Invoking a JavaScript function from oncomplete handler of p:remoteCommand - simulating the same using some JavaScript code

Question

Caution : Although this question covers long textual information with a mess of Java code snippets, it is merely targeted to JavaScript/jQuery and a bit of PrimeFaces stuff (just <p:remoteCommand>) as mentioned in the introductory part in the beginning.

---

I am receiving a JSON message from WebSockets (Java EE 7 / JSR 356 WebSocket API) as follows.

if (window.WebSocket) {
    var ws = new WebSocket("wss://localhost:8181/ContextPath/AdminPush");

    ws.onmessage = function (event) {
        jsonMsg=event.data;
        var json = JSON.parse(jsonMsg);        
        var msg=json["jsonMessage"];

        if (window[msg]) {
            window[msg](); //It is literally interpreted as a function - updateModel();
        }
    };
}

In the above code, event.data contains a JSON string {"jsonMessage":"updateModel"}. Thus, msg will contain a string value which is updateModel.

In the following segment of code,

if (window[msg]) {
    window[msg](); //It is literally interpreted as a JavaScript function - updateModel();
}

window[msg](); causes a JavaScript function associated with a <p:remoteCommand> to be invoked (which in turn invokes an actionListener="#{bean.remoteAction}" associated with the <p:remoteCommand>).

<p:remoteCommand name="updateModel"
                 actionListener="#{bean.remoteAction}" 
                 oncomplete="notifyAll()"
                 process="@this"
                 update="@none"/>

_{update="@none" is not necessarily needed.}

---

After receiving this message, I need to notify all the associated clients about this update. I use the following JavaScript function to do so which is associated with the oncomplete handler of the above <p:remoteCommand>.

var jsonMsg;

function notifyAll() {
    if(jsonMsg) {
       sendMessage(jsonMsg);
    }
}

Notice that the variable jsonMsg is already assigned a value in the first snippet - it is a global variable. sendMessage() is another JavaScript function that actually sends a notification about this update to all the associated clients through WebSockets which is not needed in this question.

---

This works well but is there a way to do some magic in the following condition

if (window[msg]) {
    window[msg]();

    //Do something to call notifyAll() on oncomplete of remote command.
}

so that the notifyAll() function can be invoked through some JavaScript code directly (which is currently attached to oncomplete of <p:remoteCommand> and the expected JavaScript code (or even something else) should simulate this oncomplete) basically eliminating the need to depend upon a global JavaScript variable (jsonMSg)?

---

Edit : The problem I am trying to solve (it may be considered to be additional information).

When an admin for example, makes some changes (by means of DML operations) to a JPA entity named Category, entity listeners are triggered which in turn causes a CDI event to be raised as follows.

@ApplicationScoped
public class CategoryListener {

    @PostPersist
    @PostUpdate
    @PostRemove
    public void onChange(Category category) throws NamingException {
        BeanManager beanManager = (BeanManager) InitialContext.doLookup("java:comp/BeanManager");
        beanManager.fireEvent(new CategoryChangeEvent(category));
    }
}

Needless to say that the entity Category is designated with the annotation @EntityListeners(CategoryListener.class).

_{Just one side note (completely off topic) : Getting an instance of BeanManager through a JNDI look-up as done in the preceding code snippet is temporary. The GlassFish Server 4.1 having the Weld version 2.2.2 final fails to inject the CDI event javax.enterprise.event.Event<T> which is supposed to be injected as follows.}

@Inject
private Event<CategoryChangeEvent> event;

And then, the event can be fired as follows with reference to the relevant code snippet above.

event.fire(new CategoryChangeEvent(category));

---

This event is observed in the web project as follows.

@ApplicationScoped
public class RealTimeUpdate {

    public void onCategoryChange(@Observes CategoryChangeEvent event) {
        AdminPush.sendAll("updateModel");
    }
}

Where an admin uses his own end-point as follows (AdminPush.sendAll("updateModel"); is invoked manually therein).

@ServerEndpoint(value = "/AdminPush", configurator = ServletAwareConfig.class)
public final class AdminPush {

    private static final Set<Session> sessions = new LinkedHashSet<Session>();

    @OnOpen
    public void onOpen(Session session, EndpointConfig config) {
        if (Boolean.valueOf((String) config.getUserProperties().get("isAdmin"))) {
            sessions.add(session);
        }
    }

    @OnClose
    public void onClose(Session session) {
        sessions.remove(session);
    }

    private static JsonObject createJsonMessage(String message) {
        return JsonProvider.provider().createObjectBuilder().add("jsonMessage", message).build();
    }

    public static void sendAll(String text) {
        synchronized (sessions) {
            String message = createJsonMessage(text).toString();
            for (Session session : sessions) {
                if (session.isOpen()) {
                    session.getAsyncRemote().sendText(message);
                }
            }
        }
    }
}

Here only an admin is allowed to use this end-point. All other users are prevented from creating a WebSocket session using a conditional check in the onOpen() method.

session.getAsyncRemote().sendText(message); inside the foreach loop sends a notification (in the form of a JSON message) to the admin about these changes made in the entity Category.

