Details
-
Improvement
-
Status: Open
-
Minor
-
Resolution: Unresolved
-
Impala 2.9.0
-
None
-
ghx-label-1
Description
We should consider automatically disabling codegen per plan node based on the number of input/output rows. We have the plumbing already in place (it's used to disable codegen for non-grouping merge aggregations) but choosing and implementing a good policy is trickier.
We should be conservative about doing this, because the cost of codegen is O(1) but the runtime speedup is O(n).
Attachments
Issue Links
- is related to
-
IMPALA-2651 codegen overhead can be high
- Resolved
-
IMPALA-5483 Consider automatically disabling codegen for entire query based on planner estimates
- Resolved