Description
Hi,
our cluster has different queues (normal, short, long, test), with different default allocated ressources.
With the current templating system, one needs to define resource default values for each variable that appears hard-coded in the template, or provide values in the submitJobs
call.
It would be nice if the specification of non-default resources (w.r.t. to the submission queue) could be generated dynamically, based on the optional values passed in submitJobs
.
So the template would have a simple resources line with a single meta-tag:
#<%= resources %>
which would be formated and replaced by the suitable cluster-specific arguments, e.g. for Torque:
sprintf("#PBS -l %s=%s\n", names(resources), unlist(resources))
In fact this could even be generalised to other parameters
sprintf("#PBS %s %s=%s\n", sapply(resources, '[[', 'flag'), names(resources), sapply(resources, '[[', 'value'))
Each cluster function specification should then provide the mapping of resources to universal resource names that are found across clusters (nodes, memory, job.name, etc...).
Cluster-specific templates can still be provided, but they can also leverage the use of the resources
meta-tag.
Makes sense? Is this actually alreayd impelemented?