Yes. We have QoSs setup for each PI account, limited to 80K cpu hours per month. But I have played with --time, and I never get this error with fewer than 7 nodes no matter how much time I specify, and I never fail to get it even when time is very, very low (seconds).
Also, FWIW, our slurmdb setup isn't something that has changed. ------ Original message------ From: Uwe Sauter Date: Mon, Dec 19, 2016 09:57 To: slurm-dev; Cc: Subject:[slurm-dev] Re: Strange hostlist/malloc error Do you have limits (per partition / group), QoS (with limits per user), etc configured? Am 19.12.2016 um 15:52 schrieb Wiegand, Paul: > Greetings, > > > We were running slurm 16.05.0 and just upgraded to 16.05.7 during our Fall > maintenance cycle along with other changes. > Now we are having a very strange problem: > > > * When a regular user requests 6 or fewer nodes, one is able to do so without > issue; > > > * When a regular user requests 7 or more nodes, one gets the following error: > > > hostlist.c:1007: hostrange shift: malloc failed > Aborted (core dumped) > > > * It doesn't matter which nodes are being requested ... I've verified that I > *can* allocate across any node ... just not > more than 6; > > > * The root user can request any number of nodes; > > > > I should say that I do *not* believe this is due to the upgrade per se. > There were other changes made during downtime, > including to how groups are named and handled. I expect it to be a > permissions problem somewhere, but the logs are not > helpful. Indeed, I rolled back to 16.05.0 and still get this error. > > > Any help where to look would be appreciated. > > > Thanks, > > Paul. > >
