I'm investigating a memory leak (or leaks) which seem to be core to our Forms app.
We're using FreshMvvm, which previously had a pretty bad leak due to Pages/PageModels not being garbage collected. Michael Ridland (FreshMvvm) author) fixed that, which is great.
However, we're still seeing the figure returned by GC.GetTotalMemory() steadily increase. It does decrease, but the general trend is upwards.
I've stripped things right back so that I've just got a plain Page which pushes another page and we can then pop back to the first.
Doing this, I'm seeing GC.GetTotalMemory() generally increase by 2K - 4K every cycle. This is after forcibly calling GC.Collect();
We're running Xamarin.Forms 2.3.4.270 - I know this is quite old, but we really don't want to upgrade as we don't want to introduce 'new Forms' bugs! However, I have tried experimentally upgrading to 2.5.0.121934, and I see the same behaviour (as well as aspects of the app stop working with this version).
Short of using Profiler, are there any investigative techniques I can use to find the culprit?
My next step will be to strip more and more out!
I've encountered a similar issue in the past with Xamarin.Forms memory leaks and to help figure what views/pages were the culprit this class was created:
public static class Refs
{
private static readonly List<WeakReference> _refs = new List<WeakReference>();
public static readonly BindableProperty IsWatchedProperty = BindableProperty.CreateAttached("IsWatched", typeof(bool), typeof(Refs), false, propertyChanged: OnIsWatchedChanged);
public static bool GetIsWatched(BindableObject obj)
{
return (bool)obj.GetValue(IsWatchedProperty);
}
public static void SetIsWatched(BindableObject obj, bool value)
{
obj.SetValue(IsWatchedProperty, value);
}
private static void OnIsWatchedChanged(BindableObject bindable, object oldValue, object newValue)
{
AddRef(bindable);
}
public async static void AddRef(object p)
{
GC.Collect();
await Task.Delay(100);
GC.Collect();
_refs.Add(new WeakReference(p));
foreach (var #ref in _refs)
{
if (#ref.IsAlive)
{
var obj = #ref.Target;
Debug.WriteLine("IsAlive: " + obj.GetType().Name);
}
else
{
Debug.WriteLine("IsAlive: False");
}
}
Debug.WriteLine("---------------");
}
}
Then to use it you can use the IsWatched property in your pages xaml or in code behind you can just do Refs.AddRef(this); after InitializeComponent(); and just navigate to and from the page several times. If it is being gc'ed the console should print IsAlive: False if not it will print out the type.
Most often the leaks we had were fixed by clearing the BindingContext and/or setting some things to null when leaving the page.
protected override void OnParentSet()
{
base.OnParentSet();
if (Parent == null)
{
BindingContext = null;
}
}
Related
I have the following Reporting code:
public class Reporting {
private ExtentHtmlReporter extentHtmlReporter;
private static ThreadLocal<ExtentReports> extentReports = new ThreadLocal<>();
private static ThreadLocal<ExtentTest> extentTest = new ThreadLocal<>();
public synchronized ExtentTest createInstanceReport(String testCaseName) {
System.out.println(extentReports.get());
new File(Constants.userDir + "/Reports/").mkdirs();
// To generate report with name
extentHtmlReporter = new ExtentHtmlReporter(
Constants.userDir + "/Reports/" +
"ExecutionReport_" + new SimpleDateFormat(
Constants.date).format(new Date()) + ".html");
// Setting Document Title
extentHtmlReporter.config().setDocumentTitle("Demo");
// Setting Report Name
extentHtmlReporter.config().setReportName("Demo Automation");
// Setting Theme
extentHtmlReporter.config().setTheme(Theme.STANDARD);
// Setting Chart location
extentHtmlReporter.config().setTestViewChartLocation(ChartLocation.TOP);
// Setting Chart visibility
extentHtmlReporter.config().setChartVisibilityOnOpen(false);
// Setting Time stamp
extentHtmlReporter.config().setTimeStampFormat("yyyy-MM-dd HH:mm:ss");
// Setting append exist as true
extentHtmlReporter.setAppendExisting(true);
ExtentReports extentReports = new ExtentReports();
extentReports.attachReporter(extentHtmlReporter);
// Setting system info
extentReports.setSystemInfo("Name",
BaseTest.prop.getProperty(Constants.testerName));
extentReports.setSystemInfo("Environment",
BaseTest.prop.getProperty(Constants.environment));
extentReports.setSystemInfo("Browser",
BaseTest.prop.getProperty(Constants.browser));
Reporting.extentReports.set(extentReports); // Instead of using here extentReport thread like this, Can anyone suggest to use it directly
// Add test case name in report
ExtentTest extentTest = Reporting.extentTest.get();
extentTest = Reporting.extentReports.get().createTest(testCaseName);
Reporting.extentTest.set(extentTest);
// Assigning categories
extentTest.assignCategory(MultiFunction.getProp()
.getProperty(Constants.browser));
System.out.println(Reporting.extentReports.get());
System.out.println(Reporting.extentTest.get());
return extentTest;
}
public synchronized ExtentTest getExtentTest() {
return extentTest.get();
}
public synchronized ExtentReports getInstanceReport() {
return extentReports.get();
}
public synchronized void remove() {
extentReports.remove();
extentTest.remove();
}
}
I was trying parallel testing using TestNG (and will have to use Selenium grid and sauce in future). I execute 2 test cases then only one test case result is added in the report.
