[jira] [Created] (AMBARI-5890) HDFS High Availability Provisioning Fails

["Dmitry Lysnichenko (JIRA)" <ji...@apache.org>]

Dmitry Lysnichenko created AMBARI-5890:
------------------------------------------

             Summary: HDFS High Availability Provisioning Fails
                 Key: AMBARI-5890
                 URL: https://issues.apache.org/jira/browse/AMBARI-5890
             Project: Ambari
          Issue Type: Bug
    Affects Versions: 1.6.1
            Reporter: Dmitry Lysnichenko
            Assignee: Dmitry Lysnichenko
            Priority: Blocker
             Fix For: 1.6.1




When attempting to configure an existing cluster for HDFS high availability, the "Final HA Setup" step of "Start All Services" fails because of:

Invalid transition for servicecomponent, clusterName=c1, clusterId=2, serviceName=HDFS, componentName=SECONDARY_NAMENODE, currentDesiredState=INIT, newDesiredState=STARTED

Reproduction Steps
1) Configure a cluster with enough hosts to support HA mode (HDFS, Ganglia, ZooKeeper, Nagios; bare minimum)
2) After configuration of the cluster, begin HA setup for HDFS
3) Follow the various prompts and stages; everything will work.
4) When you are on the last section, "Finalize HA Setup", the "Start All Services" will fail.

It seems like the new backup namenode is still in the "INIT" state in the database. As such, it cannot be transitioned to STARTED since only components in the "INSTALLED" state can move to this desired state.

If you manually change the value in the database from INIT to INSTALLED, and then retry "Start All Services" then the services are actually started correctly. This indicates that the workflow has indeed installed the namenode correctly, but the database was not updated.




--
This message was sent by Atlassian JIRA
(v6.2#6252)