impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tim Armstrong (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-5612: join inversion should factor in parallelism
Date Mon, 21 Aug 2017 23:33:13 GMT
Tim Armstrong has posted comments on this change.

Change subject: IMPALA-5612: join inversion should factor in parallelism

Patch Set 5:

File fe/src/main/java/org/apache/impala/planner/

Line 451:    * inversion.
> Returns false if any join input is missing relevant stats.

Line 463:    * - the join strategy is PARTITIONED and rows are distributed evenly.
> I suggest you add a quick note why BROADCAST is not particularly relevant. 
Done. It was a bit tricky to explain crisply.

Line 473:    * The estimated cost of a hash join before and after inversion, measured in an
> estimated per-host cost

Line 482:    * We choose b = 10 and C = 5 empirically because it seems to give reasonable
> You had this great spreadsheet over a wide range of inputs. Do you mind dum
I cleaned up the spreadsheet a bit and put it here:

Line 491:   private boolean isInvertedJoinCheaper(boolean isLocalPlan, JoinNode joinNode)
> nit: flip args to make them consistent with invertJoins()

Line 509:     final long CONSTANT_COST_PER_ROW = 5;

Line 518:       LOG.trace("lhsCard " + lhsCard + " lhsBytes " + lhsBytes +
> Also log the tblRefIds_ and/or tupleIds_ or the join, otherwise it will be 

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: Icacea4565ce25ef15aaab014684c9440dd501d4e
Gerrit-PatchSet: 5
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <>
Gerrit-Reviewer: Alex Behm <>
Gerrit-Reviewer: Bharath Vissapragada <>
Gerrit-Reviewer: Mostafa Mokhtar <>
Gerrit-Reviewer: Tim Armstrong <>
Gerrit-HasComments: Yes

View raw message