Gabor Kaszab has posted comments on this change. ( )

Change subject: IMPALA-5706: Spilling sort optimisations

Patch Set 12:

File be/src/runtime/
PS8, Line 1676:         --num_runs_in_one_merge;
> It's possible to have var-len slots but no var-len pages if all strings are
Thanks for the heads-up! I count for 2 buffers for an intermediate result then.
File tests/query_test/
PS12, Line 135:     doesn't consume all the memory from the second Sort. This 
query takes approx. 25s to
I gave this a second look and apparently I mis-measured something the last 
time. This in fact runs for at leas 25s according to my measurements. I spent a 
few hours to reduce this but according to my observations in order to see this 
behavior manifest in the # of merges we need this scale of data and runtime.
I'm open for suggestions about the next step here. I thought this shouldn't run 
in core tests for sure, so I restricted it for the exhaustive runs. However, 
that would run this test in a number of permutations of test configs and 
multiplying that with this 25s per run I find it too much.
Currently I'd prefer just to drop this test.

What do you think?

To view, visit
To unsubscribe, visit

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I74857c1694802e81f1cfc765d2b4e8bc644387f9
Gerrit-Change-Number: 9943
Gerrit-PatchSet: 12
Gerrit-Owner: Gabor Kaszab <>
Gerrit-Reviewer: Csaba Ringhofer <>
Gerrit-Reviewer: Gabor Kaszab <>
Gerrit-Reviewer: Tim Armstrong <>
Gerrit-Comment-Date: Thu, 17 May 2018 15:21:11 +0000
Gerrit-HasComments: Yes

Reply via email to