Re: [HTCondor-users] slot cool down time

[Stefano Dal Pra <stefano.dalpra@xxxxxxxxxxxx>]

On 23/08/21 16:39, Beyer, Christoph wrote:
> Hi,
>
> I would like a certain type of slots to cool down after a finished or failed job. Reason is these are interactive jobs and I want the 2nd try of the user (complete new job) not to run on the same node that did not succeed before.
>
> In my tiny brain I thought that something like:
>
> START = $(START) && ((time() - EnteredCurrentState) > 600)
>
> Would do the trick just fine and effectly cause a 10 minute waiting time but apparently it does not - any suggestions ?
>
> Best
> christoph
Hello Christoph,
my understanding is that EnteredCurrentState
is the job classad of a "not yet started job", so that clause would 
match jobs pending for more than 10 mins.

I had a similar problem (did not want a singlecore start immediately 
soon after a multicore left).
I defined a custom machine classad "MC_GRACE" whose numeric value is set 
by a startd cronjob running at the node;

In this case, the idea is that the cronjob (running every min) keeps a 
numeric value (say x) set to 1.0 while a given condition is False 
(example: a job of Owner "foo" is running).
When it is not, it scales the x value; for example:ÂÂ x <-- 0.75 * x
then you set the machine classad as, say: OKJOB = (x < 0.1).

In this case the OKJOB would become True 9 minutes after the condition 
who keeps x at 1.0 has gone.
Hope this idea could help, eventhought my explanation could be a bit 
confused.

Stefano