Alex Behm has posted comments on this change.

Change subject: IMPALA-3940: Fix getting column stats through views.
......................................................................


Patch Set 2:

(4 comments)

http://gerrit.cloudera.org:8080/#/c/3865/2/fe/src/test/java/com/cloudera/impala/common/FrontendTestBase.java
File fe/src/test/java/com/cloudera/impala/common/FrontendTestBase.java:

Line 66: public class FrontendTestBase {
> nice, thank you!
Done


Line 180:     CreateViewStmt createVieweStmt = (CreateViewStmt) 
AnalyzesOk(createViewSql);
> remove e in Viewe
Done


http://gerrit.cloudera.org:8080/#/c/3865/2/testdata/workloads/functional-planner/queries/PlannerTest/tpcds-all.test
File testdata/workloads/functional-planner/queries/PlannerTest/tpcds-all.test:

Line 2515: |  |  hash predicates: ss_store_sk = s_store_sk
> do you know what the perf impact of these plan changes is?
Don't know, but the plan looks saner since the join on date_dim (the scan has 
predicates) was moved before the join on store. Even if the old plan was faster 
I feel like that would be just by chance and we should accept the current saner 
looking plan.


http://gerrit.cloudera.org:8080/#/c/3865/2/testdata/workloads/functional-planner/queries/PlannerTest/tpch-views.test
File testdata/workloads/functional-planner/queries/PlannerTest/tpch-views.test:

Line 1: # TPCH-Q1
> this is going to be hard to verify by hand.
Good question. I verified the plans manually with meld and the only difference 
are in how slot refs are printed. In the view version all slots refs are fully 
qualified. In the original version some slot refs are unqualified, some are 
fully qualified and some are qualified with an explicit table alias (like n1 or 
n2 in TPCH-Q7).

I tried to implement a search+replace alternative, but unfortunately, it is not 
easy to do due. The regex gets increasingly complicated because some column 
names are prefixes of other column names and the original version has slot refs 
with explicit aliases like "n1", "n2". My feeling is that the search+replace 
path is complicated enough to be brittle itself.

As a compromise, maybe we can just keep the single-node plans in the new 
view-based TPCH test?


-- 
To view, visit http://gerrit.cloudera.org:8080/3865
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: Ie3b62a5e7e7d0e84850749108c13991647cedce6
Gerrit-PatchSet: 2
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Alex Behm <[email protected]>
Gerrit-Reviewer: Alex Behm <[email protected]>
Gerrit-Reviewer: Marcel Kornacker <[email protected]>
Gerrit-HasComments: Yes

Reply via email to