I found strange hibernate behavior and I cannot explain it.
If I create an object in default thread inside transaction and make manual flush
then I cannot find it in other thread.
If I create an object in one special thread with the same conditions then everything is all right.
Here is the code that I described above:
// transaction template with propagation required
ttNew.execute(new TransactionCallbackWithoutResult() {
#Override
protected void doInTransactionWithoutResult(TransactionStatus status) {
Assert.assertEquals(envStDao.getAll().size(), 0);
g = new Group();
g.setDescription("trial");
// in debugger I get id = 1
groupDao.save(g);
groupDao.flush();
accDao.flush();
}
});
// second stage right after the first - searching the group
Thread t2 = new Thread(new Runnable() {
#Override
public void run() {
ttNew.execute(new TransactionCallbackWithoutResult() {
#Override
protected void doInTransactionWithoutResult(TransactionStatus status) {
// here I get NULL!
Group gg = groupDao.get(1);
}
});
}
});
t2.start();
t2.join();
If I wrap first block of the code into thread just as former I get the group.
Are any ideas?
I run above code in junit test. Dao objects use HibernateTemplate.
Due to transaction isolation you cannot see uncommitted data in another transaction. you have two different transaction here , so one cannot see uncommitted data of another.
The default isolationist is read committed. flush dosnt mean commit. commit will be done only at the end of the transaction. so when you flush the data in first transaction the data is written to the db , but doesn’t commit, so transaction 2 cannot see it.
Related
I'm trying to understand how threads works in java. This is a simple database request that returns a ResultSet. I'm using JavaFx.
package application;
import java.sql.ResultSet;
import java.sql.SQLException;
import javafx.fxml.FXML;
import javafx.scene.control.Button;
import javafx.scene.control.Label;
import javafx.scene.control.TextField;
public class Controller{
#FXML
private Button getCourseBtn;
#FXML
private TextField courseId;
#FXML
private Label courseCodeLbl;
private ModelController mController;
private void requestCourseName(){
String courseName = "";
Course c = new Course();
c.setCCode(Integer.valueOf(courseId.getText()));
mController = new ModelController(c);
try {
ResultSet rs = mController.<Course>get();
if(rs.next()){
courseCodeLbl.setText(rs.getString(1));
}
} catch (SQLException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
// return courseName;
}
public void getCourseNameOnClick(){
try {
// courseCodeLbl.setText(requestCourseName());
Thread t = new Thread(new Runnable(){
public void run(){
requestCourseName();
}
}, "Thread A");
t.start();
} catch (NumberFormatException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}
This returns an exception:
Exception in thread "Thread A" java.lang.IllegalStateException: Not on FX application thread; currentThread = Thread A
How do I correctly implement threading so that every database request is executed in a second thread instead of the main thread?
I've heard of implementing Runnable but then how do I invoke different methods in run method?
Never worked with threading before but I thought it's time for it.
Threading Rules for JavaFX
There are two basic rules for threads and JavaFX:
Any code that modifies or accesses the state of a node that is part of a scene graph must be executed on the JavaFX application thread. Certain other operations (e.g. creating new Stages) are also bound by this rule.
Any code that may take a long time to run should be executed on a background thread (i.e. not on the FX Application Thread).
The reason for the first rule is that, like most UI toolkits, the framework is written without any synchronization on the state of elements of the scene graph. Adding synchronization incurs a performance cost, and this turns out to be a prohibitive cost for UI toolkits. Thus only one thread can safely access this state. Since the UI thread (FX Application Thread for JavaFX) needs to access this state to render the scene, the FX Application Thread is the only thread on which you can access "live" scene graph state. In JavaFX 8 and later, most methods subject to this rule perform checks and throw runtime exceptions if the rule is violated. (This is in contrast to Swing, where you can write "illegal" code and it may appear to run fine, but is in fact prone to random and unpredictable failure at arbitrary time.) This is the cause of the IllegalStateException you are seeing: you are calling courseCodeLbl.setText(...) from a thread other than the FX Application Thread.
The reason for the second rule is that the FX Application Thread, as well as being responsible for processing user events, is also responsible for rendering the scene. Thus if you perform a long-running operation on that thread, the UI will not be rendered until that operation is complete, and will become unresponsive to user events. While this won't generate exceptions or cause corrupt object state (as violating rule 1 will), it (at best) creates a poor user experience.
