----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/61087/#review181369 -----------------------------------------------------------
ql/src/java/org/apache/hadoop/hive/ql/exec/tez/tools/KeyValueInputMerger.java Lines 86 (patched) <https://reviews.apache.org/r/61087/#comment256909> That is not always true. splits.get(1) could have a different path. You might want to add a loop + assert there. - Gopal V On July 25, 2017, 8:01 p.m., Deepak Jaiswal wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/61087/ > ----------------------------------------------------------- > > (Updated July 25, 2017, 8:01 p.m.) > > > Review request for hive, Gopal V, Jason Dere, and Sergey Shelukhin. > > > Bugs: HIVE-16965 > https://issues.apache.org/jira/browse/HIVE-16965 > > > Repository: hive-git > > > Description > ------- > > Usually, in a JOIN with multiple inputs (partitions), the inputs are read > sequentially, however, incase of SMB join, the inputs are read based on key > ordering. This invalidates the current IOContext assumption that the input > path once set wont change unless the input changes. > This was resulting in incorrect partition information in results as it is > derived from the input path in IOContext. > The new logic changes the input path as and when input changes. > > > Diffs > ----- > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/MapRecordSource.java > add7d08c40 > > ql/src/java/org/apache/hadoop/hive/ql/exec/tez/tools/KeyValueInputMerger.java > 698fa7f69e > ql/src/test/results/clientpositive/llap/llap_smb.q.out 87b33db805 > > > Diff: https://reviews.apache.org/r/61087/diff/3/ > > > Testing > ------- > > Added a new test. > > > Thanks, > > Deepak Jaiswal > >