As shown in the first code snippet, window[msg](); invokes an action method (through a <p:remoteCommand> as shown earlier) associated with an application scoped bean - actionListener="#{realTimeMenuManagedBean.remoteAction}".

@Named
@ApplicationScoped
public class RealTimeMenuManagedBean {

    @Inject
    private ParentMenuBeanLocal service;

    private List<Category> category;
    private final Map<Long, List<SubCategory>> categoryMap = new LinkedHashMap<Long, List<SubCategory>>();
    // Other lists and maps as and when required for a dynamic CSS menu.

    public RealTimeMenuManagedBean() {}

    @PostConstruct
    private void init() {
        populate();
    }

    private void populate() {
        categoryMap.clear();
        category = service.getCategoryList();

        for (Category c : category) {
            Long catId = c.getCatId();
            categoryMap.put(catId, service.getSubCategoryList(catId));
        }
    }

    // This method is invoked through the above-mentioned <p:remoteCommand>.
    public void remoteAction() {
        populate();
    }

    // Necessary accessor methods only.
}

All other users/clients (who are on a different panel - other than the admin panel) should only be notified when actionListener="#{realTimeMenuManagedBean.remoteAction}" finishes in its entirely - must not happen before the action method finishes - should be notified through the oncomplate event handler of <p:remoteCommand>. This is the reason why two different end-points have been taken.

---

Those other users use their own end-point:

@ServerEndpoint("/Push")
public final class Push {

    private static final Set<Session> sessions = new LinkedHashSet<Session>();

    @OnOpen
    public void onOpen(Session session) {
        sessions.add(session);
    }

    @OnClose
    public void onClose(Session session) {
        sessions.remove(session);
    }

    @OnMessage
    public void onMessage(String text) {
        synchronized (sessions) {
            for (Session session : sessions) {
                if (session.isOpen()) {
                    session.getAsyncRemote().sendText(text);
                }
            }
        }
    }
}

The method annotated with @OnMessage comes to play, when a message is sent through oncomplete of <p:remoteCommand> as shown above.

Those clients use the following JavaScript code to just fetch the new values from the above-mentioned application scoped bean (the bean was already queried adequately by the admin from the database. Thus, there is no need to ridiculously query it again by each and every individual client separately (other than the admin). Hence, it is an application scoped bean).

if (window.WebSocket) {
    var ws = new WebSocket("wss://localhost:8181/ContextPath/Push");
    ws.onmessage = function (event) {
        var json = JSON.parse(event.data);
        var msg = json["jsonMessage"];

        if (window[msg]) {
            window[msg]();
        }
    };

    $(window).on('beforeunload', function () {
        ws.close();
    });
}

In conjunction with the following <p:remoteCommand>.

<p:remoteCommand name="updateModel"
                 process="@this"
                 update="parentMenu"/>

Where parentMenu - the component to be updated by this <p:remoteCommand> is an id of a container JSF component <h:panelGroup> which contains a plain CSS menu with a bunch of <ui:repeat>s.

Hope this makes the scenario clearer.

---

Update :

This question has been answered precisely here based on <p:remoteCommand> (As to the concrete question, the sole question was to remove a dependency upon a global JavaScript variable as stated in the introductory part of this question).

Answer 1

I don't think I understood every aspect of your problem, but anyway I try to help a bit. Note that I do not know PrimeFaces, so all I did was reading the docs.

What I understand is, that you try to get rid of the global variable. But I am afraid, I do not think this is possible.

The problem here is, that PrimeFaces does not allow you to pass something transparently from your invocation of the remote call further to the oncomplete call (except you pass it to a Java code of the Bean and then back to the UI, and this usually is not what you want).

However, I hope, you can come very close to it.

Part 1, JS returns early

Please also note that there probably is some misconception about Java and JavaScript.

Java is multithreaded and runs several commands in parallel, while JavaScript is singlethreaded and usually never waits for something to complete. Doing things asychronously is mandatory to get a responsive Web-UI.

Hence your remoteCommand invocation (seen from the JS side) will (usually, async case) return long before the oncomplete handler will be invoked. That means, if window[msg]() returns, you are not finished with the remoteCommand yet.