I have isolated the extentTest, extentReporter and WebDriver instances using threadPool.
Tried below with extentHtmlReporter instance:
1) Tried to make it static(no luck)
2) Tried to make it local (the same behaviour, getting only 1 test case result)
3) Tried as a non-static global variable ( no luck)
Could you suggest how to solve the above issue?
Please note: Only one report is generated. But when I tried to run parallel test cases in debug mode reports are generated for both the test case. I think because one test case gets over its killing some instance (when running in non-debug mode)
Also, I want to redesign the following place in my code:
For extentRpeort, I am using:
Reporting.extentReports.set(extentReports);
To add extentReport instance to my extentReport Thread.
Instead of adding like this I want to use it directly so as to reduce line of code.
If I understand correctly you have to generate Report from all executed TestNG cases.
However, from code which you shared, it is very visible that you will have some trouble with it. You are making a few critical mistakes and result are obvious:
For generating reports with TestNG I will suggest grabbing information about test execution from TestNG listener. Something like:
public final class TestNGListener extends TestListenerAdapter implements IInvokedMethodListener, ISuiteListener {
#Override
public void onStart(ITestContext context) {
Logger.info(buildMessage(Logger.PREFIX_TEST_STARTED, context.getName()));
}
#Override
public void onFinish(ITestContext context) {
Logger.info(buildMessage(Logger.PREFIX_TEST_FINISHED, context.getName()));
}
#Override
public void onTestStart(ITestResult result) {
Logger.info(buildMessage(Logger.PREFIX_METHOD_STARTED, getMethodName(result)));
}
#Override
public void onTestSuccess(ITestResult result) {
Logger.info(buildMessage(Logger.PREFIX_METHOD_SUCCESS, getMethodName(result)));
processTestResult(result);
}
#Override
public void onTestFailure(ITestResult result) {
Logger.info(buildMessage(Logger.PREFIX_METHOD_FAILED, getMethodName(result)));
}
You can't do everything in one method! You broke Single Responsibility Principle. Your createInstanceReport() is doing all jobs (setting report details, set system info, attach an executed test case to report) at one place. You have to redesign this logic to some logical separate operations. After redesigning your problem with the next line:
Reporting.extentReports.set(extentReports)
Could successfully disappear.
You have to consider a case, why you need to use exactly Extent, Reports Version 3. TestNG has test reports from the box. They are poor but they are presented out of the box. If you want just to improve it a little bit you could use ReportNG over TestNG.
It is quite easy to configure: Configuring ReportNG with TestNG for HTML Reports.
It isn't maintained, but it makes TestNG reports really eye candy and understandable.
Anyway, my suggestion is to use TestNGListener for getting info about test cases execution. And read more about good programming practice.
Work with TestNG/jUnit (or other runner framework that you are using) listener, here is a good example how to do it.
Do not put everything in a single class.
https://www.swtestacademy.com/extent-reports-version-3-reporting-testng/
The issue was with the flushing of extent report instance.
