Github user aljoscha commented on the pull request:
https://github.com/apache/flink/pull/202#issuecomment-76973194
Yes, this sounds good? Another thing: it has probably already come up but I
just want to make sure, you implement CoGroup and Reduce the way you do because
of performance, correct? That is, you don't do any work in the user code of a
ReduceOperator but you do it in a chained MapPartition because there you get
all the elements which makes communication with the python process more
efficient. Same with CoGroup, where you implement your own grouping logic in
python from the raw input streams.
Overall I like the architecture, the communication between the host and the
guest language is well abstracted and I can see this being reused for other
languages.
Could you rename the CoGroupPython* classes to something more generic?
Because they really are a part of the generic language binding stuff and not
specific to python, correct?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
