Robert Nettleton created AMBARI-9733:
----------------------------------------

             Summary: Exported HDFS HA Blueprint not deploying when used for 
new cluster
                 Key: AMBARI-9733
                 URL: https://issues.apache.org/jira/browse/AMBARI-9733
             Project: Ambari
          Issue Type: Bug
          Components: ambari-server
    Affects Versions: 2.0.0
            Reporter: Robert Nettleton
            Assignee: Robert Nettleton
             Fix For: 2.0.0


Blueprints exported from HDFS HA clusters are not properly working when an 
attempt is made to create a new cluster based on this exported Blueprint. 

Steps to reproduce:

1. Use the Ambari UI to create a 3-node cluster with HDFS enabled
2. Use the Ambari UI HA Wizard to enable HDFS NameNode HA
3. Export a Blueprint from the running cluster using the REST API
4. Attempt to use this Blueprint to create a new, separate cluster

There are two possible failures that this bug documents: 

1. The cluster startup attempt will fail with a "400" status error, and the 
associated error message will mention that no instances of 
"SECONDARY_NAMESERVER" are found for topology mapping.  This is an error in the 
Blueprint processor, since the "SECONDARY_NAMESERVER" cannot be present in an 
HDFS HA scenario. 

2. The cluster startup attempt will fail with a "400" status error, and the 
associated error message will mention that either no instances of "NAMESERVER" 
are found for topology mapping, or more than one instance has been found.  This 
is actually an error in the handling of an "hbase-site" property, 
"hbase.rootdir", that can refer to an HDFS HA nameservice address in HA mode, 
rather than refer to an individual NameServer.  

The Blueprint Configuration processor needs to be updated to handle these 
HA-specific cases properly.

In a future release, it may make sense to refactor the stack definitions to 
include information about properties that have different meanings/values in an 
HA scenario.  For this release (Ambari 2.0), the fixes should be made in the 
Blueprint configuration processor. 

I'm working on a patch for this now, and will be submitting shortly.  



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to