[Impala-ASF-CR] IMPALA-4167: Support insert plan hints for CREATE TABLE AS SELECT

["Lars Volker (Code Review)" <ge...@cloudera.org>]

Lars Volker has posted comments on this change. ( http://gerrit.cloudera.org:8080/8400 )

Change subject: IMPALA-4167: Support insert plan hints for CREATE TABLE AS SELECT
......................................................................


Patch Set 6:

(8 comments)

Apologies for the delayed review. I had a few more questions. Let's discuss the parser changes in person if you feel I misunderstood the approach.

http://gerrit.cloudera.org:8080/#/c/8400/6//COMMIT_MSG
Commit Message:

http://gerrit.cloudera.org:8080/#/c/8400/6//COMMIT_MSG@19
PS6, Line 19: more
add "more memory efficient"? That makes it more obvious why it is more efficient to do additional work (sorting).


http://gerrit.cloudera.org:8080/#/c/8400/6//COMMIT_MSG@27
PS6, Line 27: Add exchange node before insert even in case of unpartitioned tables.
I don't think this is true, is it? For unpartitioned tables, what would the shuffle / exchange do? Instead "shuffle" forces an exchange, "noshuffle" prevents adding one, and leaving it out results in the planner making a cost-based decision (DistributedPlanner.java:218).


http://gerrit.cloudera.org:8080/#/c/8400/6//COMMIT_MSG@35
PS6, Line 35: was moved from tbl_def_without_col_defs to statement rules.
Mention how you tested it in the commit message.


http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/main/cup/sql-parser.cup
File fe/src/main/cup/sql-parser.cup:

http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/main/cup/sql-parser.cup@1055
PS6, Line 1055: plan_hints:hints
Does opt_plan_hints not work here?


http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/main/java/org/apache/impala/analysis/CreateTableAsSelectStmt.java
File fe/src/main/java/org/apache/impala/analysis/CreateTableAsSelectStmt.java:

http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/main/java/org/apache/impala/analysis/CreateTableAsSelectStmt.java@69
PS6, Line 69:    * Helper class for parsing.
Is this class still needed or a left-over from earlier patch-sets? Can you expand the class comment to explain why we need it and what it does?


http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/test/java/org/apache/impala/analysis/AnalyzeDDLTest.java
File fe/src/test/java/org/apache/impala/analysis/AnalyzeDDLTest.java:

http://gerrit.cloudera.org:8080/#/c/8400/6/fe/src/test/java/org/apache/impala/analysis/AnalyzeDDLTest.java@1701
PS6, Line 1701: Multiple non-conflicting
Can you add one that has multiple distinct hints, like shuffle, clustered?


http://gerrit.cloudera.org:8080/#/c/8400/6/testdata/workloads/functional-planner/queries/PlannerTest/ddl.test
File testdata/workloads/functional-planner/queries/PlannerTest/ddl.test:

http://gerrit.cloudera.org:8080/#/c/8400/6/testdata/workloads/functional-planner/queries/PlannerTest/ddl.test@233
PS6, Line 233: shuffle
Is that true? Without specifying a hint there should be cases where the cost based heuristic decides against adding an exchange node.


http://gerrit.cloudera.org:8080/#/c/8400/6/testdata/workloads/functional-planner/queries/PlannerTest/ddl.test@344
PS6, Line 344: 01:EXCHANGE [UNPARTITIONED]
At first I didn't understand what this one does (I saw that insert.test:574 has the same). On my machine this exchange collects all rows at the coordinator. Maybe you want to add that in a comment?



-- 
To view, visit http://gerrit.cloudera.org:8080/8400
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I8d74bca999da8ae1bb89427c70841f33e3c56ab0
Gerrit-Change-Number: 8400
Gerrit-PatchSet: 6
Gerrit-Owner: Csaba Ringhofer <cs...@cloudera.com>
Gerrit-Reviewer: Csaba Ringhofer <cs...@cloudera.com>
Gerrit-Reviewer: Gabor Kaszab <ga...@cloudera.com>
Gerrit-Reviewer: Lars Volker <lv...@cloudera.com>
Gerrit-Comment-Date: Mon, 08 Jan 2018 14:04:10 +0000
Gerrit-HasComments: Yes