----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/41925/#review113052 -----------------------------------------------------------
My concern with StageUtils.getClusterInfo() is that if I understand correctly hostinfos retrived from topologyManager.getProjectedTopology should be used only in case if wea re in a middle of setting up tasks / stages for a BP provisioning so in the context of HostRequest.InstallTask.run(), HostRequest.StartTask.run(). Other than that topologyManager may return false, invalid info about hosts. I'm not sure if this would works but in long term the preferable solution I think would be if we can determine that we are in a context of BP initiated INSTALL / START task. These tasks: HostRequest.InstallTask.run(), HostRequest.StartTask.run() are executed on one thread so a thread context variable would be good and might be usefull in other cases as well, where we had to do something specific in case of BP. - Sandor Magyari On Jan. 5, 2016, 1:47 p.m., Oliver Szabo wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/41925/ > ----------------------------------------------------------- > > (Updated Jan. 5, 2016, 1:47 p.m.) > > > Review request for Ambari, Robert Nettleton, Sandor Magyari, Sumit Mohanty, > and Sebastian Toader. > > > Bugs: AMBARI-14491 > https://issues.apache.org/jira/browse/AMBARI-14491 > > > Repository: ambari > > > Description > ------- > > With Blueprint deployments, zookeeper cannot be installed after manual > upscale (it can happen with any other services too). > That happened because of wrong cluster info. (topology information merged > with cluster information - these are needed especially during bp install) > Added some unit tests too. > > Only 1 case which is still get the wrong cluster info: > - deploy a cluster with bluprint > - downscale (not with the blueprint API) > - number of the remaining nodes are less then the initial in the topology > -> then cluster info will contain the removed nodes (but the correct ones too) > The problem here: with the actual logic, we cannot determine that we did the > steps above, or there is a blueprint deployment in progress. (actually this > problem was present before the HDFS service check fail fix > https://issues.apache.org/jira/browse/AMBARI-12688) > > > Diffs > ----- > > ambari-server/src/main/java/org/apache/ambari/server/utils/StageUtils.java > 30843a4 > > ambari-server/src/test/java/org/apache/ambari/server/utils/StageUtilsTest.java > e1c47b9 > > Diff: https://reviews.apache.org/r/41925/diff/ > > > Testing > ------- > > Unit tests done (SUCCESS). > Functional tests: > - deploy 1 node bp cluster (with zookeeper server). added 1 new host. install > zookeper server -> success. > - deploy 3-node cluster (namenode HA) -> move namenode -> cluster info is > correct (service check contains the correct nodes) > - basically the newly added unit tests cover most of the cases. > > > Thanks, > > Oliver Szabo > >
