[ARVADOS] created: 1.1.4-748-g718e3b0c1

[Git user]

        at  718e3b0c18916a2a48d16cb5cb67d972ea5be6b8 (commit)


commit 718e3b0c18916a2a48d16cb5cb67d972ea5be6b8
Author: Tom Clegg <tclegg at veritasgenetics.com>
Date:   Sat Aug 4 11:05:45 2018 -0400

    13399: Don't try to bury jobs below very-low-priority jobs.
    
    Non-zero priorities below 20K are probably the result of renicing jobs
    that have priority=0 because they are in admin-hold state.
    
    Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tclegg at veritasgenetics.com>

diff --git a/services/crunch-dispatch-slurm/squeue.go b/services/crunch-dispatch-slurm/squeue.go
index fd4851eb0..24a056264 100644
--- a/services/crunch-dispatch-slurm/squeue.go
+++ b/services/crunch-dispatch-slurm/squeue.go
@@ -82,7 +82,7 @@ func (sqc *SqueueChecker) reniceAll() {
 			// (perhaps it's not an Arvados job)
 			continue
 		}
-		if j.priority == 0 {
+		if j.priority <= 2*slurm15NiceLimit {
 			// SLURM <= 15.x implements "hold" by setting
 			// priority to 0. If we include held jobs
 			// here, we'll end up trying to push other

-----------------------------------------------------------------------


hooks/post-receive
--