Hi Alexander,

I didn’t do extensive HDFS testing that I didn’t encounter those JIRAs in the 
Confluence page.  From a smoketest standpoint it did seem to work.

 

Allan

 

 

From: Alexander Batyrshin <[email protected]>
Date: Saturday, October 5, 2019 at 6:39 PM
To: Anu Engineer <[email protected]>
Cc: Allan Espinosa <[email protected]>, <[email protected]>
Subject: Re: Kubernetes integration for HDFS

 

ENV JAVA_HOME /usr/lib/jvm/jre-11-openjdk

Is Hadoop OK with this version?
As far i can see - this version is not compatible 
https://cwiki.apache.org/confluence/display/HADOOP/Hadoop+Java+Versions
--
Alexander Batyrshin aka bash
Biomechanical Artificial Sabotage Humanoid

 

 

On Fri, 20 Sep 2019 at 08:22, Anu Engineer <[email protected]> 
wrote:

That is very interesting. A sub-project of Hadoop called Ozone tests regularly 
with Kubernetes. If you would like to contribute your work to  Hadoop, you can 
send us a pull request and we can maintain this code as a sample under Ozone 
K8s samples directory. An advantage with that approach is that this will get 
regularly tested. Please note that K8s support is very early / experimental 
under Ozone at this point of time.

 

-- Anu

 

 

On Thu, Sep 19, 2019 at 7:30 PM Allan Espinosa <[email protected]> wrote:

Hello folks,

I just wanted to share a PoC of running HDFS on Kubernetes that I hacked on 
over the summer.

https://github.com/aespinosa/cloud-hadoop

The above repo introduces a "hdfs kube_namenode" command where the namenode 
formats itself or runs "-bootstrapStandby" depending on the state of 
provisioning an HDFS cluster.  The state information is stored inside 
Kubernetes as metadata annotations to the manifest .yml of running HDFS.

With this, it is simpler to bring up HA sandbox HDFS clusters without the need 
for externally orchestrating namenode and zkfc formatting.

Hope others find this useful.

Regards,
Allan



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to