Installing on 300+ nodes with Ambari 1.7

[William Slacum <ws...@gmail.com>]

Hi All,

I'm trying to install a hadoop distro via Ambari 1.7 on ~300 nodes. When it
comes time to install all of the services, things get weird (for lack of a
better term). I've noticed the UI will kind of give up, but if I pdsh out
to the clusters, many are still running yum commands. I'm thinking this has
to do with the yum repo I set up being on only node, and http kind of
topping out. Are there any recommendations regarding setting up the yum
repo based on the size of the cluster?

I've also noticed issues with service installations being flagged as
"warnings" (their status is orange), but there's no corresponding error in
the output from the target host. There is stdout output, but it's something
that seems fine such as "Writing file [...] because contents don't match",
and then it ends there. Is there a good set of heuristics to figure out
what caused the status of the install to be orange?

Thanks,
Bill

Re: Installing on 300+ nodes with Ambari 1.7

[Ravi Mutyala <rm...@gmail.com>]

Its mostly yum repo getting overloaded as you noticed. Retry should help
since yum does not try to reinstall the packages that it already installed.
Since repo is a webserver, you can setup a load-balanced webserver serving
yum packages but a retry would be much simpler.

On Thu, Jul 30, 2015 at 3:25 PM, William Slacum <ws...@gmail.com> wrote:

> Hi All,
>
> I'm trying to install a hadoop distro via Ambari 1.7 on ~300 nodes. When
> it comes time to install all of the services, things get weird (for lack of
> a better term). I've noticed the UI will kind of give up, but if I pdsh out
> to the clusters, many are still running yum commands. I'm thinking this has
> to do with the yum repo I set up being on only node, and http kind of
> topping out. Are there any recommendations regarding setting up the yum
> repo based on the size of the cluster?
>
> I've also noticed issues with service installations being flagged as
> "warnings" (their status is orange), but there's no corresponding error in
> the output from the target host. There is stdout output, but it's something
> that seems fine such as "Writing file [...] because contents don't match",
> and then it ends there. Is there a good set of heuristics to figure out
> what caused the status of the install to be orange?
>
> Thanks,
> Bill
>