Hello. One of our users has submitted a job which requires 2 nodes. While there was actually about four nodes free, that job was in PENDING state sith PRIORITY reason.
We have backfill sheduler enabled and basic priority plugin in use. SLURM version is 2.5.7. I thought backfill must deal with things like promoting a job with higher JobID that requires 2 nodes to launch it before a job with lower JobID which requires a number of nodes that is not available at the moment, am I wrong? 1. How is such SLURM behavoir explained? 2. Is there a way to instantly promote a job so that it would start immediately? And another bunch of questions, unrelated to the current issue: 1. Can I specify a partition so that it consumes half of cores on a node, not the whole node? 2. If so, how will a job submitted with --esclusive switch behave? Will it consume the whole node nevertheless, or will it allow other jobs on another partition's "pool" on that node? 3. Shall I do something with users' strong "--exclusive" addiction? Thanks in advance! Vsevolod Nikonorov. -- Всеволод Никоноров, ОИТТиС, НИКИЭТ <[email protected]>
