[ARVADOS] updated: 1.1.4-223-gf39ddfd

[Git user]

Summary of changes:
 services/crunch-dispatch-slurm/squeue.go | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

       via  f39ddfd2ab8479c6c6bbef44fc479a6f20aa2527 (commit)
       via  016ac3cd9e885192e069ba314593de64bb9e94b1 (commit)
      from  524c20020594ba67a2a822eccb632f8a5f5dc3ce (commit)

Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.


commit f39ddfd2ab8479c6c6bbef44fc479a6f20aa2527
Merge: 524c200 016ac3c
Author: Tom Clegg <tclegg at veritasgenetics.com>
Date:   Mon May 7 10:40:17 2018 -0400

    Merge branch '13278-launch-failed'
    
    refs #13278
    refs #13399
    
    Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tclegg at veritasgenetics.com>


commit 016ac3cd9e885192e069ba314593de64bb9e94b1
Author: Tom Clegg <tclegg at veritasgenetics.com>
Date:   Tue Mar 27 13:01:38 2018 -0400

    13278: Release jobs that are pending because "launch failed".
    
    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 9514da8..742943f 100644
--- a/services/crunch-dispatch-slurm/squeue.go
+++ b/services/crunch-dispatch-slurm/squeue.go
@@ -157,7 +157,7 @@ func (sqc *SqueueChecker) check() {
 		replacing.nice = n
 		newq[uuid] = replacing
 
-		if state == "PENDING" && reason == "BadConstraints" && p == 0 && replacing.wantPriority > 0 {
+		if state == "PENDING" && ((reason == "BadConstraints" && p == 0) || reason == "launch failed requeued held") && replacing.wantPriority > 0 {
 			// When using SLURM 14.x or 15.x, our queued
 			// jobs land in this state when "scontrol
 			// reconfigure" invalidates their feature
@@ -171,6 +171,10 @@ func (sqc *SqueueChecker) check() {
 			// reappeared, so rather than second-guessing
 			// whether SLURM is ready, we just keep trying
 			// this until it works.
+			//
+			// "launch failed requeued held" seems to be
+			// another manifestation of this problem,
+			// resolved the same way.
 			log.Printf("releasing held job %q", uuid)
 			sqc.Slurm.Release(uuid)
 		} else if p < 1<<20 && replacing.wantPriority > 0 {

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


hooks/post-receive
--