Re: Local master node

From: Vasisht Reddy T. <no email>
Date: Thu, 17 Mar 2016 12:36:55 -0400

You could either run load balancer on the cluster to kill the cluster when the jobs are done or re-launch the small node when the cluster needs to be shutdown and use starcluster to shutdown the cluster. This node doesn’t need to be running unless you need to add additional nodes to the cluster while it’s running.

On Thursday, March 17, 2016 at 12:29 PM, Adrian Draghici wrote:

> Hi,
>
> Thanks for the answers!
>
> I'm not sure I understand how the intermediate node could be shut down though? Assume the following scenario where all nodes are in EC2:
> A is the intermediate node with the new AMI, receiving and executing commands like "starcluster start testcluster" etc.
> B is the master node spawned by the cluster
> C, D, E are slave nodes
>
>
> Surely A would need to be alive until the end since the cluster needs to be turned off from it?
>
>
> On Thu, Mar 17, 2016 at 3:08 PM MacMullan IV, Hugh <hughmac_at_wharton.upenn.edu (mailto:hughmac_at_wharton.upenn.edu)> wrote:
> > I agree: you can shut that node down while the cluster is running, too.
> >
> > -----Original Message-----
> > From: starcluster-bounces_at_mit.edu (mailto:starcluster-bounces_at_mit.edu) [mailto:starcluster-bounces_at_mit.edu] On Behalf Of Cedar McKay
> > Sent: Thursday, March 17, 2016 11:00 AM
> > To: Adrian Draghici <draghici.adrian.b_at_gmail.com (mailto:draghici.adrian.b_at_gmail.com)>
> > Cc: starcluster_at_mit.edu (mailto:starcluster_at_mit.edu)
> > Subject: Re: [StarCluster] Local master node
> >
> > A t2.nano costs about 2/3 of a penny per hour. In my opinion it's not worth doing any extra engineering effort to avoid that cost.
> >
> > Best,
> > Cedar
> >
> > > On Mar 17, 2016, at 7:45 AM, Adrian Draghici <draghici.adrian.b_at_gmail.com (mailto:draghici.adrian.b_at_gmail.com)> wrote:
> > >
> > > Hi all,
> > >
> > > I’m currently trying to find a way to start and manage a Starcluster fleet from machines where installing Starcluster itself is not an option.
> > >
> > > One way of doing this could be snapshotting the standard AMI with the Starcluster tool installed and then manually bring up an instance from this snapshot. The newly brought up instance will now be a coordinator node that can take commands to manage a cluster. However, I would need this node to also be the master in order to avoid having the overhead of one extra box.
> > >
> > > Is there any option for allowing the machine starting the cluster to also be the master (assuming it’s also in EC2)?
> > >
> > > I’ve seen slight variations of this question asked before here, but with no definitive answer. I would also really appreciate any other ideas on how to solve this!
> > >
> > > Thanks,
> > > Adrian
> > > _______________________________________________
> > > StarCluster mailing list
> > > StarCluster_at_mit.edu (mailto:StarCluster_at_mit.edu)
> > > http://mailman.mit.edu/mailman/listinfo/starcluster
> >
> > _______________________________________________
> > StarCluster mailing list
> > StarCluster_at_mit.edu (mailto:StarCluster_at_mit.edu)
> > http://mailman.mit.edu/mailman/listinfo/starcluster
> _______________________________________________
> StarCluster mailing list
> StarCluster_at_mit.edu (mailto:StarCluster_at_mit.edu)
> http://mailman.mit.edu/mailman/listinfo/starcluster
>
>
Received on Thu Mar 17 2016 - 12:36:58 EDT