+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
package main
// Dispatcher service for Crunch that submits containers to the slurm queue.
}
}
+func niceness(priority int) int {
+ if priority > 1000 {
+ priority = 1000
+ }
+ if priority < 0 {
+ priority = 0
+ }
+ return (1000 - priority) * 1000
+}
+
// sbatchCmd
func sbatchFunc(container arvados.Container) *exec.Cmd {
mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+container.RuntimeConstraints.KeepCacheRAM) / float64(1048576)))
var disk int64
for _, m := range container.Mounts {
- disk += m.Capacity
+ if m.Kind == "tmp" {
+ disk += m.Capacity
+ }
}
disk = int64(math.Ceil(float64(disk) / float64(1048576)))
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))
+ sbatchArgs = append(sbatchArgs, fmt.Sprintf("--nice=%d", niceness(container.Priority)))
if len(container.SchedulingParameters.Partitions) > 0 {
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.SchedulingParameters.Partitions, ",")))
}
return exec.Command("scancel", "--name="+container.UUID)
}
+// scontrolCmd
+func scontrolFunc(container arvados.Container) *exec.Cmd {
+ return exec.Command("scontrol", "update", "JobName="+container.UUID, fmt.Sprintf("--nice=%d", niceness(container.Priority)))
+}
+
// Wrap these so that they can be overridden by tests
var sbatchCmd = sbatchFunc
var scancelCmd = scancelFunc
+var scontrolCmd = scontrolFunc
// Submit job to slurm using sbatch.
func submit(dispatcher *dispatch.Dispatcher, container arvados.Container, crunchRunCommand []string) error {
if ctr.State == dispatch.Locked && !sqCheck.HasUUID(ctr.UUID) {
log.Printf("Submitting container %s to slurm", ctr.UUID)
if err := submit(disp, ctr, theConfig.CrunchRunCommand); err != nil {
- log.Printf("Error submitting container %s to slurm: %s", ctr.UUID, err)
+ text := fmt.Sprintf("Error submitting container %s to slurm: %s", ctr.UUID, err)
+ log.Print(text)
+
+ lr := arvadosclient.Dict{"log": arvadosclient.Dict{
+ "object_uuid": ctr.UUID,
+ "event_type": "dispatch",
+ "properties": map[string]string{"text": text}}}
+ disp.Arv.Create("logs", lr, nil)
+
disp.Unlock(ctr.UUID)
return
}
}
- log.Printf("Start monitoring container %s", ctr.UUID)
+ log.Printf("Start monitoring container %v in state %q", ctr.UUID, ctr.State)
defer log.Printf("Done monitoring container %s", ctr.UUID)
// If the container disappears from the slurm queue, there is
} else if updated.Priority == 0 {
log.Printf("Container %s has state %q, priority %d: cancel slurm job", ctr.UUID, updated.State, updated.Priority)
scancel(ctr)
+ } else if niceness(updated.Priority) != sqCheck.GetNiceness(ctr.UUID) && sqCheck.GetNiceness(ctr.UUID) != -1 {
+ // dynamically adjust priority
+ scontrolUpdate(updated)
}
}
}
}
}
+func scontrolUpdate(ctr arvados.Container) {
+ sqCheck.L.Lock()
+ cmd := scontrolCmd(ctr)
+ msg, err := cmd.CombinedOutput()
+ sqCheck.L.Unlock()
+
+ if err != nil {
+ log.Printf("%q %q: %s %q", cmd.Path, cmd.Args, err, msg)
+ time.Sleep(time.Second)
+ } else if sqCheck.HasUUID(ctr.UUID) {
+ log.Printf("Container %s priority is now %v, niceness is now %v",
+ ctr.UUID, ctr.Priority, sqCheck.GetNiceness(ctr.UUID))
+ }
+}
+
func readConfig(dst interface{}, path string) error {
err := config.LoadFile(dst, path)
if err != nil && os.IsNotExist(err) && path == defaultConfigPath {