-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/46544/#review130117
-----------------------------------------------------------


Ship it!




Ship It!

- Robert Nettleton


On April 21, 2016, 10:24 p.m., Jayush Luniya wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/46544/
> -----------------------------------------------------------
> 
> (Updated April 21, 2016, 10:24 p.m.)
> 
> 
> Review request for Ambari, Alejandro Fernandez, Robert Nettleton, and Sumit 
> Mohanty.
> 
> 
> Bugs: AMBARI-16028
>     https://issues.apache.org/jira/browse/AMBARI-16028
> 
> 
> Repository: ambari
> 
> 
> Description
> -------
> 
> 1. During Namenode HA blueprint deployment, we configure the name nodes to 
> start in active/standby mode based on the following properties
>      {
>         "hadoop-env": {
>           "properties" : {
>             "dfs_ha_initial_namenode_active" : "host1",
>             "dfs_ha_initial_namenode_standby" : "host2”
>           }
>         }
>       }
> 2. The current logic is to always bootstrap the name node marked as standby.
> 3. This will lead to the Namenode marked as Standby to never start under the 
> following situation
> 
> - Cluster is deployed successfully
> - Both name nodes are stopped
> - Start the name node marked as standby. Namenode will never start.
> - This is because the standby name node will try to bootstrap again.
> - However to bootstrap a name node an active name node is required. Based on 
> the HDFS logic the first step done when bootstrapping is to connect to the 
> Active Namenode.
> - Also there is no need to bootstrap here as the name node should already be 
> bootstrapped and should come back up as “Active"
> 
> 
> Fix:
> - The fix is to maintain a bootstrap marker file (similar to the way we keep 
> a name node formatted marker file)
> - In the INITIAL_START phase (during cluster deployment) we will always force 
> bootstrap so as to enforce the name node marked as Standby to wait for the 
> Active name node to come up, bootstrap and start in STANDBY node.
> - Once we are out of INITIAL_START phase, we will bootstrap only if this name 
> node has not been bootstrapped in the past.
> - We will not enforce bootstrapping only in the INITIAL_START phase because 
> there is a possibility during cluster deployment that both name nodes don’t 
> start and hence bootstrapping out of INITIAL_START phase would be required in 
> this case.
> 
> 
> Diffs
> -----
> 
>   
> ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/package/scripts/hdfs_namenode.py
>  8b6c924 
>   
> ambari-server/src/main/resources/common-services/HDFS/2.1.0.2.0/package/scripts/params_linux.py
>  d8ff3c5 
>   ambari-server/src/test/python/stacks/2.0.6/HDFS/test_namenode.py 1c08d57 
> 
> Diff: https://reviews.apache.org/r/46544/diff/
> 
> 
> Testing
> -------
> 
> Verified scenarios on live cluster.
> 
> mvn clean test -DskipSurefireTests
> [INFO] 
> ------------------------------------------------------------------------
> [INFO] BUILD SUCCESS
> [INFO] 
> ------------------------------------------------------------------------
> [INFO] Total time: 54.784s
> [INFO] Finished at: Thu Apr 21 15:22:52 PDT 2016
> [INFO] Final Memory: 64M/1172M
> [INFO] 
> ------------------------------------------------------------------------
> 
> 
> Thanks,
> 
> Jayush Luniya
> 
>

Reply via email to