static PendingIntent pendingIntent = null;
public static Thread                infiniteRunningThread;
public static PowerManager          pm;
public static PowerManager.WakeLock wl;


@Override
public void onCreate(){
    try{
        Thread.setDefaultUncaughtExceptionHandler(
                (thread, e)->restartApp(this, "MyApp uncaughtException:", e));
    }catch(SecurityException e){
        restartApp(this, "MyApp uncaughtException SecurityException", e);
        e.printStackTrace();
    }
    pm = (PowerManager) getSystemService(Context.POWER_SERVICE);
    if(pm != null){
        wl = pm.newWakeLock(PowerManager.PARTIAL_WAKE_LOCK, "TAG");
        wl.acquire(10 * 60 * 1000L /*10 minutes*/);
    }

    infiniteRunningThread = new Thread();

    super.onCreate();
}

public static void restartApp(Context ctx, String callerName, Throwable e){
    Log.w("TAG", "restartApp called from " + callerName);
    wl.release();
    if(pendingIntent == null){
        pendingIntent =
                PendingIntent.getActivity(ctx, 0,
                                          new Intent(ctx, ActivityMain.class), 0);
    }
    AlarmManager mgr = (AlarmManager) ctx.getSystemService(Context.ALARM_SERVICE);
    if(mgr != null){
        mgr.set(AlarmManager.RTC_WAKEUP,
                System.currentTimeMillis() + 10, pendingIntent);
    }
    if(e != null){
        e.printStackTrace();
    }
    System.exit(2);
}
}

And then in your service:

ServiceTrackerTest.java:

import android.app.Service;
import android.content.Context;
import android.content.Intent;
import android.graphics.BitmapFactory;
import android.os.Handler;
import android.os.IBinder;
import android.os.Looper;
import android.os.PowerManager;
import android.support.v4.app.NotificationCompat;
import android.support.v4.content.WakefulBroadcastReceiver;

public class ServiceTrackerTest extends Service{

private static final int SERVICE_ID = 2018;
private static PowerManager.WakeLock wl;

@Override
public IBinder onBind(Intent intent){
    return null;
}

@Override
public void onCreate(){
    super.onCreate();
    try{
        Thread.setDefaultUncaughtExceptionHandler(
                (thread, e)->MyApp.restartApp(this,
                                              "called from ServiceTracker onCreate "
                                              + "uncaughtException:", e));
    }catch(SecurityException e){
        MyApp.restartApp(this,
                         "called from ServiceTracker onCreate uncaughtException "
                         + "SecurityException", e);
        e.printStackTrace();
    }
    PowerManager pm = (PowerManager) getSystemService(Context.POWER_SERVICE);
    if(pm != null){
        wl = pm.newWakeLock(PowerManager.PARTIAL_WAKE_LOCK, "TAG");
        wl.acquire(10 * 60 * 1000L /*10 minutes*/);
    }


    Handler h = new Handler();
    h.postDelayed(()->{

        MyApp.infiniteRunningThread = new Thread(()->{
            try{
                Thread.setDefaultUncaughtExceptionHandler(
                        (thread, e)->MyApp.restartApp(this,
                                                      "called from ServiceTracker onCreate "
                                                      + "uncaughtException "
                                                      + "infiniteRunningThread:", e));
            }catch(SecurityException e){
                MyApp.restartApp(this,
                                 "called from ServiceTracker onCreate uncaughtException "
                                 + "SecurityException "
                                 + "infiniteRunningThread", e);
                e.printStackTrace();
            }

            Looper.prepare();
            infiniteRunning();
            Looper.loop();
        });
        MyApp.infiniteRunningThread.start();
    }, 5000);
}

@Override
public void onDestroy(){
    wl.release();
    MyApp.restartApp(this, "ServiceTracker onDestroy", null);
}

@SuppressWarnings("deprecation")
@Override
public int onStartCommand(Intent intent, int flags, int startId){
    if(intent != null){
        try{
            WakefulBroadcastReceiver.completeWakefulIntent(intent);
        }catch(Exception e){
            e.printStackTrace();
        }
    }
    startForeground(SERVICE_ID, getNotificationBuilder().build());
    return START_STICKY;
}


private void infiniteRunning(){
    //do your stuff here
    Handler h = new Handler();
    h.postDelayed(this::infiniteRunning, 300000);//5 minutes interval
}

@SuppressWarnings("deprecation")
private NotificationCompat.Builder getNotificationBuilder(){
    return new NotificationCompat.Builder(this)
                   .setContentIntent(MyApp.pendingIntent)
                   .setContentText(getString(R.string.notification_text))
                   .setContentTitle(getString(R.string.app_name))
                   .setLargeIcon(BitmapFactory.decodeResource(getResources(),
                                                              R.drawable.ic_launcher))
                   .setSmallIcon(R.drawable.ic_stat_tracking_service);
}

}

Ignore "Deprecation" and stuff, use them when you have no choice. I think the code is clear and doesn't need to be explained. It's just about workaround suggestions and solutions.

M D P
  • 4,525
  • 2
  • 23
  • 35
  • As far as I can see this restarts the service after 5 to 10 minutes. Is this faster or more reliable than the system wise restart for `START_STICKY`? – leonardkraemer Apr 12 '18 at 12:45
  • Also since lolipop the alarm manager can not be abused in this way see https://stackoverflow.com/questions/29914724/scheduling-alarm-for-every-second-in-android-5-1 – leonardkraemer Apr 12 '18 at 12:59
  • @leoderprofi yes, `START_STICKY` is just a joke, and no it doesn't restart after interval, I copied from my own app that is being used for over 2 years now, it saves last best location on every interval inside `infiniteRunning()` method, it only restarts when an error happens. and don't worry about lollipop, I've set `minSdkVersion` to 14 and `targetSdkVersion` and `compileSdkVersion` to 27. and it's working just fine even on android N. – M D P Apr 12 '18 at 16:44
  • It certainly should be reliable since people are getting paid based on the location history my app reports to server. so yes it's more reliable. – M D P Apr 12 '18 at 16:52
  • @M D P that is not how `minSdkVersion` works, it doesn't change how a certain version of android behaves, just on what devices the app runs. Can you elaborate what mechanism restarts the app, when it was closed due to memory pressure and what mechanisms are used to prevent closing? – leonardkraemer Apr 12 '18 at 16:59
  • Looks like the limitation to 1 minute is only for repeating alarms. Considering this I think the `AlarmManager` approach can work. – leonardkraemer Apr 12 '18 at 17:10
  • @leoderprofi exactly, `AlarmManager` is an important key, no app can stop your service, even when the memory is low, system will hold that until it can be run, then runs the task you set to `AlarmManager`, only user can force stop it, in that case if you are building this service for a company, you can make your service `Administrator`, then if user removes admin or force stops the service, you get notified and can report it to the company's server. – M D P Apr 12 '18 at 18:27
  • @MDP Thanks for your input but this seems kind of pointless TBH. Surely your `infiniteRunningThread` is killed along with the rest of your apps process...? – user1202032 Apr 16 '18 at 18:00
  • @user1202032 idk what exactly you want, but it doesn't matter if gets killed, because of the `AlarmManager`, it always will be restarted unless user force-stop it. – M D P Apr 16 '18 at 23:19
  • @MDP No it will not. Its not a recurring alarm so it will only restart if restartApp() is called. If the process is killed, so is `infiniteRunningThread` which means restartApp() is never called and the service is never restarted. – user1202032 Apr 17 '18 at 06:32