Thus if you have a long-running operation (such as accessing a database) that needs to update the UI on completion, the basic plan is to perform the long-running operation in a background thread, returning the results of the operation when it is complete, and then schedule an update to the UI on the UI (FX Application) thread. All single-threaded UI toolkits have a mechanism to do this: in JavaFX you can do so by calling Platform.runLater(Runnable r) to execute r.run() on the FX Application Thread. (In Swing, you can call SwingUtilities.invokeLater(Runnable r) to execute r.run() on the AWT event dispatch thread.) JavaFX (see later in this answer) also provides some higher-level API for managing the communication back to the FX Application Thread.
General Good Practices for Multithreading
The best practice for working with multiple threads is to structure code that is to be executed on a "user-defined" thread as an object that is initialized with some fixed state, has a method to perform the operation, and on completion returns an object representing the result. Using immutable objects, in particular, a record, for the initialized state and computation result is highly desirable. The idea here is to eliminate the possibility of any mutable state being visible from multiple threads as far as possible. Accessing data from a database fits this idiom nicely: you can initialize your "worker" object with the parameters for the database access (search terms, etc). Perform the database query and get a result set, use the result set to populate a collection of domain objects, and return the collection at the end.
In some cases it will be necessary to share mutable state between multiple threads. When this absolutely has to be done, you need to carefully synchronize access to that state to avoid observing the state in an inconsistent state (there are other more subtle issues that need to be addressed, such as liveness of the state, etc). The strong recommendation when this is needed is to use a high-level library to manage these complexities for you.
Using the javafx.concurrent API
JavaFX provides a concurrency API that is designed for executing code in a background thread, with API specifically designed for updating the JavaFX UI on completion of (or during) the execution of that code. This API is designed to interact with the java.util.concurrent API, which provides general facilities for writing multithreaded code (but with no UI hooks). The key class in javafx.concurrent is Task, which represents a single, one-off, unit of work intended to be performed on a background thread. This class defines a single abstract method, call(), which takes no parameters, returns a result, and may throw checked exceptions. Task implements Runnable with its run() method simply invoking call(). Task also has a collection of methods which are guaranteed to update state on the FX Application Thread, such as updateProgress(...), updateMessage(...), etc. It defines some observable properties (e.g. state and value): listeners to these properties will be notified of changes on the FX Application Thread. Finally, there are some convenience methods to register handlers (setOnSucceeded(...), setOnFailed(...), etc); any handlers registered via these methods will also be invoked on the FX Application Thread.
So the general formula for retrieving data from a database is:
Create a Task to handle the call to the database.
Initialize the Task with any state that is needed to perform the database call.
Implement the task's call() method to perform the database call, returning the results of the call.
Register a handler with the task to send the results to the UI when it is complete.
Invoke the task on a background thread.
For database access, I strongly recommend encapsulating the actual database code in a separate class that knows nothing about the UI (Data Access Object design pattern). Then just have the task invoke the methods on the data access object.
So you might have a DAO class like this (note there is no UI code here):
public class WidgetDAO {
// In real life, you might want a connection pool here, though for
// desktop applications a single connection often suffices:
private Connection conn ;
public WidgetDAO() throws Exception {
conn = ... ; // initialize connection (or connection pool...)
}
public List<Widget> getWidgetsByType(String type) throws SQLException {
try (PreparedStatement pstmt = conn.prepareStatement("select * from widget where type = ?")) {
pstmt.setString(1, type);
ResultSet rs = pstmt.executeQuery();
List<Widget> widgets = new ArrayList<>();
while (rs.next()) {
Widget widget = new Widget();
widget.setName(rs.getString("name"));
widget.setNumberOfBigRedButtons(rs.getString("btnCount"));
// ...
widgets.add(widget);
}
return widgets ;
}
}
// ...
public void shutdown() throws Exception {
conn.close();
}
}
Retrieving a bunch of widgets might take a long time, so any calls from a UI class (e.g a controller class) should schedule this on a background thread. A controller class might look like this:
public class MyController {
private WidgetDAO widgetAccessor ;
// java.util.concurrent.Executor typically provides a pool of threads...
