Bjørn-Helge,
We haven't had time to look at this patch yet. One concern I do have
without looking at it is these limits are stored for the job instead of
the job step. I can image at least the last 2 being different for each
step. I would say the first two, tmp and constraint, are probably not
used differently on each step though.
Does anyone else have an opinion on this? I suppose if you only are
using this to analysis wait time then the job table is the correct spot
though.
Related to the topic of not having patches fall through the cracks it
might be useful to submit them to our bug tracking system
http://bugs.schedmd.com so you can follow along what is happening and
they don't fall through the cracks.
Danny
On 01/20/12 03:03, Bjørn-Helge Mevik wrote:
Here is a patch that adds the following fields to the
<cluster>_job_table in slurmdbd:
req_disk_pn: the value of --tmp
req_features: the value of --constraint
req_mem_per_cpu: the value of --mem-per-cpu
req_ntasks_pn: the value of --ntasks-per-node
These fields are very useful for doing analysis of wait times,
etc. (e.g., what effects the expected wait time before jobs start).
We've used this patch for a year with slurm 2.2.x, and have now tested
it with 2.3.2 and 2.4.0-0.pre2. The patch file has been created against
2.3.2, but applies cleanly on 2.4.0-0.pre2 as well.
We hope you can accept this for inclusion in 2.4.x.
