Hi Todd,
many thanks for your help. Now I start understanding, its the niceuser=true statement!
Unfortunatelly this is our default, beause our standard usage are serveral users with thousands of not too long running ( <2h ) jobs from each user in a cluster with around 200 cores. They all start with niceuser=true and this have the desired effect, that there are allways serveral cores free for time critical jobs (or impatient users), which can just start as normal user and mostly allways get enough jobs running.
Unfortunately I have not found how one can set the MaxJobRetirementTime for a job with niceuser=true. Have I overlooked something? Or does this mean that every user have to set it in the descrition file? Or can I set it with condor_qedit? Any suggestions?
best regards
Harald
On Friday 19 August 2016 00:45:32 Todd Tannenbaum wrote: > Hi Harold, > > Reading the below, I still say that something is already setting > MaxJobRetirementTime in your job classads to 0. Just because you are not > explicitly setting it in your job submit file doesn't mean it isn't > getting set. For instance, condor_submit itself will set > MaxJobRetirementTime=0 if your submit file has "nice_user=true" or > "universe=standard". This is why I suggested you look in your job > classads via "condor_q -l" or condor_history. > > But at any rate, to answer your question about how to have condor_submit > automatically set MaxJobRetirementTime to 9999999 in job ads by default, > append the following to your condor_config: > > MaxJobRetirementTime = 9999999 > SUBMIT_EXPRS = $(SUBMIT_EXPRS) MaxJobRetirementTime > > See URL : > > https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=HowToInsertCustomClassA > dIntoJobs > > regards, > Todd >
<snip> |