private Executor exec ;
#FXML
private TextField widgetTypeSearchField ;
#FXML
private TableView<Widget> widgetTable ;
public void initialize() throws Exception {
widgetAccessor = new WidgetDAO();
// create executor that uses daemon threads:
exec = Executors.newCachedThreadPool(runnable -> {
Thread t = new Thread(runnable);
t.setDaemon(true);
return t ;
});
}
// handle search button:
#FXML
public void searchWidgets() {
final String searchString = widgetTypeSearchField.getText();
Task<List<Widget>> widgetSearchTask = new Task<List<Widget>>() {
#Override
public List<Widget> call() throws Exception {
return widgetAccessor.getWidgetsByType(searchString);
}
};
widgetSearchTask.setOnFailed(e -> {
widgetSearchTask.getException().printStackTrace();
// inform user of error...
});
widgetSearchTask.setOnSucceeded(e ->
// Task.getValue() gives the value returned from call()...
widgetTable.getItems().setAll(widgetSearchTask.getValue()));
// run the task using a thread from the thread pool:
exec.execute(widgetSearchTask);
}
// ...
}
Notice how the call to the (potentially) long-running DAO method is wrapped in a Task which is run on a background thread (via the accessor) to prevent blocking the UI (rule 2 above). The update to the UI (widgetTable.setItems(...)) is actually executed back on the FX Application Thread, using the Task's convenience callback method setOnSucceeded(...) (satisfying rule 1).
In your case, the database access you are performing returns a single result, so you might have a method like
public class MyDAO {
private Connection conn ;
// constructor etc...
public Course getCourseByCode(int code) throws SQLException {
try (PreparedStatement pstmt = conn.prepareStatement("select * from course where c_code = ?")) {
pstmt.setInt(1, code);
ResultSet results = pstmt.executeQuery();
if (results.next()) {
Course course = new Course();
course.setName(results.getString("c_name"));
// etc...
return course ;
} else {
// maybe throw an exception if you want to insist course with given code exists
// or consider using Optional<Course>...
return null ;
}
}
}
// ...
}
And then your controller code would look like
final int courseCode = Integer.valueOf(courseId.getText());
Task<Course> courseTask = new Task<Course>() {
#Override
public Course call() throws Exception {
return myDAO.getCourseByCode(courseCode);
}
};
courseTask.setOnSucceeded(e -> {
Course course = courseTask.getCourse();
if (course != null) {
courseCodeLbl.setText(course.getName());
}
});
exec.execute(courseTask);
The API docs for Task have many more examples, including updating the progress property of the task (useful for progress bars..., etc.
Related
JavaFX - Background Thread for SQL Query
Sample for accessing a local database from JavaFX using concurrent tasks for database operations so that the UI remains responsive.
Exception in thread "Thread A" java.lang.IllegalStateException: Not on FX application thread; currentThread = Thread A
The exception is trying to tell you that you are trying to access JavaFX scene graph outside the JavaFX application thread. But where ??
courseCodeLbl.setText(rs.getString(1)); // <--- The culprit
If I can't do this how do I use a background thread?
The are different approaches which leads to similar solutions.
Wrap you Scene graph element with Platform.runLater
There easier and most simple way is to wrap the above line in Plaform.runLater, such that it gets executed on JavaFX Application thread.
Platform.runLater(() -> courseCodeLbl.setText(rs.getString(1)));
Use Task
The better approach to go with these scenarios is to use Task, which has specialized methods to send back updates. In the following example, I am using updateMessage to update the message. This property is bind to courseCodeLbl textProperty.
Task<Void> task = new Task<Void>() {
#Override
public Void call() {
String courseName = "";
Course c = new Course();
c.setCCode(Integer.valueOf(courseId.getText()));
mController = new ModelController(c);
try {
ResultSet rs = mController.<Course>get();
if(rs.next()) {
// update message property
updateMessage(rs.getString(1));
}
} catch (SQLException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
return null;
}
}
public void getCourseNameOnClick(){
try {
Thread t = new Thread(task);
// To update the label
courseCodeLbl.textProperty.bind(task.messageProperty());
t.setDaemon(true); // Imp! missing in your code
t.start();
} catch (NumberFormatException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
This has nothing to do with database. JavaFx, like pretty much all GUI libraries, requires that you only use the main UI thread to modify the GUI.
