[FLINK-3296] Remove 'flushing' behavior of the OutputFormat in DataStream API

[rmetzger]

I removed the FileSinkFunctionByMillis and removed all the millis arguments on the writing functions.

The whole "buffering" and "flushing" functionality was broken: Elements were kept in an ArrayList and send to the OutputFormat on "flush()". However, the flush was not really called periodically. It was only checked when new records arrived. So when a stream is not having elements for a certain time, the last few elements would just stay in the list until new elements arrive again.

[mbalassi]

With the RollingHDFSSink this functionality is not needed any more and as you suggested apparently was a misleading implmentation anyway. I like your not to the docs. 👍
Could you add [api-breaking] to the commit msg though please, when merging?

[StephanEwen]

May be worth adding that this means usually "at-least-once" , but may also mean data loss in cases where the output formats buffer data and do not immediately persist it.

[StephanEwen]

Code looks good. I would like to update the docs to include a bit more info (like in the inline comment) and at least refer to the addSink(...) method, for fault tolerant sinks.

[tillrohrmann]

Changes look good to me

[rmetzger]

Thank you for the review. I've addressed the comments and rebased the change.

Once travis has passed, I'll merge it!

[aljoscha]

I think the writeOutputFormat method name could be misleading. To me it implies writing the OutputFormat not writing something by using the OutputFormat.

[rmetzger]

I renamed the method to writeUsingOutputFormat and rebased to current master.

[rmetzger]

I'll merge the PR.

[coveralls]

Changes Unknown when pulling df49d5b on rmetzger:flink3296 into ** on apache:master**.