[ https://issues.apache.org/jira/browse/FLINK-4505?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15464459#comment-15464459 ]
ASF GitHub Bot commented on FLINK-4505: --------------------------------------- Github user tillrohrmann commented on the issue: https://github.com/apache/flink/pull/2461 Hi @wangzhijiang999, maybe we should start simple without introducing a factory method, because there might actually be not many cases to distinguish. Maybe we could rename the `TaskManagerFactory` into `TaskManagerRunner` which has static methods to create the `TaskManagers` components and does the network selection. That way we keep the initialization and the actual `TaskManager` logic separated. For testing purposes I guess we don't need to setup any components because they are usually mocked or one is using testing components. Passing these components to the constructor of the `TaskManager` should not be a big deal. Does this make sense? > Implement TaskManagerFactory to bring up TaskManager for different modes > ------------------------------------------------------------------------ > > Key: FLINK-4505 > URL: https://issues.apache.org/jira/browse/FLINK-4505 > Project: Flink > Issue Type: Sub-task > Components: Cluster Management > Reporter: Zhijiang Wang > Assignee: Zhijiang Wang > Priority: Minor > > Implement {{TaskExecutorFactory}} that should be an abstract class with the > helper methods to bring up the {{TaskManager}}. The factory can be > implemented by some classes to start a {{TaskManager}} in different modes > (testing, standalone, yarn). -- This message was sent by Atlassian JIRA (v6.3.4#6332)