You need to pass the data from the database back to the main UI thread. Use Platform.runLater() to schedule a Runnable to be run in the main UI thread.
public void getCourseNameOnClick(){
new Thread(new Runnable(){
public void run(){
String courseName = requestCourseName();
Platform.runLater(new Runnable(){
courseCodeLbl.setText(courseName)
});
}
}, "Thread A").start();
}
Alternatively, you can use Task.
A toggle button in a JavaFX operation will be accessed by 2 separate threads.
1.
One thread will be invoked as soon as user clicks (toggles button state) and will
a) do something in the OS
b) check if (a) succeeded
c) exit on success / exit and return toggle button to previous state on failure
2
The other thread will monitor events asynchronous to the previous operation(s) and in case of a particular event it will change the button state.
Do I need to provide synchronization between threads 1 and 2 in terms of locking the button state?
EDIT: The idea proposed by James_D seems reasonable, but I just wanted to propose an alternative (whose effectiveness remains to be proved however).
How about using synchronized code blocks, and using as lock the reference to the particular button, i.e. something like:
// getting the reference to the button
#FXML
private ToggleButton tButtonToBeSynchronized
// Thread1
synchronized(tButtonToBeSynchronized) {
// do stuff with button upon user click
}
// Thread2
synchronized(tButtonToBeSynchronized) {
// poll system every X seconds
// when asynchronous event occurs (not related to UI events)
// update tButtonToBeSynchronized state
}
Would that work in case these are called by different Controller classes? (assuming the reference to the tButtonToBeSynchronized is passed by reference - and not by value by the FXML framework?
Like most UI toolkits, JavaFX assumes a single threaded model. You should only ever access the state of nodes that are part of a scene graph from the FX Application Thread. So, toggling a button is not an atomic operation, and the code you describe is not guaranteed to work as you currently have it set up. In Java 8, it will likely throw a RuntimeException.
JavaFX provides functionality to enable interoperability with background threads. The lowest level of these is Platform.runLater(Runnable r), which executes r on the FX Application Thread. So, your monitor thread (item 2 in your question) should change the state of the toggle button with
Platform.runLater( () -> toggleButton.setSelected(...) );
There is also a javafx.concurrent API. This provides a Task class, among others, which acts as both a Runnable and a java.util.concurrent.FutureTask, and additionally has a collection of callback methods for submitting code to be executed on the FX Application Thread at various points in the Task's lifecycle.
So you should implement item 1 in your question as:
ExecutorService exec = ... ; // e.g. Executors.newCachedThreadPool();
toggleButton.selectedItemProperty().addListener((obs, wasSelected, isNowSelected) -> {
if (isNowSelected) {
Task<Void> task = new Task<Void>() {
#Override
public Void call() throws Exception {
// do something on OS
// throw exception if failed
return null ;
}
};
task.setOnFailed(event -> toggleButton.setSelected(wasSelected));
exec.submit(task);
}
});
If you prefer to return a value indicating success or failure, you can do
Task<Boolean> task = new Task<Boolean>() {
#Override
public Boolean call() {
// do work...
boolean successful = ... ;
return successful ;
}
};
task.setOnSucceeded( event -> {
boolean wasSuccessful = task.getValue();
// ...
});
I'm trying to do some optimization in my code and would like to spawn a thread where I do a time consuming operation. During the implementation of that optimization I was running into an issue which was driving me crazy. I simplified the issue and created a test case for that specific issue: (I'm using SpringJUnit4ClassRunner so the transaction is properly started at the beginning of the testCRUD method)
Could someone help me understand why the foundParent is null in the thread ?
private Semaphore sema = new Semaphore(0, false);
private long parentId;
#Test
public void testCRUD() {
//create
DBParent parent = null;
{
parent = new DBParent();
parentDao.persist(parent);
parentId = parent.getId();
assertTrue(parentId > 0);
parentDao.flush();
}
(new Thread(
new Runnable() {
public void run()
{
System.out.println("Start adding childs !");
DBParent foundParent = parentDao.findById(parentId);
assertTrue(foundParent != null); //ASSERTION FAILS HERE !!!!
