Together with some people here at UPPMAX, Stockholm, we're trying to
code proper support for SLURM via DRMAA on Galaxy.
Thanks to Martin, we've managed to make it work so far, not without
putting some intrusive burden into it (see pic attached).
The memory allocation is not something the average user really needs, so
we can take the radio buttons away and assume "reasonable defaults" (i.e
go with job manager's defaults). Some goes for the core granularity,
which is ok in most cases (sending tool jobs to individual CPU cores
instead of whole nodes).
The only "problems" are time allocation and project id's. The first
helps the job manager on better scheduling, the latter is how we
partition research projects (budgeting and core hours allowed per project).
The general config file allows us to set a fixed project:
default_cluster_job_runner = drmaa://-A a2010002 -p core
And even set per-tool job settings. But we would like each user to have
the ability to change those settings.
What is the least intrusive way to set per-user native (site-specific)
job manager settings ?
Which classes/scripts should we look at/tweak/extend ?
Thanks in advance !
Martin and Roman