Github user hequn8128 commented on the issue:
https://github.com/apache/flink/pull/6075
Hi @zhangminglei , thanks very much for your PR.
As for dependencies, I think it is better to move ORC related classes into
the module of `flink-orc`, so we don't need to add orc dependency in module of
`flink-connector-filesystem`.
Moreover, the `OrcFileWriter` should implement interface
`CheckpointedFunction` to prevent loss of data in `rowBatch` during job
failover. During checkpointing, you can store the data in `rowBatch` into
state, or call `flush()`. Similar logic can be found in `BucketingSink`.
Best, Hequn.
---