So what you want to manage with following code

if (window[msg]) {
    window[msg]();

    //Do something to call notifyAll() on oncomplete of remote command.
    dosomethinghere();
}

will fail. dosomethinghere() will not be invoked when the remoteCommand returned (as JS does not want to wait for some event, which might never happen). This means, dosomethinghere() will be invoked when the Ajax-request was just opened to the remote (to the Java application).

To run something after the Ajax call finished, this must be done in the oncomplete routine (or onsuccess). This is why it's there.

Part 2, validate msg

Please note something different about window[msg](). This can be considered a bit dangerous if you cannot trust the pushed message completely. window[msg]() essentially runs any function named with the contents of the variable msg. For example if msg happen to be close then window.close() will be run, which probably is not what you want.

You should make sure, msg is one expected word, and decline all other words. Example code for this:

var validmsg = { updateModel:1, rc:1 }

[..]

if (validmsg[msg] && window[msg])
  window[msg]();

Part 3: How to handle multiple JSON messages in parallel

The global variable has some drawback. There is only one. If you happen to receive another JSON message on the WebSocket while the previous message still is processing in the remoteCommand, this will overwrite the previous message. So the notifyAll() will see the newer message twice, the old one is lost.

A classical race condition. What you must do is, to create something like a registry to register all the messages, and then pass some value to notifyAll() to tell, which of the registered messages shall be processed.

With only a little change, you can either parallely (here) or serially (Part 4) process the messages.

First, create a counter to be able to distinguish the messages. Also an object to store all the messages. And we declare all valid messages we expect (see Part 2):

var jsonMsgNr = 0;
var jsonMessages = {};
var validmsg = { updateModel:1 }

Now add a message each time we receive one:

if (window.WebSocket) {
    var ws = new WebSocket("wss://localhost:8181/ContextPath/AdminPush");

    ws.onmessage = function (event) {
        var jsonMsg = event.data;
        var json = JSON.parse(jsonMsg);        
        var msg=json["jsonMessage"];

        if (validmsg[msg] && window[msg]) {
            var nr = ++jsonMsgNr;
            jsonMessages[nr] = { jsonMsg:jsonMsg, json:json };

To be able to pass the nr to NotifyAll() an additional parameter needs to be passed to the Bean. Let's call it msgNr:

            // Following might look a bit different on older PrimeFaces
            window[msg]([{name:'msgNr', value:nr}]);
        }
    }
}

Perhaps have a look into https://stackoverflow.com/a/7221579/490291 for more on passing values this way.

The remoteAction bean now gets an additional parameter msgNr passed, which must be passed back via Ajax.

Unfortunately I have no idea (sorry) how this looks in Java. So make sure, your answer to the AjaxCall copies the msgNr out again.

Also, as the documentation is quiet about this subject, I am not sure how the parameters are passed back to the oncomplete handler. According to the JavaScript debugger, notifyAll() gets 3 parameters: xhdr, payload, and pfArgs. Unfortunately I was not able to setup a test case to find out how things look like.

Hence the function looks a bit like (bear with me, please):

function notifyAll(x, data, pfArgs) {
   var nr = ???; // find out how to extract msgNr from data

   var jsonMsg = jsonMessages[nr].jsonMsg;
   var json = jsonMessages[nr].json;
   jsonMessages[nr] = null;  // free memory

   sendMessage(jsonMsg);

   dosomething(json);
}

If you split this into two functions, then you can invoke the notifyAll() from other parts in your application:

function notifyAll(x, data, unk) {
   var nr = ???; // find out how to extract msgNr from data

   realNotifyAll(nr);
}

function realNotifyAll(nr) {
  if (!(nr in jsonMessages)) return;

  var jsonMsg = jsonMessages[nr].jsonMsg;
  var json = jsonMessages[nr].json;
  delete jsonMessages[nr];  // free memory

  sendMessage(jsonMsg);

  dosomething(json);
}

Some things here are a bit redundant. For example you perhaps do not need the json element in jsonMessages or want to parse the json again to spare some memory in case the json is very big. However the code is meant not to be optimal but to be easy to adjust to your needs.

