RE: Failing JMS Continuations test (Was : Re: Back to normal.....)
Thanks Dan for finding the actual issues with the ConduitSelector. One thing this test has confirmed is that the way JMS continuations are handled on the server side is pretty rock solid... Cheers, Sergey -Original Message- From: Daniel Kulp [mailto:dk...@apache.org] Sent: 03 September 2009 18:54 To: dev@cxf.apache.org Cc: Sergey Beryozkin; Benson Margulies Subject: Re: Failing JMS Continuations test (Was : Re: Back to normal.) OK. I'm pretty sure Sergey and I tracked this down to some thread safety issues in the ConduitSelector. However, it was made even worse on 2.2.x/2.1.x due to some additional thread safety issues in the JMSConduit on those branches. (trunk has a completely updated conduit that doesn't suffer the problem) Anyway, I'm hoping the next round of hudson builds will be good. :-) Dan On Thu September 3 2009 6:41:03 am Sergey Beryozkin wrote: > >> > There is a "randomly" failing continuations test that I've asked > >> > Sergey to look at, but it's failing on on the branches. If he cannot > >> > find a fix tomorrow, I'll @Ignore it for a bit. > > Looking into it now... The initial observation is that it is always green > if the test server (Server2) is started in the in-process mode, all the 5 > client threads get their expected responses back after firing at the same > time but it fails as soon as the Server2 is launched in a seperate > process... Not quite sure yet what does it indicate at... > > cheers, Sergey > -- Daniel Kulp dk...@apache.org http://www.dankulp.com/blog
Re: Failing JMS Continuations test (Was : Re: Back to normal.....)
OK. I'm pretty sure Sergey and I tracked this down to some thread safety issues in the ConduitSelector. However, it was made even worse on 2.2.x/2.1.x due to some additional thread safety issues in the JMSConduit on those branches. (trunk has a completely updated conduit that doesn't suffer the problem) Anyway, I'm hoping the next round of hudson builds will be good. :-) Dan On Thu September 3 2009 6:41:03 am Sergey Beryozkin wrote: > >> > There is a "randomly" failing continuations test that I've asked > >> > Sergey to look at, but it's failing on on the branches. If he cannot > >> > find a fix tomorrow, I'll @Ignore it for a bit. > > Looking into it now... The initial observation is that it is always green > if the test server (Server2) is started in the in-process mode, all the 5 > client threads get their expected responses back after firing at the same > time but it fails as soon as the Server2 is launched in a seperate > process... Not quite sure yet what does it indicate at... > > cheers, Sergey > -- Daniel Kulp dk...@apache.org http://www.dankulp.com/blog
Re: Failing JMS Continuations test (Was : Re: Back to normal.....)
After messing with setting up a logger for an out-process server (logging.properties were ignored after setting a java.util.logging.config.file property in a server launcher properties, perhaps there's some contention teher between the EmbeddedBroker and Server2 for the same file handler, not really sure), I found that if there was just a single client thread doing an invocation then the test would pass and the server would log/confirm the continuations were successfully created/suspended/etc from inside the application code. Having just 2 client threads results in no invocations coming into the application code at all - sounds like a configuration issue leading to the rejection or even the deadlock...Does anyone have the idea what might be going on ? Here is the config file : http://svn.apache.org/repos/asf/cxf/trunk/systests/uncategorized/src/test/java/org/apache/cxf/systest/jms/continuations/jms_test_config.xml (check the bean with id='jmsConf1', in my local snapshot I increased the values for (max)ConcurrentConsumers) I don't see what other config properties may need to be added (http://cxf.apache.org/docs/using-the-jmsconfigfeature.html) Just for the record here's the client test code : http://svn.apache.org/repos/asf/cxf/trunk/systests/uncategorized/src/test/java/org/apache/cxf/systest/jms/continuations/HelloWorldContinuationsClientServerTest.java and the service one : http://svn.apache.org/repos/asf/cxf/trunk/systests/uncategorized/src/test/java/org/apache/cxf/systest/jms/continuations/HelloWorldWithContinuationsJMS.java cheers, Sergey - Original Message - From: "Sergey Beryozkin" To: Cc: "Benson Margulies" Sent: Thursday, September 03, 2009 11:41 AM Subject: Failing JMS Continuations test (Was : Re: Back to normal.) > There is a "randomly" failing continuations test that I've asked Sergey > to look at, but it's failing on on the branches. If he cannot find a > fix tomorrow, I'll @Ignore it for a bit. Looking into it now... The initial observation is that it is always green if the test server (Server2) is started in the in-process mode, all the 5 client threads get their expected responses back after firing at the same time but it fails as soon as the Server2 is launched in a seperate process... Not quite sure yet what does it indicate at... cheers, Sergey -- Daniel Kulp dk...@apache.org http://www.dankulp.com/blog