installcdk is ridiculous!

I find it quite preposterous that the installer wants to run /bin/date on every single machine in my cluster! I have over 2700 machines!

Now because we have to install new software on ‘buildhosts’ because our shared software directory is shared to the cluster READ-ONLY, the buildhost has never before now connected to any of the compute nodes, meaning I have to ‘accept’ new entries to my known_hosts file.

This is quite frankly stupid and should be changed. At least give the user another option, or a way to provide the necessary data in another fashion.

I don’t think it is a good idea to ‘assume’ that the install script is run on the ‘master-node’, as situations like mine probably exist on larger clusters. Assuming everyone is on a small baby cluster is a good way to keep the larger cluster admins from wanting to install PGI. I know I sure don’t want to deal with this again …

Lovely! Now I am getting tons of errors that has resulted in me using ‘control-c’ to escape from the entire installation.

I guess it is time to contact support …


You are right, we do not see clusters with 2700 nodes.

If you can create a commonly mounted directory for all the nodes to
access the compilers, install on the master node as ‘non-root’ and
it should NOT try to test your configuration by running the ‘/bin/date’.

The resulting installation should work on all the nodes, since they
should be running the same linux (and therefore gcc) version.