Byte-based GroupByKey in Flink Streaming
Open
flink
improvement
P3
runners
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 196
Description
Currently Flink Streaming GBK is implemented using `SystemReduceFn#buffering`, that works on de-serialized values (we need to de-serialize value after shuffle and then serialize again when appending to bag state).
We can remove this unnecessary serde cycle by using binary representation instead of de-serialized payload.
Imported from Jira [BEAM-11443](https://issues.apache.org/jira/browse/BEAM-11443). Original Jira may contain additional context.
Reported by: dmvk.
Contributor guide
Assessment
This issue has not been assessed yet.