Hello Alex Behm,
I'd like you to do a code review. Please visit
http://gerrit.cloudera.org:8080/2628
to review the following change.
Change subject: WIP: IMPALA-3148. Fix selectivity computation for pushed Kudu
predicates
......................................................................
WIP: IMPALA-3148. Fix selectivity computation for pushed Kudu predicates
This follows up on a TODO from the Kudu merge and also fixes a bug:
IMPALA-976 changed the computation of selectivities for a combined
list of conjuncts to better handle expressions with no selectivity
estimate. The Kudu implementation was forked from before this change
and thus did not have an equivalent change.
This refactors the algorithm to a new static method and calls it from
both PlanNode and KuduScanNode so that the selectivity estimate
behavior is the same regardless of whether Kudu can evaluate the
predicate server-side.
The kudu-selectivity test had to be modified since it now produces
'unavailable' cardinality instead of '1' cardinality. As far as I can
tell, there are no stats available on this table, so the previous
estimate of '1' may have been a mistaken multiplication of -1 * -1
due to a confluence of bugs. Given this, it probably makes sense
to somehow add a query test which runs against a kudu table with stats.
I tested this on TPCH 3TB and verified that the plans are reasonable
now where they used to be nonsense.
Change-Id: Id507077b577ed5804fc80517f33ea185f2bff41a
---
M fe/src/main/java/com/cloudera/impala/planner/KuduScanNode.java
M fe/src/main/java/com/cloudera/impala/planner/PlanNode.java
M
testdata/workloads/functional-planner/queries/PlannerTest/kudu-selectivity.test
3 files changed, 14 insertions(+), 19 deletions(-)
git pull ssh://gerrit.cloudera.org:29418/Impala refs/changes/28/2628/1
--
To view, visit http://gerrit.cloudera.org:8080/2628
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-MessageType: newchange
Gerrit-Change-Id: Id507077b577ed5804fc80517f33ea185f2bff41a
Gerrit-PatchSet: 1
Gerrit-Project: Impala
Gerrit-Branch: cdh5-trunk
Gerrit-Owner: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Alex Behm <[email protected]>
