HIVE-25125:PTF: Vectorize percent_rank function - #6752
Conversation
0fcbf1b to
8bbdd10
Compare
| -- 9. Test Windowing Functions | ||
| -- SORT_QUERY_RESULTS | ||
|
|
||
| -- Vector PTF does not buffer PARTITION BY columns (constant within a partition). When a partition |
There was a problem hiding this comment.
should we add an explicit guard over letting it fail with a ClassCastException or better even fall back to non-vector PTF ?
There was a problem hiding this comment.
I was thinking it should be evaluator specific in vectorised mode like here I ran into this in case of sum which should be computed for partition column not via buffered column values but via (constant partition column value* no. of rows in the partition) but since currently it only works via buffered column approach where we don't have partition column so index mapping is wrong in that case,
I think it would need some investigation regarding which evaluators should support it or not and should be made evaluator specific rather than a generic exception thrown?
I didn't created a ticket yet for it, if you think that makes then I'll create one and add the todo here, wdyt?
There was a problem hiding this comment.
please check that unsupported case is fall-backed to non-vectorized mode, otherwise it's a functional regression
There was a problem hiding this comment.
done added non-vectorized fallback in Vectorizer.java and also tests to cover that case in vector_ptf_unbuffered_partition_col.q , please have a look
asf-ci-hive
added
tests pending
tests passed
and removed
tests passed
tests pending
labels
b75d3e6 to
1062efe
Compare
asf-ci-hive
added
tests pending
tests failed
and removed
tests passed
tests pending
tests failed
labels
bca1c3b to
0eb8e72
Compare
asf-ci-hive
added
tests failed
tests pending
and removed
tests pending
tests failed
labels
asf-ci-hive
added
tests unstable
tests pending
and removed
tests unstable
tests pending
labels
Change-Id: I6fb4153d87bf92fd1c1993c79396465319d6b45a
Change-Id: I7a7cda2764c072a7b6aa5d9f0a8eb1dc61299640
Change-Id: Ic5ea75acc155ef3acda512312acf00cf47c49eaf
Change-Id: Ia2d3cb09ee72b6e1d6347ac39321e1822b236049
4a45162 to
c9fc6cb
Compare
Change-Id: Ib172b92c10e6d34fd3c43e40b587936cd233a651
c9fc6cb to
5902a28
Compare
|
3 participants
Change-Id: I6fb4153d87bf92fd1c1993c79396465319d6b45a
What changes were proposed in this pull request?
Vectorize percent_rank() function.
Why are the changes needed?
percent_rank() currently executes in non vectorized fashion, so enabling it to also be supported in vectorized manner
Does this PR introduce any user-facing change?
No
How was this patch tested?
added q test and ci results