Github user StephanEwen commented on the issue: https://github.com/apache/flink/pull/6075 @zhangminglei For your interest - there is a new Bucketing Sink in the Flink master (called `StreamingFileSink`), with a different design: Managing all state in Flink state (so it is consistent), with a new File System writer abstraction to generalize across HDFS, POSIX, and S3 (S3 still WIP) and with a more pluggable way to add encoders, like parquet and orc. As an example, we added a Parquet writer, which is quite straightforward and flexible with the new interface. Would be great to get your opinion on that and see if your ORC writer code also works with that. If it works out, the new StreamingFileSink could replace the current BucketingSink.
---