12 Replies Latest reply on Jun 9, 2004 1:21 PM by hbaxmann

    Wrong Multicast IP UDP address

    hbaxmann

      I am starting JBoss 3.2.4 out-of-the-box on WinXP with two NICs.
      Want to have JBoss on 192.168.1.102, other NIC is 192.168.1.100.

      run -c all -b 192.168.1.102

      TcpView says:

      java.exe:4120 UDP 192.168.1.100:4570 *:*
      java.exe:4120 TCP 192.168.1.102:1098 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:1099 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:1100 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:3528 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:4444 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:4445 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:4571 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8009 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8080 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8083 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8090 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8092 0.0.0.0:0 LISTENING
      java.exe:4120 TCP 192.168.1.102:8093 0.0.0.0:0 LISTENING

      Log says:

      07:40:06,094 INFO [DefaultPartition] Initializing
      07:40:06,325 INFO [UDP] unicast sockets will use interface 192.168.1.100
      07:40:06,335 INFO [UDP] socket information:
      local_addr=Tabletti:4570 (additional data: 18 bytes), mcast_addr=228.1.2.3:45566, bind_addr=/192.168.1.100, ttl=32
      socket: bound to 192.168.1.100:4570, receive buffer size=150000, send buffer size=800000
      multicast socket: bound to 192.168.1.100:45566, send buffer size=800000, receive buffer size=150000
      07:40:06,335 INFO [STDOUT]
      -------------------------------------------------------
      GMS: address is Tabletti:4570 (additional data: 18 bytes)
      -------------------------------------------------------
      07:40:08,388 INFO [DefaultPartition] Number of cluster members: 1
      07:40:08,398 INFO [DefaultPartition] New cluster view (id: 0, delta: 0) : [192.168.1.102:1099]
      07:40:08,408 INFO [DefaultPartition] Other members: 0
      07:40:08,408 INFO [DefaultPartition] Fetching state (will wait for 60000 milliseconds):
      07:40:08,438 INFO [DefaultPartition:ReplicantManager] Dead members: 0
      07:40:08,708 INFO [HANamingService] Listening on /192.168.1.102:1100
      07:40:08,708 INFO [DetachedHANamingService$AutomaticDiscovery] Listening on /192.168.1.102:1102, group=230.0.0.4, HA-JNDI address=1
      92.168.1.102:1100


      I do not expect listening anything on 192.168.1.100.

      The Testsuite (build testsuite) has the same expectation ;-)

      bax

        • 1. Re: Wrong Multicast IP UDP address
          belaban

          JGroups conveniently ignores -b, you have to set it manually in cluster-service.xml. Add a bind_addr=x.x.x.102 to the UDP properties.

          I just changed JGroups, and JBoss head and 3.2 so JGroups actually does use jboss.bind.address if it is set (via -b or --host). This will be in the next release.

          Bela

          • 2. Re: Wrong Multicast IP UDP address
            hbaxmann

            Hi Bela,

            thanks a lot, I was hoping to demonstrate this kind of functionality to a customer, than he has to wait till the next (3.2.5 ??) release :-(

            Any timeframe for this?

            wbr
            bax

            • 3. Re: Wrong Multicast IP UDP address
              hbaxmann

              Looks ok so far. But why does JGroups rely on a reverse DNS lookup:

              18:10:26,918 INFO [DefaultPartition] Initializing
              18:10:27,128 INFO [UDP] unicast sockets will use interface 192.168.1.102
              18:10:27,128 INFO [UDP] socket information:
              local_addr=tabletti:1479 (additional data: 18 bytes), mcast_addr=228.1.2.3:45566, bind_addr=/192.168.1.102, ttl=32
              socket: bound to 192.168.1.102:1479, receive buffer size=150000, send buffer size=800000
              multicast socket: bound to 192.168.1.102:45566, send buffer size=800000, receive buffer size=150000
              18:10:27,138 INFO [STDOUT]
              -------------------------------------------------------
              GMS: address is tabletti:1479 (additional data: 18 bytes)
              -------------------------------------------------------

              thank you

              bax

              • 4. Re: Wrong Multicast IP UDP address
                hbaxmann

                ... because on the 2nd one (X.X.X.102) I get endless:

                18:23:27,571 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1647
                18:23:27,661 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1647
                18:23:43,293 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1647

                anything wrong with this?

                bax

                • 5. Re: Wrong Multicast IP UDP address
                  belaban

                   

                  "hbaxmann" wrote:
                  Hi Bela,

                  thanks a lot, I was hoping to demonstrate this kind of functionality to a customer, than he has to wait till the next (3.2.5 ??) release :-(

                  Any timeframe for this?

                  wbr
                  bax



                  You can use -DUDP.bind_addr for now, e.g. provide your own script (copy of run.sh). The change will be part of 3.2.5.
                  Bela

                  • 6. Re: Wrong Multicast IP UDP address
                    belaban

                     

                    "hbaxmann" wrote:
                    Looks ok so far. But why does JGroups rely on a reverse DNS lookup:

