Hi,

It depends on how your scheduler works.  If it is an external polling
scheduler, one that periodically gets job and node information from SLURM,
then decides when and where to run jobs, then you might want to use the
wiki or wiki2 interfaces.  These are described here:

http://slurm.schedmd.com/schedplugins.html

I've integrated my own scheduler with SLURM through the wiki interface.
This is a while back, so I'm not sure if the specifics have changed.
Back then, this involved activating wiki by modifying
slurm.conf and wiki.conf.

slurm.conf:
...
SchedulerType=sched/wiki
SchedulerPort=7321
...

wiki.conf:
...
AuthKey=<some auth key>
...

This causes submitted jobs to go into JobHeld, with Priority=0.
Now, SLURM commands or API calls can be used to start jobs.

Something like:
scontrol update JobId=<job id> ReqNodeList=<nodelist> Priority=1

would then cause the job to start on the specified nodes.

I have a slideshow with some of this info:
http://users.sdsc.edu/~kenneth/ipn.2010/workshop/slideshow.html

Not sure how you would go about it, if your scheduler is meant to
replace pieces of SLURM.

You might also be interested in the PySlurm package, which provides
a Python interface to SLURM:
http://www.gingergeeks.co.uk/pyslurm/

Kenneth



On Fri, Apr 01, 2016 at 04:44:44AM -0700, Naajil Aamir wrote:
> Hi there
> I want to know how can i embed my own job scheduler into slurm which i
> coded myself. I have an fyp in which i have to code a scheduler on slurm.
> Ive changed the existing builtin code to see what happens where and I would
> like for that code to run instead of the builtin code, Secondly what
> benchmarks are suitable to check slurm scheduler performance.

Reply via email to