[
https://issues.apache.org/jira/browse/FLINK-2615?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14728925#comment-14728925
]
ASF GitHub Bot commented on FLINK-2615:
---------------------------------------
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/1088#issuecomment-137418968
The current behavior is actually desired. You can explicitly start the
local executor, and explicitly stop it. When you did not explicitly start it,
you get an auto-shut down executor. This is quite flexible, no?
The LocalExecutionEnvironment is a local one-time environment. There is
code to extend it to sessions, which is really what we want to "keep context
and data around".
For that, there is part of the work here:
https://github.com/apache/flink/pull/858
It follows a very principled approach, where the responsibility of deciding
clearing / disposal of cached context and data is bound to a session. Sessions
have timeouts and scope reapers, and are managed by the execution environments,
not the executors.
> Multiple restarts of Local Cluster for a single program
> -------------------------------------------------------
>
> Key: FLINK-2615
> URL: https://issues.apache.org/jira/browse/FLINK-2615
> Project: Flink
> Issue Type: Bug
> Reporter: Sachin Goel
> Assignee: Sachin Goel
>
> Currently, the LocalExecutor which executes program in a local environment
> starts a new cluster and tears it down on every call to execute. This can
> take up a lot of time since a program might trigger execution several times.
> Further, it is semantically different from the RemoteExecutor as in the
> latter, the actors always stay the same.
> A restart of actors in itself prevents any sharing of results across
> different executions. [A Persist operation.]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
