You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tephra.apache.org by "Andreas Neumann (JIRA)" <ji...@apache.org> on 2017/08/31 22:40:01 UTC

[jira] [Commented] (TEPHRA-242) Transaction Service sometimes does not shutdown

    [ https://issues.apache.org/jira/browse/TEPHRA-242?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16149721#comment-16149721 ] 

Andreas Neumann commented on TEPHRA-242:
----------------------------------------

It is hard to say what happened in the scenario that we observed: The pruning service was still running after the transaction server had shutdown. The only way to explain this is that in this code:
{code}
        if (server != null && server.isRunning()) {
          server.stopAndWait();
        }
        if (pruningService != null && pruningService.isRunning()) {
          // Wait for pruning service to stop after un-registering from discovery
          stopFuture = pruningService.stop();
        }
{code}
the stopping of the server threw an exception (we were facing OutOfMemory issues) or an InterruptedException. In that case, the exception goes uncaught and pruningService.stop() is never called. It seems safe to stop the prune service first (it needs the tx server anyway). 



> Transaction Service sometimes does not shutdown
> -----------------------------------------------
>
>                 Key: TEPHRA-242
>                 URL: https://issues.apache.org/jira/browse/TEPHRA-242
>             Project: Tephra
>          Issue Type: Bug
>          Components: manager
>    Affects Versions: 0.12.0-incubating
>            Reporter: Andreas Neumann
>            Assignee: Andreas Neumann
>             Fix For: 0.13.0-incubating
>
>
> If shutting down encounters an exception (for example, OutOfMemory), it can happen that the process does not exit. We have observed this after an NPE from DFS (closing the transaction log), the process stayed alive and the\ prune thread (executor) was not shut down. It was not clear why the process stayed alive, but it would certainly help to handle exceptions better in the shutdown sequence. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)