[
https://issues.apache.org/jira/browse/FLINK-999?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14053532#comment-14053532
]
Stephan Ewen commented on FLINK-999:
------------------------------------
+1 as well.
The configurability is kind of an issue in all places. Having the singleton
global configuration may work for the cluster setups, but it rather suboptimal
for the local embedded setups, where the configuration may change between
different runs.
I think it would be really nice of we did not have the singleton global
configuration (which became mutable in order to support the local embedded
setups), but a configuration instance that we pass to the JobManager /
TaskManager.
> Configurability of LocalExecutor
> --------------------------------
>
> Key: FLINK-999
> URL: https://issues.apache.org/jira/browse/FLINK-999
> Project: Flink
> Issue Type: Wish
> Reporter: Till Rohrmann
> Priority: Minor
>
> I noticed when running locally some examples with the Scala API that the
> TaskManagers ran out of network buffers when increasing the number of slots
> on one machine. Trying to adjust the value, I stumbled across a more
> naturally grown interface in LocalExecutor and NepheleMiniCluster. Some
> parameters are specified as attributes of NepheleMiniCluster, others are
> loaded from a configuration file and others use the default value specified
> in ConfigConstants. I think it would be good to give the user the ability to
> easily control the different parameters in a unified fashion by providing a
> configuration object. Furthermore, the NepheleMiniCluster is not consistent
> with the use of parameter values: For example, getJobClient sets the
> jobmanager rpc port with respect to the attribute jobManagerRpcPort whereas
> the start method can retrieve it from a specified configuration file.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
