PollInterval: time.Duration(theConfig.PollPeriod),
DoneProcessing: make(chan struct{})}
- if _, err := daemon.SdNotify("READY=1"); err != nil {
+ if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
log.Printf("Error notifying init daemon: %v", err)
}
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("--cpus-per-task=%d", container.RuntimeConstraints.VCPUs))
- if container.RuntimeConstraints.Partition != nil {
- sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.RuntimeConstraints.Partition, ",")))
+ if container.SchedulingParameters.Partitions != nil {
+ sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.SchedulingParameters.Partitions, ",")))
}
return exec.Command("sbatch", sbatchArgs...)
b, _ := ioutil.ReadAll(stdoutReader)
stdoutReader.Close()
stdoutChan <- b
+ close(stdoutChan)
}()
stderrChan := make(chan []byte)
b, _ := ioutil.ReadAll(stderrReader)
stderrReader.Close()
stderrChan <- b
+ close(stderrChan)
}()
// Send a tiny script on stdin to execute the crunch-run command
io.WriteString(stdinWriter, execScript(append(crunchRunCommand, container.UUID)))
stdinWriter.Close()
- err = cmd.Wait()
-
stdoutMsg := <-stdoutChan
stderrmsg := <-stderrChan
- close(stdoutChan)
- close(stderrChan)
+ err = cmd.Wait()
if err != nil {
submitErr = fmt.Errorf("Container submission failed: %v: %v (stderr: %q)", cmd.Args, err, stderrmsg)
// Mutex between squeue sync and running sbatch or scancel.
squeueUpdater.SlurmLock.Lock()
- err := scancelCmd(container).Run()
+ cmd := scancelCmd(container)
+ msg, err := cmd.CombinedOutput()
squeueUpdater.SlurmLock.Unlock()
if err != nil {
- log.Printf("Error stopping container %s with scancel: %v",
- container.UUID, err)
+ log.Printf("Error stopping container %s with %v %v: %v %v",
+ container.UUID, cmd.Path, cmd.Args, err, string(msg))
if squeueUpdater.CheckSqueue(container.UUID) {
log.Printf("Container %s is still in squeue after scancel.",
container.UUID)