I was using ThreadLocal for storing extent report instance and was flushing the wrong instance.
Looking at the docs for Parcel#writeMap(Map), it clearly states that "The Map keys must be String objects", however looking at the code there seems to be absolutely no reason for this aside from completely unnecessary casting:
/* package */ void readMapInternal(Map outVal, int N,
ClassLoader loader) {
while (N > 0) {
Object key = readValue(loader);
Object value = readValue(loader);
outVal.put(key, value);
N--;
}
}
/* package */ void writeMapInternal(Map<String,Object> val) {
if (val == null) {
writeInt(-1);
return;
}
Set<Map.Entry<String,Object>> entries = val.entrySet();
writeInt(entries.size());
for (Map.Entry<String,Object> e : entries) {
writeValue(e.getKey());
writeValue(e.getValue());
}
}
Is there anything I am missing here? It's important for me to know since I am writing a library that heavily uses this class and it seems like a massive restriction for no reason.
EDIT: Similarly, when writing SparseArray objects, this class forces you
to pass in only SparseArray<Object>:
public final void writeSparseArray(SparseArray<Object> val)
This is another restriction that makes no sense. Are these Android API bugs?
EDIT: This answer seems to also imply there is a bug
My Android app has a simple "loader" NativeActivity with a very simple android_main() which only loads a different shared object and passes control to it:
typedef void (*Tandroid_main)( android_app*);
void android_main( android_app* state )
{
void* glib = dlopen("libmain.so", RTLD_NOW);
void* fmain = dlsym(glib, "android_main");
Tandroid_main libmain = (Tandroid_main)fmain;
libmain(state)
}
This works well.. about half of the times. Other times it crashes since dlopen() fails and return NULL with errno=2 (No such file).
Due to the strange inconsistency of this occurrence I suspected a timing issue and indeed, adding a sleep(1) before dlopen() stopped it from happening. Something more robust than sleep(1) would be just trying it in a loop:
int count = 0;
void* glib = dlopen(soName, RTLD_NOW);
while(glib == NULL) {
sched_yield();
++count;
glib = dlopen(soName, RTLD_NOW);
}
The count I'm getting from this loop is usually something in the range of 10-70 on my device. But this is a hackish ugly solution.
What is really going on here? How come I can only load other shared objects only slightly after the NativeActivity starts? Is there a better way to find when when it's safe to load it?
It should be noted that I am also calling System.loadLibrary("main") from my NativeActivity's onCreate()
Not sure, but it is recommended to call loadLibrary() from a static initializer:
public class MainActivity extends Activity {
static {
System.loadLibrary("main")
}
...
}
Does it help?
I am trying to use a Singleton to share a large data object between Activities. But when I open the new Activity, the singleton comes up as empty. It seems to me that the Singleton should be the same no matter where in the Application I call if from.
It seems like the Scope of the Singleton is being limited to the individual Activity. Working around this is making my App very complicated. I must be doing something wrong. I even tried instantiating them in an extended Application class... Google says I should not have to use that though...
Can someone please point out where I am going wrong? i.e. Why does this singletom not contain the same data in each Activity?
I call it from an Activity with...
DataLog dataLog = DataLog.getInstance(this);
I have...
public class DataLog extends ArrayList<String> implements Serializable {
private static final long serialVersionUID = 0L;
private static DataLog sInstance;
private static Context mContext;
public static DataLog getInstance(Context context) {
mContext = context.getApplicationContext();
prefs = PreferenceManager.getDefaultSharedPreferences(mContext);
if (sInstance == null) {
sInstance = new DataLog();
}
return sInstance;
}
private DataLog() {
}
public boolean add(String entry) {
super.add(entry);
return true;
}
public void add(int index, String entry) {
if (index > 0)
super.add(index, entry);
else
super.add(entry);
}
public void clear() {
super.clear();
}
...
}
Its highly advisable to avoid singleton for sharing large data sets in android.
Singletons are used for short life-cycle objects.
Switch to SharedPrefferences, SQLite DB's or file storing. You are not the only to have experienced this behavior, and the reason lies in the nature of android Activities and the system itself(managing activities and its data).
