On Mon, Jan 10, 2011 at 4:57 PM, Alex Dean <a...@crackpot.org> wrote:
> > On Jan 10, 2011, at 5:02 AM, Jeff Rose wrote: > > > On Sat, Jan 8, 2011 at 2:11 AM, Alex Dean <a...@crackpot.org> wrote: > > On Dec 14, 2010, at 9:48 AM, Jeff Rose wrote: > > > > > Hi Alex, > > > Thanks for the response. We looked into it a bit more based on your > > > suggestion, and we discovered that there is an error occurring when the > > > continuation is resumed in onMessageAvailable in AjaxListener.java. It > > > looks like this onMessageAvailable handler is being called every time a > new > > > message arrives on the queue, and then it is trying to resume a > continuation > > > that has already been resumed, which causes an illegal state exception. > > > We've been trying to fix the bug by checking first: > > > > > > !continuation.isResumed() > > > > > > but we are still getting the error occasionally (although it does often > get > > > all of the messages now). This seems to be a pretty fundamental bug in > the > > > AJAX support of ActiveMQ. Definitely worth posting a ticket (although > I > > > don't know where that is done). > > > > Jeff: I attached a patch to AMQ-3094 which I think fixes this problem. A > continuation can be in more states than just initial & resumed, so checking > !continuation.isResumed() can still result in illegal state exceptions. I > changed your suggestion so AjaxListener only tries to resume a continuation > "if( continuation.isSuspended() )". Otherwise it buffers the message for > delivery in the next poll (building on some code Dejan committed a few days > ago). > > > > Let me know if you have time to test this patch, and if it solves the > issue for you. It looks good in my tests so far. > > > > > https://issues.apache.org/jira/secure/attachment/12467757/all_messages_are_delivered_to_ajax_clients.patch > > > > Hi Alex, > > I'm not able to apply your patch, either to the latest from SVN or to > revision 1052368 which is specified in the patch. For example, in > MessageListenerServlet.java I don't have any reference to the > unconsumedMessages linked list. Can you generate the patch against the > latest from SVN so I can give it a try? I've got everything setup again to > test with my initial PHP and javascript configuration. > > Sure. I'll update to trunk HEAD and re-send the patch. > > unconsumedMessages was added to AjaxListener in r1052259, so if you've got > HEAD you should have it. > http://svn.apache.org/viewvc?view=revision&revision=1052259 > > Updated patch (now based on trunk r1057227) is at > https://issues.apache.org/jira/secure/attachment/12467892/all_messages_are_delivered_to_ajax_clients.2.patch > > alex Nice, this is working much better. Now I can see that responses are sometimes spread across multiple <ajax-response> elements, and I'm not getting random dropped messages like before. The last issue I do see is that fairly often the consumer stops listening to the topic even though the ajax poll requests are still coming. I can see in jconsole that the consumer count on my topic drops to zero, even though at the same time I'm watching empty ajax responses coming back periodically. -Jeff
