Thanks for the info Gordon. I was able to make a HA cluster work, but I found that all of my clients (and the qpid-* tools) would not connect properly when I used a round robin dns entry to advertise all of the brokers. I thought they were supposed to fail to connect and then retry, but my logs looked like things were connecting, authenticating, and then getting disconnected from non-primary brokers and never trying again. I solved this by going to a rgmanager virtual ip, but still seems odd to me. And it contradicts all of the documentation that I could find... I'm running this on Centos 6.4 if it matters.
Otherwise the various docs were good, but it'd be nice to have everything consistent. I.e. remove references to active-active clusters from 0.20, and add a note the recently posted "replacing old cluster with HA" to make it obvious that the old cluster was actually removed in .20, and you need to use HA there, not just in 0.22. -Darrell On May 14, 2013, at 4:45 AM, Gordon Sim <[email protected]> wrote: > On 05/13/2013 05:22 PM, Darrell Budic wrote: >> A couple of list newbie questions about clustering in 0.20: >> >> So I figured out that the old "active-active" clustering was removed >> from 0.20, which took a little while since all the documentation on >> the web site >> (http://qpid.apache.org/books/0.20/AMQP-Messaging-Broker-CPP-Book/html/chap-Messaging_User_Guide-Active_Active_Cluster.html) >> still references it. > > Sorry for the frustration! > >> I wasn't able to find any information on why this happened on the >> website or users lists, what's the story? > > There are two parts to the answer. The first is that it was being replaced > anyway. The original active-active design required complete synchronization > of all broker state and this made it very fragile to changes in unexpected > ways. It was also limited in its scalability as the bulk of the work for each > node was done on a single thread. > > The second part is that some significant refactoring was required in order to > get AMQP 1.0 support (plus fix some long outstanding structural issues). > Keeping the original cluster through this would have added significant effort > while also destabilising the cluster code. > > I therefore suggested removing the feature (along with some others) in a > thread on the user list last year: > http://qpid.2158936.n2.nabble.com/proposal-to-remove-certain-features-from-qpidd-tt7579928.html > > I do apologise for the inconvenience this change presents. > >> Are there any options for Active-Active clustering I'm missing? Is >> the intent for HA load distribution now federated groups only? > > One point to mention is that in practice the load-balancing offered by the > active-active solution was limited. All traffic was replicated to all nodes > and processed by a single thread at each node. > >> We've been running 0.14 with 3 node active-active clusters, are there >> any migration guides available to an equivalent setup in 0.20? > > Alan wrote some notes to that purpose: > https://svn.apache.org/repos/asf/qpid/trunk/qpid/cpp/README-HA.txt. Any > comments or feedback will I'm sure be gratefully received. > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
