EC2 node names

[Paolo Di Tommaso]

Dear all, 

Is there a way to provide a naming strategy for the nodes in a cluster? 

I mean, let's suppose that I want to create a cluster with 10 nodes named like that: 

master, node01, node02, .. node09

How to do that ? 

Cheers,

Paolo

[Adrian Cole]

Hi, paulo.

there's no nice way to do this, right now.  The logic is somewhat tangled in CreateNodesWithGroupEncodedIntoNameThenAddToSet.  Can you add an issue for this feature? 

http://code.google.com/p/jclouds/issues

Cheers,
-A

--
You received this message because you are subscribed to the Google Groups "jclouds" group.
To view this discussion on the web visit https://groups.google.com/d/msg/jclouds/-/ZrpnVfj0HtUJ.
To post to this group, send email to jcl...@googlegroups.com.
To unsubscribe from this group, send email to jclouds+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/jclouds?hl=en.

[Paolo Di Tommaso]

Hi Adrian, 

I've just opened the issue. http://code.google.com/p/jclouds/issues/detail?id=846

Anyway my idea was to have a way to distinguish the nodes role in a group so that I can apply the configuration accordingly. 

For example consider the following use case:  Create and configure a cluster in which one node is the master the others are clients.  The configuration for master node has to be different form the client nodes. 

I was thinking to use the name as discriminator with the method runScriptOnNodesMatching. 

Does it make sense?   Would be better to use to separate groups to do that ? 

In general, what is the best approach to handle a use case like this with JClouds.

Thank you a lot. 

Paolo 

[Adrian Cole]

Hi, Paolo.

Thanks for logging the issue and thinking this through.  In jclouds, the idiomatic way would be to make a group for each identically configured set of nodes.  So, you'd have a group called master with one node in it and a group called client with a variable count.

Make sense?
-A

[Paolo Di Tommaso]

Quite clear. 

But if so how to distinguish different cluster instances? Let's suppose you have a NFS based cluster with 10 nodes and a Hadoop based cluster with 20 nodes. 

I was supposing to use the group name to do that. 

Thanks, Paolo

[Adrian Cole]

in this case, the nodes have different roles, so they'd be easiest as different groups.

If you want the same group to have different roles inside, you'd be best using a config management solution to distinguish the difference, and utiliize its database as a part of the Predicate<NodeMetadata> used in jclouds commands.

ex. if your config management has a search, you'd say to jclouds:

runScriptOnNodesMatching(new Predicate<NodeMetadata>(){
public boolean apply(NodeMetadata in){
return configQuery.idsWithRole("NFS").contains(in.getId());
}

you can look at jclouds-chef for examples like this, or pallet, whirr, etc.

I hope this helps.
-A

[Paolo Di Tommaso]

OK. Thanks. 

Actually I'm thinking that using the UserMetadata could be a valid alternative to have different roles inside a group. 

It is enough start the master node setting role=master as user metadata, and then create the remaining nodes with role=clients . 

Doing so this information would be showed in the AWS web console that could useful to administer the cluster. 

Thanks for your tips. 

Cheers,

Paolo

[Adrian Cole]

yep. that would work, too. good idea.

-a

On Thu, Feb 23, 2012 at 3:59 PM, Paolo Di Tommaso