Best approach to modifying StarCluster code

From: Ramon Ramirez-Linan <no email>
Date: Thu, 11 Dec 2014 16:17:31 -0500

Hi,

We are currently using SciDB in the AWS cloud. I am using StarCluster to
configure the infrastructure on AWS. With SciDB the compute nodes are
required to have the data locally, so every compute node has a data EBS
attached.

The problem that I am having is that StarClusters terminates the node
including the EBS volume when they become no-responsive (after 15 minutes).
This force us to ingest the data again in the whole cluster.

I want to change this behaviour so instead of terminating non-responsive
nodes it will
-detach the EBS data volume,
-terminate the failed EC2,
-start another EC2,
-attach the data volume to the new node
-and add the node to the cluster.

What is the best approach to modifying StarCluster in this manner? Should I
just hack the base code or else?

Thanks in advance

Ramon
Received on Thu Dec 11 2014 - 16:17:33 EST