On Fri, 3 Aug 2018 at 19:18, Dyslin, Mike <mike.dys...@hpe.com> wrote:
> Rob, > > Our consumer client version is 1.37: > > bash 0 3: rpm -qa | grep -i qpid > qpid-cpp-client-devel-1.37.0-1.el7.x86_64 > python-qpid-1.37.0-1.el7.noarch > qpid-qmf-1.37.0-1.el7.x86_64 > qpid-cpp-client-1.37.0-1.el7.x86_64 > python-qpid-qmf-1.37.0-1.el7.x86_64 > qpid-cpp-server-1.37.0-1.el7.x86_64 > qpid-proton-c-0.18.1-1.el7.x86_64 > qpid-tools-1.37.0-1.el7.noarch > > Our producer client version is 6.1.5: > -rw-r--r-- 1 NSDA.NSDA NSDA 570873 Jul 26 15:42 > qpid-client-6.1.5.jar > -rw-r--r-- 1 NSDA.NSDA NSDA 864493 Jul 26 15:42 > qpid-common-6.1.5.jar > > Both producer and consumer clients use AMQP_0_10 protocol. > OK - so my assumption here is that there is some issue in the management of credit in the broker (or possibly in the client). AMQP 0-10 has two distinct credit flow modes "credit" and "window" and also allows the consumer to separately set limits for both "message" and "byte" credit. I'm not very familiar with the C++ API, but do you know how your consumer is managing credit? One thing that would be very helpful in trying to diagnose this problem is getting protocol logging for the consumer (at least for the start of the consumer where it sets up the credit flow mode, and towards the end where it would be interesting to see the credit being allocated just before message flow stops). -- Rob > > Thanks for your assistance, > Mike > > FYI - There may be a typo on the past releases web page. I believe "2017" > should be "2018" in "Qpid JMS AMQP 0-x 6.3.2, July 2017". URL: > https://qpid.apache.org/releases/index.html#past-releases > > -----Original Message----- > From: Rob Godfrey [mailto:rob.j.godf...@gmail.com] > Sent: Thursday, August 2, 2018 5:57 PM > To: users@qpid.apache.org > Cc: Mears, David B <david.me...@hpe.com>; Herren, Elaine < > elaine.her...@hpe.com>; Rao, Shobha (NonStop) <shobha.jayatheer...@hpe.com > > > Subject: Re: Java Broker (7.0.6) stops delivering queue/consumer messages > after 4 GB data transfer > > Hi Mike, > > On Fri, 3 Aug 2018 at 01:25, Dyslin, Mike <mike.dys...@hpe.com> wrote: > > > This is my first submit to this email group. Hopefully this is the > > correct place to post this problem. > > > > > This is exactly the right place to post this problem. > > > > We are running a continuous stream of message (about 5K each) from > > producer to consumer over a single java broker queue at a rate of > > about 600 messages/second. Outbound message flow stops after > > transferring 4 GB of message data (about 770,000 messages in 25 > > minutes). The Web Management Console page for our consumer connection > shows the total "Outbound Bytes" > > growing steadily until it reaches 4.0 GB and stops with "Last I/O time" > > unchanging thereafter. > > > > After outbound messages stop: > > Inbound messages continue on the producer connection (well past 4.0 > > GB) and are kept in the queue until they expire with a time-to-live > > value of 3 minutes. The queue grows until is stabilizes with a steady > > 600 m/s inbound, and 600 m/s expiring and being deleted from the queue > > (as expected). The Web Management Console shows that the consumer > > connection remains open and is a consumer on the queue, and the queue > > shows the connection as a consumer on the queue. > > > > If I run the exact same test replacing the Java Broker with a C++ > > broker (1.37.0), message flow continues well past the 4 GB barrier. I > > kept it running for about 17 hours reaching about 37 million messages, > > about 180 GB data transferred on the queue. > > > > Since the only difference seems to be the broker, this seems to point > > to a problem with the Java Broker, and not issues with our producer, > > consumer or network issues. Could there be some problem with our java > > broker configuration that would explain this behaviour? > > > > Unfortunately this sounds like it may be a bug in the Java Broker :-( > > > > > > Has anyone out there experienced more than 4 GB of outbound data on a > > single java broker connection or queue? > > > > > Can you confirm which client you are using, and which version of AMQP is > in use (as you have identified I don't expect this to be a client problem, > but knowing the client will help us track down the issue in the broker)? > > > > Any help would be appreciated. > > > > Other comments/observations: > > > > I do not know if the 4 GB barrier is associated with the connection > > and/or the queue because all our message traffic is over one consumer > > connection and one queue. I could determine this by changing our > > consumer code to spread message traffic over one connection and multiple > queues. > > > > > > > > We are using the heartbeat feature with a 5 minute timeout. Since the > > connection stays open beyond the 5 minute timeout after the messages > > stop, I assume the heartbeat messages are still being sent between > > consumer and broker, indicating that the consumer and broker are able > > to communicate over the socket. It has been awhile since I have > > tested that the heartbeat feature is working correctly. > > > > If I close the consumer connection from the Web Management Console, > > the broker deletes the queue (I believe) and our consumer detects the > > closed connection, establishes a new connection and new queue, and > > messages start flowing again until . . . we reach the 4 GB barrier and > > messages stop being delivered once again. > > > > We have run with the Java Broker on both Linux (RHEL 7.4) and > > proprietary NonStop POSIX platform with the same results. > > Unfortunately, the C++ broker is not yet an option on the NonStop > > POSIX platform where we require the broker to be. > > > > > Hopefully we can quickly track down the issue in the Java Broker and push > out a fix, > > > > Thanks, > > Mike > > > > > Apologies you've run into this issue, > Rob > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: users-unsubscr...@qpid.apache.org For > > additional commands, e-mail: users-h...@qpid.apache.org > > > > >
