-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/34849/
-----------------------------------------------------------
(Updated May 31, 2015, 1:02 a.m.)
Review request for Ambari, Mahadev Konar, Robert Levas, and Sumit Mohanty.
Changes
-------
Further testing with HDFS NameNode HA clusters revealed filtering out the HDFS
NameNode (non-HA) properties in an HA environment appears to break HDFS HA
cluster startups with Blueprints.
I've updated the patch to remove the filter for HDFS Namenode properties, and
left the HBase/Hive filtering intact. I'll re-introduce the HDFS filter in a
separate patch once I've determined the cause of the stack script failures I'm
seeing, since the scripts seem to depend upon these non-HA properties in HA
mode.
Bugs: AMBARI-11563
https://issues.apache.org/jira/browse/AMBARI-11563
Repository: ambari
Description
-------
This patch addresses AMBARI-11563.
The Blueprints processor currently has no notion of "optional" properties in a
cluster's configuration. The Blueprints processor uses the stack definitions to
determine the set of properties to include and extend in the process of
deploying a cluster.
There are some properties in HBase and Hive that are "conditional" in nature.
An example of this is the "hbase.coprocessor.regionserver.classes" config
property in "hbase-site". This property should only be set when HBase
authorization is configured to be enabled.
Currently, a Blueprints deployment of HBase will set this property to the
default, empty value, and this will be included in the Cluster configuration.
This causes problems after the cluster deployment completes, since the Ambari
UI will display this property as required, even through authorization is not
configured to be enabled in this cluster.
The Blueprints configuration processor needs to be modified to exclude any
properties that don't meet certain conditions:
1. HBase Authorization (on/off)
2. Hive Authentication (on/off)
3. HDFS NameNode properties that are only useful in a non-HA scenario must be
filtered out in an HDFS NameNode HA deployment.
This patch addresses this problem by:
1. Updating the Stack class used by Blueprints to read in the new property
dependencies defined in the stack
2. Extending the existing PropertyFilter interface in the
BlueprintConfigurationProcessor, such that it can now be used to filter out
properties during the cluster startup phase, as well as a Blueprint export.
3. Implemented filters to detect the HBase Authorization, Hive Authentication,
and HDFS NameNode HA cases described above.
4. Adds unit tests to verify this change.
Diffs (updated)
-----
ambari-server/src/main/java/org/apache/ambari/server/controller/internal/BlueprintConfigurationProcessor.java
6fb4807
ambari-server/src/main/java/org/apache/ambari/server/controller/internal/Stack.java
38b5a3c
ambari-server/src/test/java/org/apache/ambari/server/controller/internal/BlueprintConfigurationProcessorTest.java
1de2b29
ambari-server/src/test/java/org/apache/ambari/server/controller/internal/StackTest.java
3df9ec1
Diff: https://reviews.apache.org/r/34849/diff/
Testing
-------
1. Run the Blueprints-specific unit tests with this patch applied, all passing.
(I plan on running the full ambari-server unit test suite prior to merging).
2. Deployed a single-node HBase and Hive cluster with
authentication/authorization turned off, and verified that the properties are
not visible or required in the Ambari UI. Verified that the cluster REST
resource reports the same results.
3. Unit tests for these fixes cover the various cases (Hive Authentication
on/off, HBase Authorization on/off, etc).
Thanks,
Robert Nettleton
