Jason Lowe commented on YARN-2902:

The key with the new proposal is that the LocalizerRunner thread is the one 
issuing the deletes and only after the localizer process exits.

bq. 2. Container is killed. Associated resources are stuck in downloading state 
and a deletion task is launched for them.
That deletion task would not be launched because the localizer has not exited.  
The LocalizerRunner will still be waiting on the localizer process to exit.

bq. Localizer doesnt exit immediately as of now when container is killed, even 
though we interrupt the thread.
Yes, and that's fine.  We won't issue the deletion requests until the localizer 
process eventually exits.

The key is this code in LocalizerRunner:
    public void run() {
      Path nmPrivateCTokensPath = null;
      Throwable exception = null;
      try {
[...localizer pre-startup code removed for brevity...]
        if (dirsHandler.areDisksHealthy()) {
          exec.startLocalizer(new LocalizerStartContext.Builder()
        } else {
          throw new IOException("All disks failed. "
              + dirsHandler.getDisksHealthReport(false));
      // TODO handle ExitCodeException separately?
      } catch (FSError fe) {
        exception = fe;
      } catch (Exception e) {
        exception = e;
      } finally {
        if (exception != null) {
          LOG.info("Localizer failed", exception);
          // On error, report failure to Container and signal ABORT
          // Notify resource of failed localization
          ContainerId cId = context.getContainerId();
          dispatcher.getEventHandler().handle(new ContainerResourceFailedEvent(
              cId, null, exception.getMessage()));
        for (LocalizerResourceRequestEvent event : scheduled.values()) {
        delService.delete(null, nmPrivateCTokensPath, new Path[] {});

startLocalizer won't return until the localizer process exits, so when it 
iterates the {{scheduled}} map in the finally block to unlock the resources we 
can issue deletions for the local resource paths at the same time.

> Killing a container that is localizing can orphan resources in the 
> ------------------------------------------------------------------------------------
>                 Key: YARN-2902
>                 URL: https://issues.apache.org/jira/browse/YARN-2902
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: nodemanager
>    Affects Versions: 2.5.0
>            Reporter: Jason Lowe
>            Assignee: Varun Saxena
>         Attachments: YARN-2902.002.patch, YARN-2902.03.patch, 
> YARN-2902.04.patch, YARN-2902.05.patch, YARN-2902.06.patch, 
> YARN-2902.07.patch, YARN-2902.patch
> If a container is in the process of localizing when it is stopped/killed then 
> resources are left in the DOWNLOADING state.  If no other container comes 
> along and requests these resources they linger around with no reference 
> counts but aren't cleaned up during normal cache cleanup scans since it will 
> never delete resources in the DOWNLOADING state even if their reference count 
> is zero.

This message was sent by Atlassian JIRA

Reply via email to