[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Unable to get _HOOK_PREPARE_JOB to add something to job ClassAd


I think the real thing Carsten is thinking about is producing a version of RemoteWallClockTime that is weighted by the performance of each machine it touches. The main idea being to guide future purchases based upon what we think the jobs would have done on some standard machine.

I think the appropriate way is to rely on BENCHMARKS_JOBLIST in some capacity. Regardless, I will add Carsten's concern to the "what we need/want/etc" document I intend to circulate later this week for LIGO consumption and then pass on to the HTCondor team.

Tom Downes
Senior Scientist and Data CenterÂManager
Center for Gravitation, Cosmology and Astrophysics
University of Wisconsin-Milwaukee

On Mon, Sep 4, 2017 at 8:15 AM, Carsten Aulbert <carsten.aulbert@xxxxxxxxxx> wrote:
Hi Greg

(sorry for the delay, a lot of things kept interfering).

On 08/02/17 15:12, Greg Thain wrote:
> The job ad changed by the hook only changes the copy on the execute
> node, and doesn't make it back to the submit side.
> It is probably easier to advertise this attribute in a machine classad
> with startd cron or STARTD_ATTRS, and then use
> on the submit side to copy that attribute into the job ad.

Works like a charm with one caveat - many time/timing related attributes
are undefined at the time the machine attributes are copied into the job
ad, e.g. JobStart is undefined but MyCurrentTime is (good).

Is there an "easy" way to record the start (MyCurrentTime may already be
good enough) as well as the stop times of a job in the job ad?

My ultimate goal would be to track the slot usage across a few exec
hosts, i.e. in principle all I would need are a few machine attributes
along with the start and stop times (or the run time).

Do you know a way to accomplish this?



HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@cs.wisc.edu with a
subject: Unsubscribe
You can also unsubscribe by visiting

The archives can be found at: