On Wed, 14 Oct 2020 at 15:55, Petrenko, Vadim <vadim.petre...@ns.nl> wrote: > > Hi Ken, > > Yes, we're pre-settling messages already: > connectionfactory.myFactoryLookupSender = > amqp://localhost:5674?jms.presettlePolicy.presettleProducers=true > (dockerized Qpid with an exposed port). > > Making the sending client first check for available credits is a possibility, > but not the ideal option we'd like to use. These clients are not in our > control, it's other teams that develop their applications using our Qpid > network. So we can't really enforce them. Only advise. >
The JMS client can't do that anyway. > I also noticed some strange behavior which I don't understand. Maybe you guys > can explain. > I've taken the Sender and Receiver from qpid-jms-examples, connected them to > our router network and observed how this works. > 1. When I start the Sender, it gets blocked because there is no receiver (in > the logs I see "Holding Message send until credit is available"). This is as > expected. > 2. I start the Receiver, messages begin to flow, all good. > 3. I kill the Receiver process, but the Sender is still sending messages. To > nowhere. It doesn't timeout either. It's sending them to the router, which it still has outstanding credit to do. It knows nothing of the receiver. It won't time out a send unless you configure it to, and note it certainly won't time out in this case if it has credit to send and is sending pre-settled messages, since there is also no response to wait for. > > I then proceeded debugging AmqpFixedProducer, which works with credits, and > noticed that on step 1 number of credits it gets via > getEndpoint().getCredit() is 0. So it holds the message. > On the next step Receiver gives it 250 credits. After that, even though the > Receiver process is killed, the number of credits is still 250 and doesn't > get decremented. The Sender continues to send. I expect that credits would be > exhausted at some point. > Is this as it should be? Are you sure it 'doesn't get decremented' (which happens locally as sends occur), or are you perhaps just not observing the credit is being replenished by the router? You can see the protocol trace on router and/or the client by running them with env variable PN_TRACE_FRM=1 set (or also for client, see: http://qpid.apache.org/releases/qpid-jms-0.54.0/docs/index.html#logging) As you are using message-routing, the router itself is what grants the sender credit to send. The receiver also grants credit but that is to the router, and is quite independent of the credit the sender will see from the router (credit would be aligned end to end for a link-routing case, e.g through router to/from a broker). https://issues.apache.org/jira/browse/DISPATCH-1090 is the last I can think of the behaviour in this area being tweaked, which was to try and drain the outstanding credit in similar cases, however in checking the vicinity of those changes now, I see edge routers being called out and handled differently, indicating it will replenish the credit for such messages. Your other thread mentioned edge routers so I think perhaps this is in play here: https://github.com/apache/qpid-dispatch/blob/1.14.0/src/router_core/transfer.c#L467-L476 > > Thanks. > > > -----Original Message----- > From: Ken Giusti <kgiu...@redhat.com> > Sent: woensdag 14 oktober 2020 15:18 > To: users <users@qpid.apache.org> > Subject: Re: Non-blocking Sender with no Receivers > > On Wed, Oct 14, 2020 at 8:25 AM Petrenko, Vadim <vadim.petre...@ns.nl> > wrote: > > > Good day, > > > > We’re trying to implement our Qpid dispatch router network in such a > > way that a sender can connect to it at any moment and start sending > > messages to a topic. > > There might be no receivers at this moment, so the messages would be > > just lost. This is ok. > > Then an any moment a receiver(s) can connect and start receiving > > messages currently being sent, then drop off. The sender will just > > continue sending messages, probably into the void. > > > > It’s about sending data from sensors. Sensor data is short living and > > there is no need to buffer it or persist. If no one consumes this data > > in realtime, it can be lost. > > Senders and receivers in our situation are quite volatile, they appear > > and drop off. > > > > As per docs of Qpid: > > === > > Dispatch Router uses a credit-based flow control mechanism to ensure > > that producers can only send messages to a router if at least one > > consumer is available to receive them. Because Dispatch Router does > > not store messages, this credit-based flow control prevents producers > > from sending messages when there are no consumers present. > > === > > > > I was wondering, maybe there is still a workaround or a trick to > > implement the router network the way I described in the beginning? > > If possible without extra brokers and only with Qpid dispatch routers? > > > > Thanks. > > > > > Can your sending client first check whether or not there is credit on the > link and if there is no credit simply drop the message rather than sending it? > Getting the value of available credit depends on the client API you are > using. For example in Proton-C your sender would call > pn_link_credit(sending_link_ptr) to get the credit for the sending link. > > Also keep in mind that since the application is OK with messages being > dropped you'll want the sender to settle the message after it is sent (a.k.a. > "pre-settled). > > > > > > > ________________________________ > > > > Deze e-mail, inclusief eventuele bijlagen, is uitsluitend bestemd voor > > (gebruik door) de geadresseerde. De e-mail kan persoonlijke of > > vertrouwelijke informatie bevatten. Openbaarmaking, vermenigvuldiging, > > verspreiding en/of verstrekking van (de inhoud van) deze e-mail (en > > eventuele bijlagen) aan derden is uitdrukkelijk niet toegestaan. > > Indien u niet de bedoelde geadresseerde bent, wordt u vriendelijk > > verzocht degene die de e-mail verzond hiervan direct op de hoogte te > > brengen en de e-mail (en eventuele bijlagen) te vernietigen. > > > > Informatie vennootschap<http://www.ns.nl/emaildisclaimer> > > > > > -- > -K > > ________________________________ > > Deze e-mail, inclusief eventuele bijlagen, is uitsluitend bestemd voor > (gebruik door) de geadresseerde. De e-mail kan persoonlijke of vertrouwelijke > informatie bevatten. Openbaarmaking, vermenigvuldiging, verspreiding en/of > verstrekking van (de inhoud van) deze e-mail (en eventuele bijlagen) aan > derden is uitdrukkelijk niet toegestaan. Indien u niet de bedoelde > geadresseerde bent, wordt u vriendelijk verzocht degene die de e-mail verzond > hiervan direct op de hoogte te brengen en de e-mail (en eventuele bijlagen) > te vernietigen. > > Informatie vennootschap<http://www.ns.nl/emaildisclaimer> --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@qpid.apache.org For additional commands, e-mail: users-h...@qpid.apache.org
