Hi Joe, Thanks for your time. My application requires a hub & spoke architecture with around 100 publishers & 500 consumers at peak load. All publishers would be pushing messages to different topics. One consumer would be durable as it needs to collect all messages. The rest would be online sporadically (actually they are end-user GUIs & I intend to stream messages in real time to them). So I intended to make them non-durable. What I want is that if my broker goes down, all my consumers should failover to the slave & not loose any messages. I am using NFS based master/slave. I am observing that my non-durable subscribers are losing messages on failover if they are lagging behind the producers. I am concerned about the cost of using 500 durable subscribers. I'll be considering the use of retroactive consumers & virtual destinations. What would be the recommended thing to do in my case? And BTW, the user guide is awesome!
Thanks, Ramit ttmdev wrote: > > Your non-durable subscriber will only get those messages that were > published while it is active. When you bounce the broker it does not > retain the identity of your non-durable subscriber. So when the broker > comes back up, it doesn't consider your non-durable subscriber to have > been active when the persisted messages were published. > > Can you use durable subscribers? If not, look into retroactive consumers > or better yet, virtual topics. With the latter you get the functionality > of both a topic and queue and you don't have to worry about setting up > unique identifiers as you would have to when using durable subscriptions. > > http://activemq.apache.org/virtual-destinations.html > > http://activemq.apache.org/retroactive-consumer.html > > Joe > Goto www.ttmsolutions.com for a free ActiveMQ user guide > > > > Ramit Arora wrote: >> >> Hi All, >> I have a situation where I have to consume persisent messages from many >> non-durable subscribers. My publishers push persistent messages to a >> topic, which are consumed asynchronously by non-durable subscribers. I am >> using the AMQ Persistence Adapter & CLIENT_ACKNOWLEDGE. If I kill a >> non-durable subscriber & bring it up again, some messages are lost, which >> is expected. But consider the following case: >> 1. Start a publisher publishing persistent messages to a topic. >> 2. Start a non-durable subscriber which is slow compared to the >> publisher. >> 3. Kill the broker when 100 messages have been published but only 50 have >> been consumed. >> 4. Since the publisher & subscriber are failover clients, they wait for >> the broker to come up again. >> 5. Restart the broker. >> 6. Try to get the remaining 50 messages to the non-durable subscriber. >> >> Step 6 does not work in both 5.0 & the latest 5.1 snapshot. Shouldn't the >> remaining 50 messages be delivered? The JMS spec says: >> "A JMS provider must deliver a PERSISTENT message once-and-only-once. >> This >> means a JMS provider failure must not cause it to be lost, and it must >> not >> deliver it twice. PERSISTENT (once-and-only-once) and NON_PERSISTENT >> (at-most-once) >> message delivery are a way for a JMS client to select between delivery >> techniques that may lose a messages if a JMS provider dies and those >> which >> take extra effort to insure that messages can survive such a failure. >> There is >> typically a performance/reliability trade-off implied by this choice. >> When a >> client selects the NON_PERSISTENT delivery mode, it is indicating that it >> values performance over reliability; a selection of PERSISTENT reverses >> the >> requested trade-off." >> >> Queues donot suffer from this problem. Please let me know if I am missing >> something. >> >> Thanks, >> Ramit >> >> >> > > -- View this message in context: http://www.nabble.com/Consuming-persistent-messages-from-a-non-durable-subscriber-tp16348721s2354p16350322.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
