<div dir="ltr">Using Slurm 20.02 on CentIOS 7.7 with Bright Cluster. We changed the following options to enable MPS:<br>SelectType=select/cons_tres<br>GresTypes=gpu,mic,mps<br><br>I restarted slurmctld and ran scontrol reconfigure, however all jobs get the below error:<br>[2020-04-07T15:29:00.741] debug:  backfill: no jobs to backfill<br>[2020-04-07T15:29:03.051] Resending TERMINATE_JOB request JobId=3056 Nodelist=node[001-002]<br>[2020-04-07T15:29:03.051] Resending TERMINATE_JOB request JobId=3061 Nodelist=node003<br>[2020-04-07T15:29:03.051] debug:  sched: Running job scheduler<br>[2020-04-07T15:29:03.063] agent/is_node_resp: node:node003 RPC:REQUEST_TERMINATE_JOB : Header lengths are longer than data received<br>[2020-04-07T15:29:03.071] agent/is_node_resp: node:node002 RPC:REQUEST_TERMINATE_JOB : Header lengths are longer than data received<br>[2020-04-07T15:29:03.071] agent/is_node_resp: node:node001 RPC:REQUEST_TERMINATE_JOB : Header lengths are longer than data received<br><br>Do any other options need changing? What causes these header length errors? </div>