Uh oh!
There was an error while loading. Please reload this page.
[WIP][SPARK-24721][SPARK-25213][SQL] extract python UDF at the end of optimizer - #22244
[WIP][SPARK-24721][SPARK-25213][SQL] extract python UDF at the end of optimizer#22244cloud-fan wants to merge 1 commit into
Conversation
cloud-fan
commented
Aug 27, 2018
cc @icexelloss@HyukjinKwon@rdblue @icexelloss feel free to take this over and verify if it can pass the tests you added in #22104 , thanks! |
SparkQA
commented
Aug 27, 2018
Test build #95292 has finished for PR 22244 at commit
|
icexelloss
commented
Aug 27, 2018
@cloud-fan Thanks! I will take a look later today and incorporate this with my patch. |
rdblue
commented
Aug 27, 2018
@cloud-fan, I like this solution better than adding a special case in the v2 conversion to physical plan. This explains why the Python exec nodes weren't already in the tree! I'd much rather commit this or something like it than go with my work-around solution, which is fairly brittle. For now, I'm going to assume that this is going to be the fix. Please let me know if you want me to do further work to get my patch ready. We could use it if this work ends up being more complex than we want to wait on for the release. @icexelloss, also feel free to take the test case from #22206, and thank you for working on this! |
cloud-fan
commented
Aug 28, 2018
closing in favor of #22104 |
What changes were proposed in this pull request?
In #12127 , we moved the
ExtractPythonUDFsrule to the physical phase, while there was another option: doExtractPythonUDFsat the end of optimizer.Currently we hit 2 issues when exacting python UDFs at physical phase:
FileSourceScanExecand fail the planner when try to extract it later. See [SPARK-24721][SQL] Extract Python UDFs at the end of optimizer #22104This PR proposes to move
ExtractPythonUDFsto the end of optimizer.How was this patch tested?
TODO