tree 9e012dd92494749eb2efee9613006459d1f4514b
parent 7123269b5cc69806499bffc0757da63eaa1b9306
author chitralverma <chitralverma@gmail.com> 1597027782 +0800
committer William Guo <guoyp@apache.org> 1597027782 +0800

[GRIFFIN-305] Standardize sink hierarchy

**What changes were proposed in this pull request?**
Currently, the implementation of `Sinks` in Griffin poses the below issues. This PR aims at fixing these issues.
- `Sinks` are based on the recursive MultiSink class which is a sink itself but the underlying implementation is that of a `Seq` which causes ambiguity and isn't much useful. This has been removed.
- Some unused code like `SinkContext` has been removed.
- Data is converted from the performant DataFrame to RDD while persisting in both streaming and batch pipelines. A new method `sinkBatchRecords` has been added to allow operations directly on DataFrame for batch pipelines. Streaming will still use the old implementation which will be replaced with structured streaming.
- Refactored the methods of `Sink` like changed `start`/ `finish` to `open`/ `close` and `jobName` was incorrectly passed as `metricName`.
- Presently, only one instance of a sink with a given type can be defined in the env config. This will not allow the cases where you want to configure multiple sinks of same type like HDFS or JDBC. Added sink `name` to env config which is used to define the sink that should be used in the job config also.
- Updated all sinks as per the changes above. With some additional changes to ConsoleSink

**Does this PR introduce any user-facing change?**
Yes. As mentioned above, the sink config has changed in env and job configs.

**How was this patch tested?**
Griffin test suite and additional unit test cases

Author: chitralverma <chitralverma@gmail.com>

Closes #575 from chitralverma/standardize-sink-hierarchy.