13078: Add test case for "release job held because BadConstraints".

[arvados.git] / services / crunch-dispatch-slurm / squeue.go

diff --git a/services/crunch-dispatch-slurm/squeue.go b/services/crunch-dispatch-slurm/squeue.go
index 8862d166a4e75e1908b240302e209dc44a132c2f..365b4e85706ccd46f8a77ad30af2700cfaf1ac60 100644 (file)
--- a/services/crunch-dispatch-slurm/squeue.go
+++ b/services/crunch-dispatch-slurm/squeue.go
@@ -121,7 +121,7 @@ func (sqc *SqueueChecker) check() {
        sqc.L.Lock()
        defer sqc.L.Unlock()
 
-       cmd := sqc.Slurm.QueueCommand([]string{"--all", "--noheader", "--format=%j %y %Q"})
+       cmd := sqc.Slurm.QueueCommand([]string{"--all", "--noheader", "--format=%j %y %Q %T %r"})
        stdout, stderr := &bytes.Buffer{}, &bytes.Buffer{}
        cmd.Stdout, cmd.Stderr = stdout, stderr
        if err := cmd.Run(); err != nil {
@@ -135,9 +135,9 @@ func (sqc *SqueueChecker) check() {
                if line == "" {
                        continue
                }
-               var uuid string
+               var uuid, state, reason string
                var n, p int64
-               if _, err := fmt.Sscan(line, &uuid, &n, &p); err != nil {
+               if _, err := fmt.Sscan(line, &uuid, &n, &p, &state, &reason); err != nil {
                        log.Printf("warning: ignoring unparsed line in squeue output: %q", line)
                        continue
                }
@@ -148,6 +148,23 @@ func (sqc *SqueueChecker) check() {
                replacing.priority = p
                replacing.nice = n
                newq[uuid] = replacing
+
+               if state == "PENDING" && reason == "BadConstraints" && p == 0 && replacing.wantPriority > 0 {
+                       // When using SLURM 14.x or 15.x, our queued
+                       // jobs land in this state when "scontrol
+                       // reconfigure" invalidates their feature
+                       // constraints by clearing all node features.
+                       // They stay in this state even after the
+                       // features reappear, until we run "scontrol
+                       // release {jobid}".
+                       //
+                       // "scontrol release" is silent and successful
+                       // regardless of whether the features have
+                       // reappeared, so rather than second-guessing
+                       // whether SLURM is ready, we just keep trying
+                       // this until it works.
+                       sqc.Slurm.Release(uuid)
+               }
        }
        sqc.queue = newq
        sqc.Broadcast()