Thanks for the input. However, I doubt that a member variable approach is feasible, because when the Storm topology is translated into a Flink program (in `FlinkBuilder.createTopology()`) the Storm configuration is not available yet. And adding the configuration later to each operator would be cumbersome.
If there are no better ideas, I guess the current usage of JobConfiguration is the best way to handle it (because extending TaskConfiguration seems to be no option) -Matthias On 09/06/2015 10:51 PM, Aljoscha Krettek wrote: > Hi, > I think the possibility to use a Configuration object is a legacy from the > past where the API was a bit closer to how Hadoop works. In my opinion this > is not necessary anymore since User Code objects can just contain > configuration settings in fields. > > The feature for the Storm API could probably be implemented by just storing > a Configuration object in the user code function. > > Regards, > Aljoscha > > On Sun, 6 Sep 2015 at 18:29 Matthias J. Sax <mj...@apache.org> wrote: > >> Hi, >> >> I observed, that DataSet API offers a nice way to configure >> UDF-Operators by providing the method ".withParameters()". However, >> Streaming API does not offer such a method. >> >> For a current PR (https://github.com/apache/flink/pull/1046) this >> feature would be very helpful. >> >> As a workaround, PR #1046 can also be finished using JobConfiguration. >> However, this seems to be somewhat unnatural. Furthermore, I think that >> this feature would be nice to have in general. What do you think about it? >> >> If we introduce this feature, we can either open a new JIRA of just >> include it into the current PR #1046. What would be the better way? >> >> >> -Matthias >> >> >
signature.asc
Description: OpenPGP digital signature
