[torqueusers] qalter -s "comment"?
David.Singleton at anu.edu.au
Wed Oct 15 14:01:35 MDT 2008
I guess you are assuming that the scheduler wont change the
job comment once the job is running. In general, a manually
set job comment could be overwritten by the scheduler soon
after (unless your scheduler can be told not to set job
comments). Is the comment the right place for this?
For a Quadrics RMS system, we modified OpenPBS to put the
"RMS resource id" (analogous to the ALPS reservation id) in the
PBS altid job attribute. It seems to me this is compatible with
how altid was used for IRIX arrays and UNICOS jobs. We modified
qalter (added a -i option) to allow the setting of altid manually
if necessary. Mostly we handled RMS directly from the scheduler
and MOM using the RMS API. I believe ALPS has a fairly extensive
API as well.
Troy Baer wrote:
> Can anybody think of a reason why qalter shouldn't be allowed to modify
> the comment field in a job object if run by a user with manager privs?
> I know that's historically been reserved for the scheduler, but what if
> you've got a scheduler that doesn't go through the scheduler C API?
> For instance, right now I help administer a Cray XT4 where the resource
> management environment has a quirky three-way interaction between
> TORQUE, Moab, and a Cray-specific allocator/placement service called
> ALPS via a set of glue scripts. ALPS has its own view of the machine
> that is more or less disjoint from that of TORQUE, although it has
> "reservation ids" that ought to have a one-to-one correlation to running
> TORQUE jobids. Moab knows the mapping between a TORQUE jobid and an
> ALPS resid thanks to one of the glue scripts, but I'd like to modify
> said script to put that information into something that will show up in
> the output of qstat -f (and maybe even the accounting logs) as well.
> Let me know if I'm totally off in left field here.
More information about the torqueusers