### What changes were proposed in this pull request? In [celeborn-955](https://github.com/apache/incubator-celeborn/pull/1924), GetShuffleId RPC was introduced to generate a celeborn shuffle id from app shuffle id to support spark stage rerun GetShuffleId RPC assumes that Shuffle Write operation always happens before Shuffle Read operation, but this is not true for empty shuffle data in celeborn, which causes GetShuffleId RPC to throw NPE and fail the Job This PR fixes this bug ### Why are the changes needed? to avoid spark job failure with empty shuffle data ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? a new test case is included for empty shuffle data Closes #2136 from ErikFang/fix-GetShuffleId-RPC-NPE-for-empty-shuffle. Lead-authored-by: Erik.fang <fmerik@gmail.com> Co-authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com> Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com> |
||
|---|---|---|
| .. | ||
| src | ||
| pom.xml | ||