Tim Armstrong has posted comments on this change.

Change subject: IMPALA-3567 Part 2, IMPALA-3899: factor out PHJ builder
......................................................................


Patch Set 4:

I put together a spreadsheet of results for memory usage before and after this 
patch: 
https://docs.google.com/spreadsheets/d/17jyrLpDVUut_TKHrIhwROAa3BHh4ibVu9MqbLZxCGzE/edit?usp=sharing

Overall there are lots of small improvements.

There are a couple of bad regressions (q8 and q9), which are a bit surprising 
because on paper this should use a lot less memory to spill when there are many 
join nodes in the plan.

I dug into those regressions.

Q8: the difference is that scan memory usage was much higher for the scan node 
at the bottom of the join pipeline (~407M v 126M), so the problem was 
IMPALA-3209. 
Q9: I couldn't reproduce the lower memory usage before the patch

-- 
To view, visit http://gerrit.cloudera.org:8080/3873
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I1e02ea9c7a7d1a0f373b11aa06c3237e1c7bd4cb
Gerrit-PatchSet: 4
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Michael Ho
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-HasComments: No

Reply via email to