I’m looking for guidance on managing software packages across cluster members. Specifically, managing software on dynamic compute clusters (e.g., a cluster in AWS with AutoScaling in Alces Launch. You could gain similar functionality by using User Data on instance launch or pdsh with a static cluster, but are there best practices with managing software across dynamic cluster with Alces Launch?
There are a couple of options for managing cluster-wide software package installation:
- Use of the
alces gridwaretool to install your applications and libraries – these are installed in the
/opt/gridwaretree and NFS exported to the cluster.
- Manually build your apps into the
/opt/appstree – similarly, this directory is NFS exported to the cluster.
There shouldn’t be any need to install apps on the compute/slave nodes as they are able to access the software over the network.
Having said that, it’s possible you might need distribution package (RPM) support for manually installed apps. If you need to do this, we’d recommend managing the process through customization profiles – let us know if you need some pointers on configuring profiles that will handle distro package installations.