Part 4: serialize requests

Now to the changes to serialize things. That's quite easy by adding some semaphore. Semaphores in JavaScript are just variables. This is because there is only one global thread.

var jsonMsgNr = 0;
var jsonMessages = {};
var validmsg = { updateModel:1 }
var jsonMsgNrLast = 0;           // ADDED

if (window.WebSocket) {
    var ws = new WebSocket("wss://localhost:8181/ContextPath/AdminPush");

    ws.onmessage = function (event) {
        var jsonMsg = event.data;
        var json = JSON.parse(jsonMsg);        
        var msg=json["jsonMessage"];

        if (validmsg[msg] && window[msg]) {
            var nr = ++jsonMsgNr;
            jsonMessages[nr] = { jsonMsg:jsonMsg, json:json };

            if (!jsonMsgNrLast) {    // ADDED
                jsonMsgNrLast = nr;  // ADDED
                window[msg]([{name:'msgNr', value:nr}]);
            }
        }
    }
}

function realNotifyAll(nr) {
  if (!(nr in jsonMessages)) return;

  var jsonMsg = jsonMessages[nr].jsonMsg;
  var json = jsonMessages[nr].json;
  delete jsonMessages[nr];  // free memory

  sendMessage(jsonMsg);

  dosomething(json);

  // Following ADDED
  nr++;
  jsonMsgNrLast = 0;
  if (nr in jsonMessages)
    {
      jsonMsgNrLast = nr;
      window[jsonMessages[nr].json.msg]([{name:'msgNr', value:nr}]);
    }
}

Note: jsonMsgNrLast could be just a flag (true/false). However having the current processed number in a variable perhaps can help somewhere else.

Having said that, there is a starvation problem in case something fails in sendMessage or dosomething. So perhaps you can interleave it a bit:

function realNotifyAll(nr) {
  if (!(nr in jsonMessages)) return;

  var jsonMsg = jsonMessages[nr].jsonMsg;
  var json = jsonMessages[nr].json;
  delete jsonMessages[nr];  // free memory

  nr++;
  jsonMsgNrLast = 0;
  if (nr in jsonMessages)
    {
      jsonMsgNrLast = nr;
      // Be sure you are async here!
      window[jsonMessages[nr].json.msg]([{name:'msgNr', value:nr}]);
    }

  // Moved, but now must not rely on jsonMsgNrLast:
  sendMessage(jsonMsg);
  dosomething(json);
}

This way the AJAX request is already send out while sendMessage is running. If now dosomething has a JavaScript error or similar, the messages are still processed correctly.

Please note: All this was typed in without any tests. There might be syntax errors or worse. Sorry, I tried my best. If you find a bug, edit is your friend.

Part 5: Direct Invocation from JS

Now, with all this in place and a serialized Run, you can always invoke the previous notifyAll() using realNotifyAll(jsonMsgNrLast). Or you can display the jsonMessages in a list and choose any arbitrary number.

By skipping the call to window[jsonMessages[nr].json.msg]([{name:'msgNr', value:nr}]); (and above window[msg]([{name:'msgNr', value:nr}]);) you also can halt the Bean processing and run it on-demand using the usual JQuery callbacks. For this create a function and change the code a bit again:

var jsonMsgNr = 0;
var jsonMessages = {};
var validmsg = { updateModel:1 }
var jsonMsgNrLast = 0;
var autoRun = true;        // ADDED, set false control through GUI

if (window.WebSocket) {
    var ws = new WebSocket("wss://localhost:8181/ContextPath/AdminPush");

    ws.onmessage = function (event) {
        var jsonMsg = event.data;
        var json = JSON.parse(jsonMsg);        

        if (validmsg[msg] && window[msg]) {
            var nr = ++jsonMsgNr;
            jsonMessages[nr] = { jsonMsg:jsonMsg, json:json };

            updateGuiPushList(nr, 1);

            if (autoRun && !jsonMsgNrLast) {
                runRemote(nr);
            }
        }
    }
}

function realNotifyAll(nr) {
  if (!(nr in jsonMessages)) return;

  var jsonMsg = jsonMessages[nr].jsonMsg;
  var json = jsonMessages[nr].json;
  delete jsonMessages[nr];  // free memory

  updateGuiPushList(nr, 0);

  jsonMsgNrLast = 0;
  if (autoRun)
    runRemote(nr+1);

  // Moved, but now must not rely on jsonMsgNrLast:
  sendMessage(jsonMsg);
  dosomething(json);
}

function runRemote(nr) {
  if (nr==jsonMsgNrLast) return;
  if (nr in jsonMessages)
    {
      if (jsonMsgNrLast) { alert("Whoopsie! Please wait until processing finished"); return; }
      jsonMsgNrLast = nr;

      updateGuiPushList(nr, 2);

      // Be sure you are async here!
      window[jsonMessages[nr].json.msg]([{name:'msgNr', value:nr}]);
    }
}

Now you can start the processing with runRemote(nr) and invoke the completion function with realNotifyAll(nr).

The function updateGuiPushList(nr, state) with state=0:finished 1=added 2=running is the callback to your GUI code which updates the on-screen list of waiting pushes to process. Set autoRun=false to stop automatic processing and autoRun=true for automatic processing.

Note: After setting autoRun from false to true you need to trigger runRemote once with the lowest nr, of course.