System.out.println("Releasing semaphore !");
sema.release();
System.out.println("End adding childs !");
}
})).start();
try {
System.out.println("Acquiring semaphore !");
sema.acquire();
}
catch (InterruptedException e) {
e.printStackTrace();
}
}
=============================EDITED===================================
As per one comment suggestion, I created a threadManager bean which spawn the thread. Here is the code of the threadManager:
public class ThreadManager {
#Transactional(propagation=Propagation.REQUIRES_NEW)
public void executeTask(String Name, Runnable task) {
(new Thread(task, Name)).start();
}
}
Then in the previous test, instead of staring the thread manually, I just post it in the thread manager like this:
#Autowired private ParentDao parentDao;
#Autowired private ThreadManager threadManager;
private Semaphore sema = new Semaphore(0, false);
private long parentId;
#Test
public void testCRUD() {
//create
DBParent parent = null;
{
parent = new DBParent();
parentDao.persist(parent);
parentId = parent.getId();
assertTrue(parentId > 0);
parentDao.flush();
}
threadManager.executeTask("BG processing...",
new Runnable() {
public void run()
{
System.out.println("Start adding childs !");
DBParent foundParent = parentDao.findById(parentId);
assertTrue(foundParent != null); //ASSERTION FAILS HERE !!!!
System.out.println("Releasing semaphore !");
sema.release();
System.out.println("End adding childs !");
}
});
try {
System.out.println("Acquiring semaphore !");
sema.acquire();
}
catch (InterruptedException e) {
e.printStackTrace();
}
}
Unfortunately this doesn't work either !!! :-(
The transaction context is bound to the thread. So the code in the spawned thread doesn't run in the same transaction context as the code in the initial thread. So, due to transaction isolation (the I in ACID), the spawned thread doesn't see what the initial thread's transaction is inserting in the database.
You can bind Spring transaction to a new thread, to run transactions & Hibernate/JPA access in it. But this has to be a different TX and JPA/HB session from other threads.
Spring code for OpenSessionInViewFilter, is a reasonable an example of how to bind Hibernate session to Spring's TX management. You can strip this down to fairly minimal code.
See:
org.springframework.orm.hibernate3.support.OpenSessionInViewFilter
OpenSessionInViewFilter.doFilterInternal() -- this is where it actually binds it
TransactionSynchronizationManager.bindResource()
TransactionSynchronizationManager.unbindResource()
TransactionSynchronizationManager.getResource()
In one project (IIRC) I wrapped this functionality into a 'ServerThreadHb' class, to setup & save previous thread-bindings on construction -- with a restore() method to be called in a finally block, to restore previous bindings.
For your posted code sample, there isn't much point in running work on a separate thread -- since you synchronously wait for the work to be done. However I assume you were planning to remove that constraint & extend that functionality.
I have a requirement in JavaFX 2 to display the notifications on the header. These notifications are populated in the database through an external system.
When a new notification is inserted in the database, my JavaFX application should listen to the same and refresh the header.
One solution I can think about is to implement a timeline that triggers the code to retrieve the latest notification status once every fixed time period, may be once every 2 mins.
Other than this, is there any other way to achieve this? Any hints on this would be highly appreciated.
You can create a new Task that would be listening (or checking) for changes in your database. This Task would be running in a different Thread as not to block your UI. Once a change occurs, it can then update your UI.
Your code could look like this :
//... initialize your variables.
task = new DatabaseWatcher();
executor = Executors.newSingleThreadExecutor();
executor.execute(task);
And your DatabaseWatcher :
public class DatabaseWatcher extends Task<Object> {
public DatabaseWatcher() {
//initialize here
}
#Override
protected Object call() throws Exception {
//Get data from database
//if new update :
updateUI();
return null;
}
private void updateUI() {
Platform.runLater(new Runnable() {
#Override
public void run() {
//Set your new values in your UI
//Call the method in your UI to update values.
}
});
}
}
This should get you started on the right path.
See Also
Multiple FXML with Controllers, share object, more specifically this answer
javafx, update ui from another thread
Concurrency in JavaFX
Have a table 'temp' ..
