Answers inline

On Tue, Aug 16, 2011 at 8:17 PM, tog <[email protected]> wrote:
> I looked into the flume service demo and one patch for ganglia. I was
> wondering if some docs describing the service installation lifecycle
> were available. I would be interested to now when [before,
> after]Bootstrap, Configure are called during the VM lifecycle
> procedure.

A cluster started by Whirr goes through 3 major steps: bootstrap,
configure, destroy.

In the bootstrap phase the objective is to provision the server
resources and install the services without configuring or starting
them. In the configure phase the objective is to configure and start
the services. Destroy just stops all the VMs.

The before/after events are triggered as you would expected before and
after each phase and they are used to construct a list of jclouds
statements and to change firewall settings.

I think Tom's presentation explains the process in detail.

> My interest is to:
>   - know where to define SecurityGroup rules (in ec2 dialect) - in
> flume it seems to be located in beforeConfigure while at that stage I
> would have assumed tthe VM to be already started

Yes, the VM is already running but keep in mind that by default the
SecurityGroup denies outside access and allows any kind of
communication inside. If you wish you can change the firewall settings
in afterConfigure and it should work just fine.

Warning: there is no firewall support for cloudservers and by default
there are no connection restrictions.

>   - know if I can set-up fw rules involving external/local addresses

I don't really understand this question. You can specify the source IP
for a rule.

>  - know how to decide in which order ServiceMasterHandler /
> ServiceNodeHandler shall be run (sometimes nodes need the master  ip
> or vice-versa)

This is not supported. We are currently running scripts concurrently
on all machines for each cluster lifecycle step. There is an open
issue for this. On the long term we want to be able to define a DAG
for the supported services.

https://issues.apache.org/jira/browse/WHIRR-221

I'm not sure how clear my answers are so feel free to ask more questions.

Cheers,

Andrei

>
> Thanks for your help
>
> Guillaume
>
> On Mon, Aug 15, 2011 at 8:28 AM, tog <[email protected]> wrote:
>>
>> Thanks, looks pretty easy - will start creating new services
>>
>> On Sun, Aug 14, 2011 at 11:53 PM, Andrei Savu <[email protected]> wrote:
>>>
>>> Take a look at the following presentation:
>>>
>>> http://www.oscon.com/oscon2011/public/schedule/detail/19214
>>>
>>> Explains what Whirr is and how to add a new service.
>>>
>>> -- Andrei Savu / andreisavu.ro
>>>
>>> On Sun, Aug 14, 2011 at 8:16 AM, tog <[email protected]> wrote:
>>> > i just came across whirr today - i was wondering if that could be used to
>>> > configure a compute cluster on ec2 - for example something having:
>>> >  - ntp, nfs, mpi, slurm configured
>>> >
>>> > I guess new roles have to be created for both the head node and slave 
>>> > nodes
>>> > of the cluster.
>>> >
>>> > Is there some place where I could learn that ?
>>> >
>>> > Guillaume
>>> >
>>> > --
>>> > PGP KeyID: 2048R/EA31CFC9  subkeys.pgp.net
>>> >
>>
>>
>>
>> --
>> PGP KeyID: 2048R/EA31CFC9  subkeys.pgp.net
>
>
>
> --
> PGP KeyID: 2048R/EA31CFC9  subkeys.pgp.net
>

Reply via email to