-// If the container is marked as Locked, check if it is already in the slurm
-// queue. If not, submit it.
-//
-// If the container is marked as Running, check if it is in the slurm queue.
-// If not, mark it as Cancelled.
-func monitorSubmitOrCancel(dispatcher *dispatch.Dispatcher, container arvados.Container, monitorDone *bool) {
- submitted := false
- for !*monitorDone {
- if squeueUpdater.CheckSqueue(container.UUID) {
- // Found in the queue, so continue monitoring
- submitted = true
- } else if container.State == dispatch.Locked && !submitted {
- // Not in queue but in Locked state and we haven't
- // submitted it yet, so submit it.
-
- log.Printf("About to submit queued container %v", container.UUID)
-
- if err := submit(dispatcher, container, config.CrunchRunCommand); err != nil {
- log.Printf("Error submitting container %s to slurm: %v",
- container.UUID, err)
- // maybe sbatch is broken, put it back to queued
- dispatcher.UpdateState(container.UUID, dispatch.Queued)
- }
- submitted = true
- } else {
- // Not in queue and we are not going to submit it.
- // Refresh the container state. If it is
- // Complete/Cancelled, do nothing, if it is Locked then
- // release it back to the Queue, if it is Running then
- // clean up the record.
-
- var con arvados.Container
- err := dispatcher.Arv.Get("containers", container.UUID, nil, &con)
- if err != nil {
- log.Printf("Error getting final container state: %v", err)
+// Submit a container to the slurm queue (or resume monitoring if it's
+// already in the queue). Cancel the slurm job if the container's
+// priority changes to zero or its state indicates it's no longer
+// running.
+func (disp *Dispatcher) runContainer(_ *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+ ctx, cancel := context.WithCancel(context.Background())
+ defer cancel()
+
+ if ctr.State == dispatch.Locked && !disp.sqCheck.HasUUID(ctr.UUID) {
+ log.Printf("Submitting container %s to slurm", ctr.UUID)
+ if err := disp.submit(ctr, disp.CrunchRunCommand); err != nil {
+ var text string
+ if err, ok := err.(dispatchcloud.ConstraintsNotSatisfiableError); ok {
+ var logBuf bytes.Buffer
+ fmt.Fprintf(&logBuf, "cannot run container %s: %s\n", ctr.UUID, err)
+ if len(err.AvailableTypes) == 0 {
+ fmt.Fprint(&logBuf, "No instance types are configured.\n")
+ } else {
+ fmt.Fprint(&logBuf, "Available instance types:\n")
+ for _, t := range err.AvailableTypes {
+ fmt.Fprintf(&logBuf,
+ "Type %q: %d VCPUs, %d RAM, %d Scratch, %f Price\n",
+ t.Name, t.VCPUs, t.RAM, t.Scratch, t.Price,
+ )
+ }
+ }
+ text = logBuf.String()
+ disp.UpdateState(ctr.UUID, dispatch.Cancelled)
+ } else {
+ text = fmt.Sprintf("Error submitting container %s to slurm: %s", ctr.UUID, err)