Re: [HTCondor-users] job hold, over memory limit

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

On 2/21/2019 9:17 AM, Michael Di Domenico wrote:
> for some of the jobs submitted to my pool i'm getting
> 
> "Job has gone over memory limit of 8192 megabytes. Peak usage: 8007 megabytes."
> 
> granted it's pretty close, but it doesn't seem like it should have
> held the job given that message.

Hi Michael,

Agreed!

Turns out if Linux thinks it is running out of memory, any job that is 
using more than 90% of its memory limit could be killed.

Thanks to you taking the time to report this, we are changing this in 
HTCondor v8.9.1 so this arbitrary 90% rule goes away, and jobs will need 
to be over there limit to be killed.

Details in this ticket:

  https://htcondor-wiki.cs.wisc.edu/index.cgi/tktview?tn=6925

Thanks for pointing out the above confusion!

regards,
Todd