                    18:10:26,918 INFO [DefaultPartition] Initializing
                    18:10:27,128 INFO [UDP] unicast sockets will use interface 192.168.1.102
                    18:10:27,128 INFO [UDP] socket information:
                    local_addr=tabletti:1479 (additional data: 18 bytes), mcast_addr=228.1.2.3:45566, bind_addr=/192.168.1.102, ttl=32
                    socket: bound to 192.168.1.102:1479, receive buffer size=150000, send buffer size=800000
                    multicast socket: bound to 192.168.1.102:45566, send buffer size=800000, receive buffer size=150000
                    18:10:27,138 INFO [STDOUT]
                    -------------------------------------------------------
                    GMS: address is tabletti:1479 (additional data: 18 bytes)
                    -------------------------------------------------------

                    thank you

                    bax



                    Okay, just for you: :-)
                    I added a system property resolve.dns, which disables DNS resolving. By default it is "true" (resolving enabled).
                    Bela

                    • 7. Re: Wrong Multicast IP UDP address
                      belaban

                       

                      "hbaxmann" wrote:
                      Looks ok so far. But why does JGroups rely on a reverse DNS lookup:

                      18:10:26,918 INFO [DefaultPartition] Initializing
                      18:10:27,128 INFO [UDP] unicast sockets will use interface 192.168.1.102
                      18:10:27,128 INFO [UDP] socket information:
                      local_addr=tabletti:1479 (additional data: 18 bytes), mcast_addr=228.1.2.3:45566, bind_addr=/192.168.1.102, ttl=32
                      socket: bound to 192.168.1.102:1479, receive buffer size=150000, send buffer size=800000
                      multicast socket: bound to 192.168.1.102:45566, send buffer size=800000, receive buffer size=150000
                      18:10:27,138 INFO [STDOUT]
                      -------------------------------------------------------
                      GMS: address is tabletti:1479 (additional data: 18 bytes)
                      -------------------------------------------------------

                      thank you

                      bax



                      Okay, just for you: :-)
                      I added a system property resolve.dns, which disables DNS resolving. By default it is "true" (resolving enabled).
                      Bela

                      • 8. Re: Wrong Multicast IP UDP address
                        hbaxmann

                        You are as kind with me as my mother was :-))

                        Thanks a lot.

                        [ rant ]
                        But unfortunately, putting the other quaintnesses of the build&test aside, I am disappointed. Is it not normal to have an exceptable result for an mostly out-of-the-box used product? If yes where are they and which is the canon?

                        At the very moment, I am relatively alone with 153 errors (unanticipated !) in the reference environment.
                        And I know what I am doing here - since then.

                        [ rant ]

                        could not resist

                        bax

                        • 9. Re: Wrong Multicast IP UDP address
                          hbaxmann

                          Now, cannot aplly your patch to the release version, I get 92 errors in the testsuite results. Manly brought to us by "Connection refused".

                          Is it a bug in the test.

                          The whole test gives me a failure rate of 6%, with my policy not even taggable for release process (<5%).

                          nice try

                          bax

                          • 10. Re: Wrong Multicast IP UDP address
                            hbaxmann

                            Another issue:

                            there is a
                            00:49:43,596 INFO [STDOUT]
                            -------------------------------------------------------
                            GMS: address is tabletti:1556 (additional data: 18 bytes)
                            -------------------------------------------------------
                            00:49:45,659 INFO [DefaultPartition] Number of cluster members: 2

                            versus:

                            00:50:04,156 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1534
                            00:50:04,166 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1534
                            00:50:21,170 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1534
                            00:50:21,260 WARN [UDP] discarded message from different group (TreeCache-Cluster). Sender was Tabletti:1534

                            so i think the nslookup ( getHostName() ) is case sensitive, which should not be appropriate ?
                            If I change the DNS to "Tabletti" all things go smoothly ... so far.

                            bax

                            • 11. Re: Wrong Multicast IP UDP address
                              belaban

                               

                              "hbaxmann" wrote:
                              You are as kind with me as my mother was :-))

                              Thanks a lot.

                              [ rant ]
                              But unfortunately, putting the other quaintnesses of the build&test aside, I am disappointed. Is it not normal to have an exceptable result for an mostly out-of-the-box used product? If yes where are they and which is the canon?

                              At the very moment, I am relatively alone with 153 errors (unanticipated !) in the reference environment.
                              And I know what I am doing here - since then.

                              [ rant ]

                              could not resist

                              bax


                              come again ? I don't know what you are talking about ...
                              Bela

                              • 12. Re: Wrong Multicast IP UDP address
                                hbaxmann

                                Bela,

                                this is what I am really afraid of. But you are not alone - the thread in Inst&Config died silently. As all these about build&test.

                                I am talking about having potential users of JBoss to show how unnecessary it is to get the CVS HEAD or post-Relese tag out of the repository and that there is no necessity to build&test and trying to interpret the whole stuff.

                                I am lost - I have lost.

                                ok, I am a loser so far ...

                                an angry one

                                bax
                                PS: I am not talking about simplest runs of the testsuite which is the highest level the Q&A must use. And even than there are failures and errors remaining. We know all which one the only one JBoss distribution was with zero failures and errors.