Here is an example why singleton is bad for your case:
You stored important data in it. The user knows that he can close the app on home button to call someone or whatever)maybe someone called him when he was in your app), and that when he opens your app he will come back at the same place with everything in order. (this is expected behavior from users and android apps). The system can easily kill your process and all static variables in it for memory maintenance, app inactivity etc...result=data lost. Thus its not safe to use it.
With the release of Gingerbread, I have been experimenting with some of the new API's, one of them being StrictMode.
I noticed that one of the warnings is for getSharedPreferences().
This is the warning:
StrictMode policy violation; ~duration=1949 ms: android.os.StrictMode$StrictModeDiskReadViolation: policy=23 violation=2
and it's being given for a getSharedPreferences() call being made on the UI thread.
Should SharedPreferences access and changes really be made off the UI thread?
I'm glad you're already playing with it!
Some things to note: (in lazy bullet form)
if this is the worst of your problems, your app's probably in a good spot. :) Writes are generally slower than reads, though, so be sure you're using SharedPreferenced$Editor.apply() instead of commit(). apply() is new in GB and async (but always safe, careful of lifecycle transitions). You can use reflection to conditionally call apply() on GB+ and commit() on Froyo or below. I'll be doing a blogpost with sample code of how to do this.
Regarding loading, though...
once loaded, SharedPreferences are singletons and cached process-wide. so you want to get it loaded as early as possible so you have it in memory before you need it. (assuming it's small, as it should be if you're using SharedPreferences, a simple XML file...) You don't want to fault it in the future time some user clicks a button.
but whenever you call context.getSharedPreferences(...), the backing XML file is stat'd to see if it's changed, so you'll want to avoid those stats during UI events anyway. A stat should normally be fast (and often cached), but yaffs doesn't have much in the way of concurrency (and a lot of Android devices run on yaffs... Droid, Nexus One, etc.) so if you avoid disk, you avoid getting stuck behind other in-flight or pending disk operations.
so you'll probably want to load the SharedPreferences during your onCreate() and re-use the same instance, avoiding the stat.
but if you don't need your preferences anyway during onCreate(), that loading time is stalling your app's start-up unnecessarily, so it's generally better to have something like a FutureTask<SharedPreferences> subclass that kicks off a new thread to .set() the FutureTask subclasses's value. Then just lookup your FutureTask<SharedPreferences>'s member whenever you need it and .get() it. I plan to make this free behind the scenes in Honeycomb, transparently. I'll try to release some sample code which
shows best practices in this area.
Check the Android Developers blog for upcoming posts on StrictMode-related subjects in the coming week(s).
Accessing the shared preferences can take quite some time because they are read from flash storage. Do you read a lot? Maybe you could use a different format then, e.g. a SQLite database.
But don't fix everything you find using StrictMode. Or to quote the documentation:
But don't feel compelled to fix everything that StrictMode finds. In particular, many cases of disk access are often necessary during the normal activity lifecycle. Use StrictMode to find things you did by accident. Network requests on the UI thread are almost always a problem, though.
One subtlety about Brad's answer: even if you load the SharedPreferences in onCreate(), you should probably still read values on the background thread because getString() etc. block until reading the shared file preference in finishes (on a background thread):
public String getString(String key, String defValue) {
synchronized (this) {
awaitLoadedLocked();
String v = (String)mMap.get(key);
return v != null ? v : defValue;
}
}
edit() also blocks in the same way, although apply() appears to be safe on the foreground thread.
(BTW sorry to put this down here. I would have put this as a comment to Brad's answer, but I just joined and don't have enough reputation to do so.)
