[ https://issues.apache.org/jira/browse/FLINK-1670?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14502117#comment-14502117 ]
ASF GitHub Bot commented on FLINK-1670: --------------------------------------- Github user ggevay commented on the pull request: https://github.com/apache/flink/pull/581#issuecomment-94310035 Thank you for your comments. I am very sorry for not replying earlier, but I was extremely busy this week with other things. I will try to address the issues that you mentioned: > It supports only java.io.Serializable types. This is a bit inconsistent with the current type handling and serialization in Flink. Some types that work in all other parts do not work here. Sorry, I will fix this. > It does not work in a cluster. It sends "localhost" as the name to the worker who should send the data back. In any non-local setup, this cannot work. It doesn't send "localhost". It sends InetAddress.getLocalHost().getHostName(), which will be some real hostname of the machine. (Something like the "hostname" Linux command would give.) > It requires the worker to be able to connect to the client. This may be tricky, when the client and workers do not run both in the cluster. If I put this in "contrib", as you suggested later, is it then an acceptable assumption that the worker should be able to connect to the client, or should I find some other way to make the connection? > Selecting the proper interface that opens the port for data communication is actually quite tricky. I will modify the code to open the server in a way that it listens on all interfaces. (Actually, it already does that. bind(null) binds the socket to 0.0.0.0 (which means to listen on all interfaces), according to the JDK source code. But the Java API documentation doesn't say this explicitly, so I will use the ServerSocket constructor instead (with null as bindAddr), for which the documentation agrees with the source that it will listen on all interfaces.) I will also take a look at how the TaskManager does this. > What about reworking this as a library function and add it to flink-contrib? OK, I will make the modification. > Collect method for streaming > ---------------------------- > > Key: FLINK-1670 > URL: https://issues.apache.org/jira/browse/FLINK-1670 > Project: Flink > Issue Type: New Feature > Components: Streaming > Affects Versions: 0.9 > Reporter: Márton Balassi > Assignee: Gabor Gevay > Priority: Minor > > A convenience method for streaming back the results of a job to the client. > As the client itself is a bottleneck anyway an easy solution would be to > provide a socket sink with degree of parallelism 1, from which a client > utility can read. -- This message was sent by Atlassian JIRA (v6.3.4#6332)