Eclipse Community Forums: NatTable » ConcurrentModification-Exception in ViewportLayer(?)

Home » Eclipse Projects » NatTable » ConcurrentModification-Exception in ViewportLayer(?)(A possible race-condition occurs when handling many events in a short time. )

Show: Today's Messages :: Show Polls :: Message Navigator

ConcurrentModification-Exception in ViewportLayer(?) [message #1719472]

Fri, 08 January 2016 09:02

Dennis Bo

Messages: 3
Registered: January 2016

Junior Member

Hi everyone,
I'm using NatTable to display a rather large and dynamic set of entries.
The update ratio is very variable, varying from 0 entries per minute to several hundreds per second.

My underlying data structure is a combination of GlazedLists (EventList -> ObservableList -> FilterList -> SortedList).

Now sometimes I get this exception:

index.php/fa/24539/0/

Most of the time it happens right at the beginning, but it also happens after serveral minutes.

So far as I can see it happens either if too many events travel up the layer-stack in a rather short amount of time or if 2 or more different events reach the ViewportLayer at the same time (like a PropertyChangeEvent from the ObservableList and a StructuralChangEvent from the EventList). But I'm only guessing at this point.

I build the natTable like this:

index.php/fa/24541/0/

I hope you can help me with this,
regards Dennis

Attachment: stack-trace natTable-exception.JPG
(Size: 199.82KB, Downloaded 406 times)
Attachment: NatTableExamplePart1.JPG
(Size: 149.14KB, Downloaded 557 times)
Attachment: NatTableExamplePart3.JPG
(Size: 105.46KB, Downloaded 465 times)
Attachment: NatTableExamplePart2.JPG
(Size: 123.26KB, Downloaded 586 times)

Report message to a moderator

Re: ConcurrentModification-Exception in ViewportLayer(?) [message #1719544 is a reply to message #1719472]

Fri, 08 January 2016 17:47

Dirk Fauth

Messages: 2902
Registered: July 2012

Senior Member

Looking at the code at that position it is iterated over the list of EventHandlers. Are you dynamically adding and removing EventHandlers while processing? Or what version of NatTable are you using?

Since you are using the GlazedLIstsEventLayer, there should be no problem with the event processing, because the GlazedLIstsEventLayer collects the events and only processes one event per 100ms to avoid such heavy overprocessing.

Report message to a moderator

Re: ConcurrentModification-Exception in ViewportLayer(?) [message #1719547 is a reply to message #1719544]

Fri, 08 January 2016 18:30

Dennis Bo

Messages: 3
Registered: January 2016

Junior Member

Hi Dirk,
I'm using the current version of the NatTable (or so I believe, Build #465). I think I determind the source of the problem. My AutoScrollHandler is listening to RowStructuralChangeEvents. Each time it receives one it determines the last row and triggers moveRowPositionIntoView() of the ViewportLayer. If I understand the source code there correctly this method is registering an EventHandler and 100ms later it unregisters the same one.

When at the same time any other event is handled by the ViewportLayer the ConcurrentModification-Exception occurs.

I'm fresh out of ideas I could try in my own code to prevent this, but i believe if the ViewportLayer would override handleLayerEvent(), registerEventHandler() and the unregisterEventHandler() this could work.
By using a lock object the ViewportLayer could acquire a read-lock before iterating over the eventHandler map and a write-lock before adding or removing a handler. This souldn't decrease the performance by much and it should increase thread-safety for the eventHandlers.

Regards Dennis

[Updated on: Fri, 08 January 2016 18:34]

Report message to a moderator

Re: ConcurrentModification-Exception in ViewportLayer(?) [message #1719548 is a reply to message #1719547]

Fri, 08 January 2016 18:50

Dirk Fauth

Messages: 2902
Registered: July 2012

Senior Member

Open a ticket and provide a patch if you have a solution for this.

Report message to a moderator

Re: ConcurrentModification-Exception in ViewportLayer(?) [message #1719550 is a reply to message #1719548]

Fri, 08 January 2016 19:02

Dirk Fauth

Messages: 2902
Registered: July 2012

Senior Member

BTW, I think it should be fine to to synchronize in the AbstractLayer. It should be quite uncommon to register and unregister event handler while processing events, therefore a read lock should not have too much impact.

Report message to a moderator