I know this is an old question but I want to share my approach. I had long reading times and used a combination of shared preferences and the global application class:
ApplicationClass:
public class ApplicationClass extends Application {
private LocalPreference.Filter filter;
public LocalPreference.Filter getFilter() {
return filter;
}
public void setFilter(LocalPreference.Filter filter) {
this.filter = filter;
}
}
LocalPreference:
public class LocalPreference {
public static void saveLocalPreferences(Activity activity, int maxDistance, int minAge,
int maxAge, boolean showMale, boolean showFemale) {
Filter filter = new Filter();
filter.setMaxDistance(maxDistance);
filter.setMinAge(minAge);
filter.setMaxAge(maxAge);
filter.setShowMale(showMale);
filter.setShowFemale(showFemale);
BabysitApplication babysitApplication = (BabysitApplication) activity.getApplication();
babysitApplication.setFilter(filter);
SecurePreferences securePreferences = new SecurePreferences(activity.getApplicationContext());
securePreferences.edit().putInt(Preference.FILER_MAX_DISTANCE.toString(), maxDistance).apply();
securePreferences.edit().putInt(Preference.FILER_MIN_AGE.toString(), minAge).apply();
securePreferences.edit().putInt(Preference.FILER_MAX_AGE.toString(), maxAge).apply();
securePreferences.edit().putBoolean(Preference.FILER_SHOW_MALE.toString(), showMale).apply();
securePreferences.edit().putBoolean(Preference.FILER_SHOW_FEMALE.toString(), showFemale).apply();
}
public static Filter getLocalPreferences(Activity activity) {
BabysitApplication babysitApplication = (BabysitApplication) activity.getApplication();
Filter applicationFilter = babysitApplication.getFilter();
if (applicationFilter != null) {
return applicationFilter;
} else {
Filter filter = new Filter();
SecurePreferences securePreferences = new SecurePreferences(activity.getApplicationContext());
filter.setMaxDistance(securePreferences.getInt(Preference.FILER_MAX_DISTANCE.toString(), 20));
filter.setMinAge(securePreferences.getInt(Preference.FILER_MIN_AGE.toString(), 15));
filter.setMaxAge(securePreferences.getInt(Preference.FILER_MAX_AGE.toString(), 50));
filter.setShowMale(securePreferences.getBoolean(Preference.FILER_SHOW_MALE.toString(), true));
filter.setShowFemale(securePreferences.getBoolean(Preference.FILER_SHOW_FEMALE.toString(), true));
babysitApplication.setFilter(filter);
return filter;
}
}
public static class Filter {
private int maxDistance;
private int minAge;
private int maxAge;
private boolean showMale;
private boolean showFemale;
public int getMaxDistance() {
return maxDistance;
}
public void setMaxDistance(int maxDistance) {
this.maxDistance = maxDistance;
}
public int getMinAge() {
return minAge;
}
public void setMinAge(int minAge) {
this.minAge = minAge;
}
public int getMaxAge() {
return maxAge;
}
public void setMaxAge(int maxAge) {
this.maxAge = maxAge;
}
public boolean isShowMale() {
return showMale;
}
public void setShowMale(boolean showMale) {
this.showMale = showMale;
}
public boolean isShowFemale() {
return showFemale;
}
public void setShowFemale(boolean showFemale) {
this.showFemale = showFemale;
}
}
}
MainActivity (activity that get called first in your application):
LocalPreference.getLocalPreferences(this);
Steps explained:
The main activity calls getLocalPreferences(this) -> this will read your preferences, set the filter object in your application class and returns it.
When you call the getLocalPreferences() function again somewhere else in the application it first checks if it's not available in the application class which is a lot faster.
NOTE: ALWAYS check if an application wide variable is different from NULL, reason -> http://www.developerphil.com/dont-store-data-in-the-application-object/
The application object will not stay in memory forever, it will get killed. Contrary to popular belief, the app won’t be restarted from scratch. Android will create a new Application object and start the activity where the user was before to give the illusion that the application was never killed in the first place.
If I didn't check on null I would allow a nullpointer to be thrown when calling for example getMaxDistance() on the filter object (if the application object was swiped from the memory by Android)
SharedPreferences class does some reads & writes within XML files on disk, so just like any other IO operation it could be blocking. The amount of data currently stored in SharedPreferences affects the time and resource consumed by the API calls. For minimal amounts of data it's a matter of a few milliseconds (sometimes even less than a millisecond) to get/put data. But from the point of view of an expert it could be important to improve the performance by doing the API calls in background. For an asynchronous SharedPreferences I suggest checking out the Datum library.
i do not see any reason to read them from a background thread. but to write it i would. at startup time the shared preference file is loaded into memory so its fast to access, but to write things can take a bit of time so we can use apply the write async. that should be the difference between commit and apply methods of shared prefs.