X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/83c8f1685d812c31d8bd568f3c2ac1edcd120aed..1885991dd79a75034f7650cccef16ebe3fd71959:/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go?ds=sidebyside

diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
index 7cb14fab15..cca8b3f150 100644
--- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
+++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm.go
@@ -123,44 +123,48 @@ func doMain() error {
 		log.Printf("Error notifying init daemon: %v", err)
 	}
 
-	containerTrackerTicker := trackContainers(dispatcher)
-	defer containerTrackerTicker.Stop()
+	go checkSqueueForOrphans(dispatcher, sqCheck)
 
 	return dispatcher.Run(context.Background())
 }
 
-var containerUuidPattern = regexp.MustCompile(`[a-z0-9]{5}-dz642-[a-z0-9]{15}$`)
-
-// Start a goroutine to check squeue report periodically, and
-// invoke TrackContainer for all the containers in the report.
-func trackContainers(dispatcher *dispatch.Dispatcher) *time.Ticker {
-	ticker := time.NewTicker(sqCheck.Period)
-	go func() {
-		for {
-			select {
-			case <-ticker.C:
-				for uuid := range sqCheck.AllUuids() {
-					match := containerUuidPattern.MatchString(uuid)
-					if match {
-						dispatcher.TrackContainer(uuid)
-					}
-				}
-			}
+var containerUuidPattern = regexp.MustCompile(`^[a-z0-9]{5}-dz642-[a-z0-9]{15}$`)
+
+// Check the next squeue report, and invoke TrackContainer for all the
+// containers in the report. This gives us a chance to cancel slurm
+// jobs started by a previous dispatch process that never released
+// their slurm allocations even though their container states are
+// Cancelled or Complete. See https://dev.arvados.org/issues/10979
+func checkSqueueForOrphans(dispatcher *dispatch.Dispatcher, sqCheck *SqueueChecker) {
+	for _, uuid := range sqCheck.All() {
+		if !containerUuidPattern.MatchString(uuid) {
+			continue
+		}
+		err := dispatcher.TrackContainer(uuid)
+		if err != nil {
+			log.Printf("checkSqueueForOrphans: TrackContainer(%s): %s", uuid, err)
 		}
-	}()
-	return ticker
+	}
 }
 
 // sbatchCmd
 func sbatchFunc(container arvados.Container) *exec.Cmd {
-	memPerCPU := math.Ceil(float64(container.RuntimeConstraints.RAM) / (float64(container.RuntimeConstraints.VCPUs) * 1048576))
+	mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+container.RuntimeConstraints.KeepCacheRAM) / float64(1048576)))
+
+	var disk int64
+	for _, m := range container.Mounts {
+		if m.Kind == "tmp" {
+			disk += m.Capacity
+		}
+	}
+	disk = int64(math.Ceil(float64(disk) / float64(1048576)))
 
 	var sbatchArgs []string
-	sbatchArgs = append(sbatchArgs, "--share")
 	sbatchArgs = append(sbatchArgs, theConfig.SbatchArguments...)
 	sbatchArgs = append(sbatchArgs, fmt.Sprintf("--job-name=%s", container.UUID))
-	sbatchArgs = append(sbatchArgs, fmt.Sprintf("--mem-per-cpu=%d", int(memPerCPU)))
+	sbatchArgs = append(sbatchArgs, fmt.Sprintf("--mem=%d", mem))
 	sbatchArgs = append(sbatchArgs, fmt.Sprintf("--cpus-per-task=%d", container.RuntimeConstraints.VCPUs))
+	sbatchArgs = append(sbatchArgs, fmt.Sprintf("--tmp=%d", disk))
 	if len(container.SchedulingParameters.Partitions) > 0 {
 		sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.SchedulingParameters.Partitions, ",")))
 	}