[
https://issues.apache.org/jira/browse/AMBARI-21674?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16163279#comment-16163279
]
Nate Cole commented on AMBARI-21674:
------------------------------------
Hi David, this type of initiative has always been a nice-to-have, but we've
seen requests for it drop off in the past. That said, there's a couple of
things to check (please bear in mind it's been a while since I investigated
this):
* Agent Registration. There was a time where we would compare Ambari to the
hosts in the cluster. I don't believe that's the case anymore, but should be
checked.
** The comparison should be "is the OS family known and supported by Ambari."
* JDK Install options. We allow the ability to specify a user-installed JDK.
The assumption of homogenous clusters means the user only needs to specify this
once. With hybrid, we may need a bit more flexibility here.
** What about JDK availability across platforms? Oracle JDK will likely have
consistent versions, but OpenJDK may not - eg is 1.8.0_121 available for all
platforms?
** JCE will probably fall in the same bucket.
* Repository Versions. This will be a big focus area. Today, we have the
{{repo_version}} table. This table contains a few bits of metadata (the name)
but also contains the {{version}}. The specific {{version}} is resolved at
install-time, but (admittedly, badly) is only for one specific operating
system. We do have an effort in 3.0.0 to break this table into its constituent
parts ({{repo_os}}, {{repo_detail}}, and we can move {{version}} to each
{{repo_os}} table instead.
** API implications means that {{version}} in the response today no longer
makes sense, and {{OperatingSystems}} should be updated instead.
** There are UI implications to this too, as we use the version for display.
The UI should reflect the API changes above.
For this effort, I can be your PMC point of contact. Please add me to design
discussions and code reviews as you see fit.
-Nate
> Ambari to support Hybrid cluster
> ---------------------------------
>
> Key: AMBARI-21674
> URL: https://issues.apache.org/jira/browse/AMBARI-21674
> Project: Ambari
> Issue Type: New Feature
> Components: ambari-agent, ambari-server
> Reporter: Ayappan
>
> We have been exploring Hybrid hadoop cluster recently. By hybrid, i mean the
> nodes in the cluster have different platform architectures. Some nodes are
> intel x86_64 , some are IBM Power ppc64le., and so on. Right now Ambari
> doesn't have this hybrid support. It selects the repo based on the machine
> where it is installed and then ask the users to add the nodes. I think this
> order has to be changed as the first step to provide hybrid support in Ambari.
> We were actually able to bring up a hybrid cluster using some hacks (not so
> good) like manually configuring the yum repos according to the arch,
> installing the rpms before ambari does and also manipulating the JAVA_HOME
> variable. We also ran some examples like terasort without any issues.
> So the thing here is we kind of starting to work on making Ambari provide a
> smooth hybrid support which requires some design changes. We very much
> welcome Ambari community to provide some ideas, patch contributions, or
> whatever way you can help us on.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