Code:
CREATE TABLE `temp` (
`id` bigint(20) unsigned NOT NULL AUTO_INCREMENT,
`student_id` bigint(20) unsigned NOT NULL,
`current` tinyint(1) NOT NULL DEFAULT '1',
`closed_at` datetime NOT NULL,
PRIMARY KEY (`id`),
UNIQUE KEY `unique_index` (`student_id`,`current`,`closed_at`),
KEY `studentIndex` (`student_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8
The corresponding Java pojo is http://pastebin.com/JHZwubWd . This table has a unique constraint such that only one record for each student can be active.
2) I have a test code which does try to continually add records for a student ( each time making the older active one as inactive and adding a new active record) and also in a different thread accessing some random ( non-related ) table.
Code:
public static void main(String[] args) throws Exception {
final SessionFactory sessionFactory = new AnnotationConfiguration().configure().buildSessionFactory();
ExecutorService executorService = Executors.newFixedThreadPool(1);
int runs = 0;
while(true) {
Temp testPojo = new Temp();
testPojo.setStudentId(1L);
testPojo.setCurrent(true);
testPojo.setClosedAt(new Date(0));
add(testPojo, sessionFactory);
Thread.sleep(1500);
executorService.submit(new Callable<Object>() {
#Override
public Object call() throws Exception {
Session session = sessionFactory.openSession();
// Some dummy code to print number of users in the system.
// Idea is to "touch" the DB/session in this background
// thread.
System.out.println("No of users: " + session.createCriteria(User.class).list().size());
session.close();
return null;
}
});
if(runs++ > 100) {
break;
}
}
executorService.shutdown();
executorService.awaitTermination(1, TimeUnit.MINUTES);
}
private static void add(final Temp testPojo, final SessionFactory sessionFactory) throws Exception {
Session dbSession = null;
Transaction transaction = null;
try {
dbSession = sessionFactory.openSession();
transaction = dbSession.beginTransaction();
// Set all previous state of the student as not current.
List<Temp> oldActivePojos = (List<Temp>) dbSession.createCriteria(Temp.class)
.add(Restrictions.eq("studentId", testPojo.getStudentId())).add(Restrictions.eq("current", true))
.list();
for(final Temp oldActivePojo : oldActivePojos) {
oldActivePojo.setCurrent(false);
oldActivePojo.setClosedAt(new Date());
dbSession.update(oldActivePojo);
LOG.debug(String.format(" Updated old state as inactive:%s", oldActivePojo));
}
if(!oldActivePojos.isEmpty()) {
dbSession.flush();
}
LOG.debug(String.format(" saving state:%s", testPojo));
dbSession.save(testPojo);
LOG.debug(String.format(" new state saved:%s", testPojo));
transaction.commit();
}catch(Exception exception) {
LOG.fatal(String.format("Exception in adding state: %s", testPojo), exception);
transaction.rollback();
}finally {
dbSession.close();
}
}
Upon running the code, after a few runs, I am getting an index constraint exception. It happens because for some strange reason, it does not find the latest active record but instead some older stale active record and tries marking it as inactive before saving ( though the DB actually has a new active record already present).
Notice that both the code share the same sessionfactory and the both code works on a totally different tables. My guess is that some internal cache state gets dirty. If I use 2 different sessionfactory for the foreground and background thread, it works fine.
Another weird thing is that in the background thread ( where I print the no of users), if I wrap it in a transaction ( even though it is only a read operation), the code works fine! Sp looks like I need to wrap all DB operations ( irrespective of read / write ) in a transaction for it to work in a multithreaded environment.
Can someone point out the issue?
Yes, basically, transaction demarcation is always needed:
Hibernate documentation says:
Database, or system, transaction boundaries are always necessary. No communication with the database can occur outside of a database transaction (this seems to confuse many developers who are used to the auto-commit mode). Always use clear transaction boundaries, even for read-only operations. Depending on your isolation level and database capabilities this might not be required, but there is no downside if you always demarcate transactions explicitly.
When trying to reproduce your setup I experienced some problems caused by the lack of transaction demarcation (though not the same as yours). Further investigation showed that sometimes, depending on connection pool configuration, add() is executed in the same database transaction as the previous call(). Adding beginTransaction()/commit() to call() fixed that problem. This behaviour can be responsible for your problem, since, depending on transaction isolation level, add() can work with the stale snapshot of the database taken at the begining of transaction, i.e. during the previous call().