[jira] [Created] (REEF-488) Refactor LocalRuntime to support with appattempts

["Andrew Chung (JIRA)" <ji...@apache.org>]

Andrew Chung created REEF-488:
---------------------------------

             Summary: Refactor LocalRuntime to support with appattempts
                 Key: REEF-488
                 URL: https://issues.apache.org/jira/browse/REEF-488
             Project: REEF
          Issue Type: Improvement
          Components: REEF Runtime Local
         Environment: REEF Local Runtime
            Reporter: Andrew Chung


Currently, the way local runtime is structured does not support driver restarts. As of now, LocalRuntime starts the driver process first, which instantiates the local ResourceManager, which in turn starts Evaluator processes as child processes of the driver process. From here, there are two scenarios that can happen: if the driver is force-terminated (e.g. via kill command), child processes may be orphaned, but the ResourceManager dies with it, so no context is kept. If the driver terminates "regularly" by throwing an exception, child processes will be killed by local runtime cleanup. Either way, context is lost.

LocalRuntime needs to be refactored to support the restart scenario.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)