Uploaded image for project: 'ActiveMQ'
  1. ActiveMQ
  2. AMQ-6553

ReplicatedLevelDB with Zookeeper : ActiveMQ Broker Stop when leader stop



    • Type: Bug
    • Status: Closed
    • Priority: Major
    • Resolution: Invalid
    • Affects Version/s: 5.12.1
    • Fix Version/s: None
    • Component/s: Broker, OSGi/Karaf, Transport
    • Labels:
    • Environment:

      Servicemix 5.5.2
      Activemq-osgi 5.12.1



      I m using replicatedLevelDB to use zookeeper and clustering activemq broker.

      all work fine when all is started.

      here is my configuration

      3 servicemix
      each servicemix have a bundle zookeeper 3.4.6 and a bundle zookeeper-server (my code to start zookeeper server)
      each servicemix have is own activemq broker

      Firslty, i start all my zookeeper-server bundle, and all work fine, i do

      echo stat | nc localhost 2181
      echo stat | nc localhost 2182
      echo stat | nc localhost 2183

      and i have 1 Leader and 2 Followers.

      After i configure each activemq broker with :

      hostname="[I PUT MY IP BUT HIDE IN THIS JIRA]"

      the only thing that change is the bind parameter with different port.

      I start all my broker and all work fine, if i launch the karaf command bstat on my servicemix, i can see my queue/topic on 1 broker, and the 2 others just show me

      smx@zookeeper2> bstat
      Uptime = 46 minutes

      if i stop the master broker, zookeeper elect another one to be the master and it works.

      but if i stop my zookeepr Leader, master broker is stop and another one is up, but when i say broker is stop, bstat don't show me anything anymore on this instance.

      so i have loose 1 broker and it can't be elected anymore.

      here is the log i see:

      2017-01-05 11:22:46,415 | WARN | spatch-DEFAULT-2 | MasterLevelDBStore | 96 - org.apache.activemq.activemq-osgi - 5.12.1 | Unexpected session error: java.io.IOException: Connection reset by peer
      2017-01-05 11:22:46,429 | ERROR | meSize=104857600 | TransportConnector | 96 - org.apache.activemq.activemq-osgi - 5.12.1 | Could not accept connection : java.lang.InterruptedException
      2017-01-05 11:22:46,461 | WARN | spatch-DEFAULT-3 | MasterLevelDBStore | 96 - org.apache.activemq.activemq-osgi - 5.12.1 | Unexpected session error: java.io.IOException: Connection reset by peer

      my question is :

      why, when i stop the zookeeper leader, the master broker is killed?
      why, in this case, we don't wait a little that another zookeeper become the leader and don't stop the master broker, but just reinitialize the connection to the new zookeeper master?


      when, in this case, if its necessary to stop the broker and let zookeeper reelect a new one, why the broker is kill and cannot be reelect by zookeeper anymore? does it possible to let it up to be reelected?

      possibly i miss something in the configuration that can prevent this problem, but if not, how can i solve my problem?





            • Assignee:
              Palex Mançaux Pierre-Alexandre
            • Votes:
              0 Vote for this issue
              2 Start watching this issue


              • Created: